Hi SaugaTech community,

Meetup #11 happened last week, and if you were in the room at IDEA Mississauga, you already know it ran long. “Tech Careers in 2026” was supposed to wrap on schedule. It didn’t. The panel kept going, the questions kept coming, and by the time we finally had to call it, most of the room just moved to pizza and kept the conversation going anyway. Good problem to have, and it’s got us thinking about how to scale this thing up — more on that below.

For now, we’re picking the series back up. Part 4 of AI Under the Hood is the paper that finally taught AI to behave like something you’d actually want to talk to.

🚀 First Things First - Volunteering with SaugaTech

Meetups like #11 don’t run themselves, and honestly, they’re starting to outpace what our small core team can carry alone. So we’re trying to do something we haven’t really done before: putting together a formal volunteering program as a part of our newly incorporated non-profit entity.

Three areas, specifically to start off with:

1. Event Curation — helping shape what future meetups actually look like. Picking themes, lining up speakers and panelists, thinking through format.

2. Content Management — this newsletter, recap posts, social content, the website. If writing or editing is your thing, this is where we’d love to have your help.

3. Mentoring & career guidance— working with members of the community, individually or as a group, share your own experiences, help them shape their career direction

None of these need to be a huge time commitment. A few hours a month, pointed at something you’re actually good at, goes a long way for a community this size.

Here is the link to the Volunteering Interest form. We look forward to hearing from you. Feel free to share it in your networks to those who might be interested in helping us out

Anyway, coming back to today’s topic.

The Practical Problem Statement

Picture the most well-read new hire you’ve ever had on a team. They’ve absorbed the entire internet, they can write in any style, hold a conversation on nearly any topic, and finish a sentence in exactly the way it statistically tends to get finished. Ask them a direct question, though, and something odd happens. Half the time they just keep talking, adding more and more plausible-sounding text, because that’s the only thing they were ever actually trained to do. Nobody taught them what it means to help.

That’s roughly the state language models were in before this week’s paper. Technically brilliant, occasionally useless, and this is the paper that fixed it.

If you’ve been following along, this one closes a loop we’ve been building since Part 1. Attention gave models the ability to read context. GPT-3’s few-shot learning gave them the ability to adapt on the fly. RAG gave them a way to look things up instead of guessing. This paper is the one that finally taught them to behave like something you’d actually want to talk to.

Grab a coffee. Let’s get into it.

⚓ The Paper: “Training Language Models to Follow Instructions with Human Feedback” (2022)

In early 2022, a team at OpenAI published the paper behind InstructGPT, the direct predecessor to ChatGPT. The core discovery is almost embarrassingly simple to say out loud: making a model bigger does not make it better at doing what you actually asked. Size and helpfulness turned out to be two completely different problems, and this paper is the one that finally tackled the second.

The problem the researchers identified

A raw language model like GPT-3 is trained on one objective: predict the next word on a page pulled from the internet. That’s a genuinely useful skill, but it’s not the same skill as “follow this instruction helpfully and safely.” Ask a raw model to write a professional email and it might just as easily continue your prompt with a list of other emails you could write, or a joke about email, or nothing useful at all. The researchers call this a misaligned objective. The model is optimizing for “sounds like the internet,” when what the user actually wants is “does what I asked.”

This gap shows up as three separate failure modes, and it’s worth naming them individually because you’ll recognize all three from any raw model you’ve used.

1. Models make things up.

2. Models sometimes ignore what you actually asked and answer a nearby question instead.

3. And Models can drift toward toxic, biased, or simply inappropriate output, because that’s present somewhere in a training set the size of the internet.

The core idea — teach the model with a manager, not just a textbook

The fix here has three stages, and the analogy that holds up best is a new hire’s first few months on the job.

Stage one is supervised fine-tuning. Human contractors write out example responses to a wide range of prompts, showing the model what a genuinely helpful answer actually looks like. Think of this as the onboarding binder. Here’s what a good customer email sounds like. Here’s how you handle an ambiguous request. The model gets fine-tuned directly on these examples.

Stage two is reward modelling. Contractors are shown several different model responses to the same prompt and asked to rank them from best to worst. That ranking data trains a second, separate model, called the reward model, whose entire job is to look at a response and predict how a human would rate it. This is the manager who’s read enough of your work to guess, fairly reliably, whether you’re about to get positive feedback or a note back asking you to redo it.

Stage three is reinforcement learning. The original model gets fine-tuned again, this time using the reward model as a coach, adjusting its behaviour over and over to chase a higher predicted score. The technique used here is called PPO, Proximal Policy Optimization, which is really just a careful way of nudging the model’s behaviour without letting it swing wildly or forget what it already knew how to do.

Put together, this is RLHF. Reinforcement Learning from Human Feedback.

And the researchers found something that should give every builder in this community real pause about the “just make it bigger” instinct: a 1.3 billion parameter InstructGPT model, fine-tuned this way, was preferred by human raters over the raw 175 billion parameter GPT-3, a model over a hundred times its size.

The evidence that it worked

The headline number is that stark preference gap. In direct comparisons, InstructGPT’s output was preferred over raw GPT-3 output 85% of the time, and preferred 71% of the time even when GPT-3 was given a carefully hand-crafted prompt meant to coax better behaviour out of it.

The hallucination numbers are the ones worth sitting with a little longer, especially after we spent all of Part 3 talking about RAG. On closed-domain tasks, where the correct answer should come entirely from what’s already in the prompt, InstructGPT made things up about half as often as GPT-3 did, a 21% hallucination rate compared to 41%. On the TruthfulQA benchmark specifically, it produced truthful and informative answers roughly twice as often.

There were smaller wins too. When explicitly asked to respond respectfully, InstructGPT generated about 25% fewer toxic outputs than GPT-3. And the model showed a genuinely interesting kind of generalization: even though the fine-tuning data was almost entirely English, InstructGPT got noticeably better at following instructions in other languages and at answering questions about code, tasks it was never specifically trained to improve on.

Where it fell short

InstructGPT still hallucinated, just less often, and it developed a couple of new bad habits along the way. It would sometimes hedge on simple questions that had a clear answer, giving a wishy-washy “there are many perspectives” response instead of just answering. It could also be led astray by a false premise buried in a question, going along with something untrue instead of catching it.

And there’s a bigger, more uncomfortable question sitting underneath all of it. Whose preferences was this model actually trained on? The paper is upfront that the labelers were a small, specific group, mostly English-speaking, hired through a couple of platforms, following instructions written by OpenAI researchers. “Aligned” here means aligned to that particular group’s judgment calls, not some universal standard of correctness. That distinction matters more the more these models get deployed into contexts none of those labelers ever saw.

The side effect nobody saw coming

Worth a quick detour here, because this paper didn’t just introduce a training technique. It quietly created an entire labour category. RLHF runs on a constant supply of humans ranking model outputs and writing example answers, and the paper’s own appendix is candid about it: about 40 contractors, hired through Upwork and a company called Scale AI, screened for how well their judgment matched the researchers’.

That demand didn’t stay small. Scale AI’s revenue went from roughly $250 million in early 2022, right as RLHF took off, to around $750 million by late 2023, and Meta later paid $14.3 billion for a 49% stake in the company in 2025. The broader data annotation and labeling market is projected to grow from about $1.2 billion in 2024 to $10.2 billion by 2034. Newer entrants like Surge AI, Labelbox, and Mercor have built entire businesses competing for the same kind of work: paying people to rank AI outputs so a reward model can learn what “better” looks like.

It’s a strange thing to sit with. The paper that taught machines to sound more human ended up creating a lot of very human jobs to make that possible.

⚓ Why this actually matters for what we build

This paper is the reason a raw model and a genuinely usable assistant feel like completely different products, even when the underlying architecture is identical.

1. It explains why “the base model” and “the chat model” behave so differently for the exact same task. If you’ve ever pulled a raw, non-instruction-tuned model off a hub and been confused by how oddly it responds compared to something like ChatGPT or Claude, this is why. Same architecture underneath, wildly different training on top of it. When you’re choosing between a base model and an instruction-tuned one for a product, you’re choosing between raw capability and actual usability, and for almost anything customer-facing, usability wins.

2. It’s also a genuine argument for a smaller, well-trained model over a bigger, raw one. A 1.3B model beating a 175B model on real user preference is a big deal if you’re building anything cost-sensitive. Before you reach for the biggest model on the market, it’s worth asking whether a smaller, well-aligned one might actually serve your users better, and cost a fraction as much to run.

3. It also gives a name to something every builder eventually notices the hard way: models will over-hedge or misread a false premise, and it’s not really a bug so much as a side effect of how they were trained to be careful. If your product needs a direct, confident answer to a well-defined question, it’s worth testing specifically for these two failure modes, because they show up more than people expect.

And this is where the thread to Constitutional AI actually starts. Anthropic’s Constitutional AI, which we’ll cover properly in a future edition, took this same basic RLHF framework and asked a pointed question: what if the model itself could help judge its own responses against a written set of principles, instead of leaning entirely on humans to hand-label every example? It’s not a replacement for the ideas in this paper so much as an evolution of them, aimed at reducing how much this process depends on human labelers seeing every harmful case in advance, and at making that alignment process more consistent and more scalable as these models get deployed more broadly.

✨ SaugaTech Epilogue — four papers, one continuous story of evolution

Four parts into this series now, and it’s worth stepping back and seeing the shape of the whole thing.

Attention taught a model how to actually read what’s in front of it, understanding context instead of processing words in isolation.

Few-shot learning, once that model got big enough, taught it to adapt to a brand new task on the spot, just from a couple of examples.

RAG gave it a way to reach outside its own training and check a real source instead of guessing.

And this week’s paper, RLHF, taught it to actually behave, to take everything it’s capable of and point it at what the person in front of it actually wants.

Understand, adapt, look things up, behave helpfully. None of those four ideas replaced the one before it. Each one is sitting underneath the model you used this morning, doing its job quietly, all at once.

There’s a decent lesson for builders buried in the order these ideas arrived in, too. Raw capability came first, and it wasn’t enough on its own. It took a genuinely separate effort, actually listening to what people wanted and training toward that, to turn a very well-read machine into something worth talking to. Worth remembering the next time a product feels technically impressive but somehow still misses the point.

See you soon.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Saturday afternoons at Square One usually mean the mall, the food court, maybe a walk to the Celebration Square to check out what’s going on. Last Saturday, a good chunk of Mississauga’s tech crowd had a different plan.

By the time IDEA Mississauga filled up for another packed room meetup, it had that familiar pre-match energy — familiar faces catching up, new faces getting pulled into conversations they didn’t expect to have, the kind of easy chatter that tells you a community has found its footing. Then we kickstarted the panel, and the energy shifted. Chatter gave way to attention. Casual discussions gave way to genuinely hot takes.

But this wasn’t a one-way panel — the floor was open from the start, and the folks in the room took full advantage. Questions came in as topics were being covered, not saved for the end. People shared their own experiences, pushed back, added context from their own job searches and hiring decisions. The room was as much a part of the discussion as the panelists were.

Here’s the best measure of how it went: the panel was planned to run from 1:05 to 1:45. It went till almost 2:15. The pizzas had arrived long before that — and nobody moved. Not one person. When we finally wrapped, the one-on-ones continued over pizza, panelists surrounded by people with follow-up questions, specific situations, and career decisions they wanted a real read on. That’s not something you can plan for. That’s just what happens when the conversation is genuinely good.

But honestly, some of the most valuable conversations that evening didn’t happen on stage at all. But before we get into all that, we have some news to share.

⚓ A Milestone Worth Mentioning

We also used this meetup to share something we’ve been working on quietly for a while: SaugaTech is now officially incorporated as an Ontario non-profit. What started as a Reddit post in April 2025 — 20 people at a community centre, a ragtag group of volunteers and a whole lot of goodwill — now has a real legal foundation under it. The kind that lets us build partnerships that last, plan past the next meetup, and play the long game we keep talking about. Not the next year. The next ten.

⚓ The Room Before and After

Before the session and well after the pizza was gone, something quietly important was happening on the sidelines. Folks in the middle of career transitions — some of them navigating it publicly, some of them not quite ready to say it out loud — were getting real, unfiltered guidance from panelists and community regulars who stuck around to listen and help. No awkwardness, no elevator-pitch pressure, just people talking to people. That’s been true of SaugaTech for a while now, but this evening had more of it than most.

This is what we mean when we say SaugaTech is a space where you can open up. Job searching is one of the loneliest things you can go through, and this community keeps proving that it doesn’t have to be..

⚓ The Panel — What the Room Took Away

Thank you to Anwar (Tundra Technical), Gustavo (TechBlocks), and Gurpreet Kochhar (Bell) for being genuinely candid and unfiltered. No corporate-speak, no recycled LinkedIn advice. Here’s what the room walked away with.

Authenticity is the new superpower. In a world where AI can reshape anyone’s resume to fit any job description, the panel kept circling back to this: staying truthful and playing to your actual strengths is the differentiator now, not a liability. Recruiters see through shallow claims fast. Use AI to sharpen how you present yourself - not to erase what makes you distinct.

Three pillars decide most outcomes - what you know, whom you know, and what you can convey. Technical skill alone doesn’t carry candidates the way it used to. Domain expertise still matters, but problem-solving under ambiguity and genuine culture fit often end up being the deciding factor. Networking is nearly as important as competence — meeting the right person at the right time opens doors applications alone won’t. And have a point of view: read your industry, understand the trends shaping it, and be ready to talk about it. It makes every conversation land differently.

One sharp tactical note: if you’re targeting a specific company or role, skip the VP and reach out to someone who recently cracked that same role. They remember the process and tend to be far more willing to help. And perfect your 1–2 minute elevator pitch — you never know who you’ll run into. This crowd is proof of that.

The technical bar has moved. Working alongside AI and coding agents is close to expected now. What separates candidates is system design thinking and the ability to use available tools to actually reach the right outcome. Almost every hiring manager on the panel looks at GitHub — not for polish, but to understand what someone is building and what problems they’re choosing to solve.

For recent grads, coachability is everything. Can you learn quickly, take feedback, adapt? That matters more at this stage than having every box checked. And never self-reject. This one came straight from the room — someone asked about missing the bar on one or two points in a job description, and the answer was unambiguous: if you fit 15 of the asks, the one or two you’re missing won’t sink you. JDs are frequently imperfect wish lists written by committees. Apply anyway, and let the conversation do the rest.

LinkedIn is not optional anymore. It’s the window recruiters look through before they ever talk to you. Visibility is table stakes now, not a bonus.

And finally — protect your mental health through the process. The panel didn’t skip this, and neither will we. Job searching is a process, not an event. Patience matters as much as any tactical advice on this list.

⚓ Builder Showcase: Four Stories, Straight From the Source

After pizza, the floor went to four builders in our community — Shamraiz, Umesh, Samiul, and Himanshu — who each shared what they’ve been building, how they built it, and what they learned along the way. No slide decks, no polish — just honest accounts of their motivations, what their product does, the actual journey of building and small wins that kept them going. If the panel was about getting hired, this part of the evening was a reminder that a lot of our community is out there building things instead. And that’s just as valid a path.

✨ SaugaTech Epilogue

Thank you to Anwar, Gustavo, and Gurpreet for giving the room their time and their honesty.

A big thank you to ACCES Employment and Adriana for helping bring this session together — and specifically for connecting us with Gustavo as a speaker, which really elevated the quality of the conversation.

Thank you to Invest Mississauga and IDEA Square One for continuing to support this community.

And thank you to everyone who packed that room, stayed past the clock, kept the conversations going in the hallway, and made this one of our best events yet.

From a Reddit post to a registered non-profit. The throughline hasn’t changed — a room full of people helping each other navigate something genuinely hard, together.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Hi SaugaTech community,

A few years back, I was trying to make sense of an old banking regulation, we were trying to dig through old printed regulations for a clause buried in a document from almost twenty years earlier. Nobody in the room had it memorized. What happened instead was completely ordinary - one of our office assistants dug out the filing cabinet, pulled the right folder, and read the actual line back. Nobody needed to have the answer sitting in their head. They just needed to know where to look.

That small, boring moment is basically the plot of this week’s paper. Part 3 of AI Under the Hood is here, and it’s the one that explains why “chat with your documents” became the single most common thing builders in this community have shipped over the past year.

If you’re just joining the series, Part 1 covered the Transformer and Part 2 covered how GPT-3 learned to pick up new tasks from a few examples. Neither is required reading for this one, but they’ll make a few things click faster.

Grab a coffee. Let’s get into it.

🚀 First Things First

Meetup #11 is ten days out — August 22nd at IDEA Mississauga, Square One. The topic “Tech Careers in 2026: The New Rules of Getting Hired.”

We’ve got a great panel of experienced tech managers and recruiter coming in to talk about what’s actually changed on the other side of the table. If you’ve been job hunting, thinking about a switch, or just curious, join us at IDEA Mississauga on 22nd afternoon. We are also hosting our 2nd builders showcase along with the usual Open Networking session, so its going to be another day to spend with the vibrant SaugaTech community.

RSVP here

Anyway, coming back to our today’s AI paper.

⚓ The Paper: “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” (2020)

In May 2020, a team at Facebook AI Research led by Patrick Lewis published a paper that tackled a problem every builder in this community has bumped into by now. The model sounds confident. The model is also sometimes just wrong. It has a name in the field now — hallucination — and this paper is one of the earliest and cleanest fixes for it. The shorthand everyone uses is RAG.

The problem the researchers tried to solve

Everything a language model like GPT-3 or BART knows, it learned during training. Once training ends, that knowledge is frozen. Ask about something that happened last week, something buried in your company wiki, or something that simply wasn’t common enough in the training data, and the model either admits it doesn’t know or, worse, fills the gap with something fluent and wrong.

There’d already been attempts at fixing this — REALM and ORQA both paired a language model with some kind of external lookup — but they were built specifically for extractive question answering, basically highlighting the exact span of text that contains an answer. Nobody had shown this idea working as a general-purpose approach across a wide range of tasks, including ones where the model has to actually write a sentence, not just point at one.

The core idea — two kinds of memory, working together

The paper gives a name to something worth remembering.

Parametric memory is what the model learned during training, baked into its weights. Fast, fluent, and frozen the moment training ends.

Non-parametric memory is an external, searchable collection of documents the model can look things up in — in this case, a snapshot of Wikipedia chopped into roughly 21 million chunks of about 100 words each.

RAG connects the two. Before the model writes an answer, a retriever component searches the document collection for the passages most relevant to the question and hands those passages to the generator as extra context. The generator then writes its answer with those actual documents sitting in front of it, instead of relying purely on whatever it memorized months earlier.

Our filing cabinet from the opening holds up well here. The retriever is the person who knows exactly which drawer to check. The generator is the person who reads the line back in plain English. Neither one does the whole job alone.

How it works: the library card catalogue

The retriever is built on a technique called Dense Passage Retrieval, DPR for short. Instead of matching keywords the way an old-school search engine does, DPR converts both the question and every document chunk into a vector, a long string of numbers that captures meaning rather than exact wording. A question about west GTA development ends up sitting close to a passage about Mississauga’s growth as a business hub, even when the actual words barely overlap.

Finding the closest match among 21 million vectors sounds like it should take forever. It doesn’t, thanks to a search technique called Maximum Inner Product Search, which narrows the field down in a fraction of a second using an index built ahead of time. This is the same basic machinery, evolved a fair bit since, sitting underneath every vector database you’ll hear a developer mention today.

Once the retriever hands over its top candidates, the generator — BART, a 400 million parameter model related to the ones we covered in Part 1 — simply concatenates the retrieved passages with the original question and writes the answer. The retriever and generator are trained together, end to end, with no one ever telling the retriever exactly which document is “correct.” It learns which documents are useful purely from whether they helped the generator land the right answer.

The paper tests two flavours of this. RAG-Sequence picks one set of documents and uses it for the whole answer, start to finish. RAG-Token can lean on a different document for each word it generates, which lets it blend information from more than one source into a single answer.

The evidence

RAG set a new state of the art on four open-domain question answering benchmarks — Natural Questions, TriviaQA, WebQuestions, and CuratedTrec — beating both pure “closed-book” models that only had parametric memory, and specialized extract-the-span systems that only had retrieval. On Natural Questions specifically, RAG hit 44.5% exact match, well ahead of the closed-book T5 model’s 34.5%, using a fraction of the parameters.

The most convincing evidence isn’t a benchmark score though. It’s an experiment the researchers ran almost as a side note. They built two separate document indexes — one from a Wikipedia snapshot taken in December 2016, one from December 2018 — then found 82 world leaders whose positions had changed between those dates and asked the model “Who is {position}?” against each index separately.

Using the 2016 index, the model correctly named the 2016 leader about 70% of the time. Swap in the 2018 index, and it named the 2018 leader correctly about 68% of the time. Mismatch an outdated index with a current question, and accuracy fell into single digits.

That’s the whole point of the architecture in one experiment. You can update what the model “knows” by swapping the document collection, without touching a single one of its trained weights.

Where it fell short

The paper is upfront about a failure mode they call retrieval collapse. On some tasks, especially open-ended ones like story generation, the retriever would learn to fetch the same handful of documents regardless of what was actually being asked, and the generator would learn to just ignore them, quietly reverting to acting like a plain BART model. The researchers suspect it happens more on tasks where the target output is long and the connection to any one specific fact is loose, which gives the retriever a weaker training signal to learn from. It’s a reminder that bolting on retrieval doesn’t automatically make a system reliable. The retriever still has to actually be doing something.

⚓ Why this actually matters for what we build

This paper is close to a blueprint for the majority of “AI knowledge base” products built in the last two years, including probably a few sitting in this community’s WhatsApp group right now.

The first is the most practical one for anyone reading this with a day job and limited weekends: you don’t need to train a model to build something like this. You need a way to search your own documents, and an existing language model to write the answer once the right passages are sitting in front of it. Both are sitting on the shelf already. A handful of GTA builders in this community have already stitched the two together over a weekend.

If a chatbot you’ve built keeps returning garbage, check the filing system before you blame the model. That’s usually where the fault actually sits. Chunks that are too big or too small, embeddings that miss what a question is really asking, an index that just doesn’t contain the passage you needed — nine times out of ten, that’s the failure, not the language model reading the retrieved documents wrong.

And domain expertise is what actually goes in the filing cabinet. A generic chatbot knows nothing specific about your industry. Point one at your firm’s compliance manual, your clinic’s intake procedures, or your logistics company’s routing rules, and it suddenly knows exactly what it needs to — because you’re the one who decided what goes into the index. Years spent in healthcare, manufacturing, or financial services aren’t something a prompt can replicate. That experience is the fuel. Retrieval just puts the right piece of it in front of the model at the right moment.

The world-leader experiment is also a decent argument for why you don’t need to retrain anything every time your information changes. Swap the documents, and the answers update on their own. It’s a big part of why so many teams are reaching for a vector database right now instead of waiting on the next model release to simply “know more.”

One thing worth watching before you ship, though: don’t assume retrieval is pulling its weight just because you wired it in. The paper’s own retrieval collapse finding is a decent gut check — if the connection between a question and a specific document is loose, or your outputs run long, it’s worth checking whether the system is actually using what it retrieves or quietly ignoring it.

✨ SaugaTech Epilogue — three papers, one machine

Worth pausing here, three parts in, to notice these aren’t three separate stories. They’re one machine, built in layers, and each paper we’ve covered solved the problem the last one couldn’t.

Part 1 gave us Attention — a way for a model to read an entire sentence at once and figure out which words actually matter to each other, instead of plodding through word by word and forgetting the beginning by the time it reached the end. That’s the engine. It’s what makes a model capable of understanding context in the first place.

Part 2 took that engine and asked what happens if you build it big enough. Turns out, at 175 billion parameters, something new shows up that smaller models just don’t have — the ability to learn a brand new task on the spot, from a couple of examples in the prompt, no retraining required. That’s what makes the engine flexible. One model, reused for a dozen different jobs, just by changing what you show it.

This week’s paper takes that flexible engine and gives it something neither of the first two papers had. A memory it didn’t have to be born with. Attention lets it understand what’s in front of it. Few-shot learning lets it adapt to what you’re asking. RAG lets it go check something it never saw during training at all, and come back with an answer grounded in an actual document instead of a confident guess.

Understand, adapt, look things up. That’s most of what a serious AI product needs to do well, and it’s also, not coincidentally, roughly the order these ideas arrived in the research world.

There’s a good instinct for builders buried in that order too. You don’t need to have memorized the whole industry before you start something. You need to know where the filing cabinet is, and be willing to go check it before you speak.

Next time, we go back to something the engine still doesn’t do well on its own — reliably being helpful, rather than just being fluent — and the training trick, RLHF, that taught it the difference.

See you on August 22nd.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Hope you’re doing well and finding ways to beat the on and off summer heat that GTA is experiencing off late.

We’re excited to share that SaugaTech has a brand new website — www.saugatech.ca

It’s been in the works for a while, and it now gives you a proper home base to check out upcoming meetups, browse our past sessions, see how far this community has come since that first Reddit post in March 2025, and catch up on the Compass newsletter, all in one place. Take a look and let us know what you think. Also feel free to circulate the link in your personal and professional communities to spread the word and bring in a wider talent to our community and meetups.

Speaking of upcoming meetups, we’re back with a session a lot of you have been asking for.

Tech Careers in 2026: The New Rules of Getting Hired

📅 Saturday, August 22, 2026, 1-3 pm (Doors open at 12:30 pm)

📍 IDEA Mississauga, Square One

🎟️ More Details and RSVP Here

We’re bringing the other side of the table to you. A tech recruiter and hiring managers, with decades of experience between them making real hiring calls across startups, mid-size orgs, and enterprise, will share what’s genuinely changed and what’s actually landing with candidates in 2026.

This one’s for anyone job hunting, thinking about a pivot, or just wanting a clearer read on the market right now. It’s also a chance to:

• Network with local founders, senior engineers, PMs and tech leaders.

• Collaborate on what a real job search looks like in 2026.

• Contribute to the growth of the Mississauga and GTA tech ecosystem.

RSVP & Secure Your Spot

Looking forward to see everyone on 22nd August.

Best,

The SaugaTech Team

Supported by Invest Mississauga

P.S. Join the WhatsApp Group for early announcements and community discussions, message the organizers on +14377992290 to get added to the group.

Share

Hi SaugaTech community,

It’s Friday. The long weekend is almost here — Ontario’s civic holiday Monday means three full days away from Slack, standups, and whatever your manager last pinged you about. Most of the western GTA is already mentally checked out, and honestly, fair enough.

Before you close the laptop though — this one’s worth five minutes.

Last week we kicked off a new thread in the Compass — a slow walk through the foundational research papers of modern AI, explained in plain English, with an eye toward what each one means for something you might actually build. We started with the 2017 paper that introduced the Transformer architecture. If you missed it, Part 1 is here. Worth a read before this one, though you don’t strictly need it.

This week, Part 2.

If you’ve opened ChatGPT or Claude this week and typed something like “write this in the style of my last three emails” or “here are two examples, now do a third” — congratulations, you were running a five-year-old research technique without knowing its name.

That’s actually where this edition starts. Last time in the Compass, we went back to 2017 and unpacked Attention Is All You Need — the paper that let machines read a whole sentence at once instead of word by word. Cool mechanism. But an engine on its own doesn’t tell you what car you’re driving. This week’s paper is the one that took that engine and built GPT out of it — and in the process, changed what “using AI” actually means for the rest of us.

Grab a coffee. Let’s get into it.

🚀 First Things First

Save the date — our next meetup is August 22nd at IDEA Mississauga, Square One and this one’s been requested a lot: Tech Careers in 2026: The New Rules of Getting Hired.

We’re bringing in a tech recruiter and a couple of hiring managers to share what’s actually changed in tech hiring right now — a follow-up of sorts to last year’s “Navigating Tech Careers in the Age of AI,” this time straight from the people doing the hiring. If you’re job hunting, thinking about a switch, or just curious what’s landing on the other side of the table these days, this one’s for you.

Details and RSVP will drop in the WhatsApp group early next week https://chat.whatsapp.com/HgtfWvkWZpXE2Qvis2Yjrm

⚓ The Paper: “Language Models Are Few-Shot Learners” (2020)

In May 2020, a team at OpenAI published a paper introducing GPT-3. At the time it was the largest neural network ever built — 175 billion parameters. That number got all the headlines. But honestly, the number wasn’t really the interesting part.

The interesting part was this: you could teach this thing a brand new task just by showing it a few examples, and it would pick up the pattern without anyone touching a single one of its 175 billion parameters.

That’s the exact trick that makes ChatGPT, Claude, and pretty much every AI assistant usable by a normal person. Before this paper, that’s not how any of this worked.

The problem the researchers tried to solve

Before GPT-3, if you wanted a language model to do something specific, say sort customer support tickets by urgency, you had to fine-tune it. That meant collecting thousands, sometimes hundreds of thousands, of labeled examples for that exact task, then retraining the model on them. Want it to do something else too? New dataset. Fine-tune again. Every task started the whole expensive process over from scratch.

The OpenAI team basically pointed out something kind of obvious once you say it out loud: that’s not how people learn things. Nobody hands a new employee ten thousand labeled examples to teach them what an urgent email looks like. You show them two or three, maybe give a quick explanation, and they’re off. Why should a model need it any other way?

The core idea — in-context learning

The bet the paper made: what if a model, if it’s big enough and has seen enough text, can learn a brand new task just by reading a handful of examples right there in the prompt — no retraining, no gradient updates, nothing?

They called this in-context learning and tested it three ways:

• Zero-shot — just an instruction, no examples. “Translate this to French.”

• One-shot — exactly one example of the task, then you ask it to do a new one.

• Few-shot — somewhere between ten and a hundred examples stuffed into the prompt before your actual question.

None of these touch the model’s weights. It reads your examples, figures out what you’re going for, and does it — all in one pass, all within the length of a single conversation.

If that sounds like exactly how you already prompt ChatGPT or Claude — throw in a couple examples of the format you want before asking the real thing — that’s not a coincidence. You’ve been doing in-context learning this whole time. This paper is the reason it works at all.

The evidence — and why bigger turned out to actually be different

The paper didn’t just claim this worked. It showed something specific: this ability barely existed in small models, and got dramatically stronger the bigger the model got.

They trained eight versions of the same architecture — from 125 million parameters all the way up to the full 175 billion — and ran all of them through the same tests, in zero-shot, one-shot, and few-shot settings. The gap wasn’t subtle. Take a simple task like unscrambling a word with its letters shuffled around. Small models barely improved even with a hundred examples in front of them. The full-size GPT-3 got dramatically better with more examples, and beat what any smaller model managed even with a hundred shots.

That gap between “no examples” and “a few examples” widened as the models got bigger. Small models basically ignored the extra examples. The 175B model used them well. Which told the researchers something important — this isn’t a trick you can just find in any model. It’s something that seems to emerge only once you’re big enough.

Some of the actual numbers, all from prompting alone, no retraining involved:

• Trivia questions: 71.2% accuracy few-shot, which actually beat systems that had been specifically fine-tuned for that exact task

• Arithmetic: near-perfect on 2-digit addition, and genuinely decent even on larger numbers it had never explicitly been trained to add

• Unscrambling words: putting a shuffled word back together, something the model was never directly trained to do

• Inventing new words: give it a made-up word and its definition once, and it could use that word correctly in a brand new sentence

• Writing news articles: people could only correctly guess GPT-3 had written a short article 52% of the time. Basically a coin flip.

All of it from a big model reading a few examples and figuring out the pattern on its own.

Where it fell short

The paper is refreshingly honest about this. GPT-3 was bad at tasks that required comparing two pieces of text directly — telling whether one sentence implies another, or whether the same word is being used the same way in two different sentences. On those, it barely beat random guessing even with lots of examples. It also sometimes contradicted itself over longer passages, and struggled with basic common-sense physics questions despite doing fine on benchmarks that were supposedly testing the same thing. That kind of candor is honestly part of why this paper still holds up as a reference point five years later.

⚓ Why this actually matters for what we build

This paper is basically the reason building with AI feels so different from building with any earlier generation of software.

Good prompt vs bad prompt — this paper explains the difference. Every time you drop two or three examples into a prompt before asking your real question — here’s the tone I want, here’s the format, now do it for my actual case — you’re literally running the few-shot procedure this paper described. A bad prompt is usually just an instruction on its own. “Write a professional email.” The model has to guess what “professional” means to you specifically, and it’ll guess based on whatever’s most average across everything it’s seen — which is rarely exactly what you wanted. A good prompt looks more like a small classroom. A couple of examples of the output you’re after, plus a couple of clear traits you want the model to lock onto — the tone, the structure, the things to avoid. You’re not explaining the task in more words. You’re showing it. If a model isn’t getting your format right, a longer explanation almost never fixes it. Two or three good examples almost always do.

Why the same API key can power ten completely different products. Before this paper, building an AI feature for classifying tickets and a separate one for summarizing documents meant training two different specialized models. After GPT-3 showed in-context learning scales with size, one large model could plausibly do both — same weights, no retraining, just a different prompt. That’s the whole reason a small team in Mississauga can build a support triage tool on Monday and a document summarizer on Tuesday using the exact same underlying model.

Context window size is not just a spec — it’s a product decision. Every example you give a model in a prompt takes up context. The paper showed performance improves as you give more examples — but there’s a hard ceiling on how many you can fit before the window runs out. That ceiling varies by model: GPT-3 had a 4,000 token window, modern models go much higher, but the constraint is always there. This is also why long conversations start going sideways. As the context fills up, earlier information gets less attention weight — the model’s focus narrows toward what’s most recent and the beginning of your conversation effectively fades. So “how many examples is enough” isn’t just a prompting question — it’s a function of which model you’re using and how much window you’ve already burned through. If you’re building something that relies on long multi-turn conversations or big document context, the model’s context window should be one of the first things you check, not an afterthought.

Fine-tuning still has a job — but it’s a narrower one. If your task is well-defined and you have real labeled data, a fine-tuned smaller model can still beat a big general one. But if you’re still figuring out what the task even is, few-shot prompting gets you there faster and cheaper every time.

The founders who build the most efficient AI products aren’t necessarily the ones with the cleverest prompts. Increasingly, they’re the ones who understand enough of what’s actually happening under the hood to make the right call before they write a single line of code.

✨ SaugaTech Epilogue — Bigger wasn’t just more. It was different.

There’s a temptation to hear “175 billion parameters” and think the whole story is just about scale for its own sake. Bigger number, bigger flex. This paper is basically proof that’s not what actually happened. Scale didn’t just make the model a bit better at things smaller models could already sort of do. It unlocked something smaller models genuinely couldn’t do at all — learning a brand new skill on the spot, from a handful of examples, with zero retraining.

That’s a different kind of progress than “faster” or “cheaper.” It’s the kind that changes what a technology can actually do, period. And it’s exactly the kind of shift worth paying attention to as models keep getting bigger. The next real jump probably won’t come with a press release attached. It’ll show up quietly in a benchmark table somewhere, same as this one did back in 2020 — and the builders who notice it early will already have something shipped by the time everyone else catches up.

We’ll keep going with this series. Next time, we’re looking at how models went from just predicting the next plausible word to actually following instructions like a helpful assistant would — and the training trick that made that leap possible.

See you on August 22nd for the careers panel.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Hi SaugaTech Community,

The World Cup is over. Spain edged out Argentina 1-0 in extra time to lift the trophy, Messi’s last World Cup dream ended agonisingly short, the fan zones have packed up, Celebration Square has gone back to being just a square, and the GTA has exhaled after one of the more electric summers this city has had in a while. If you were part of any of it — the watch parties, the flags, the noise — we hope you’re still riding a bit of that high.

But the last couple of weeks have also been a reminder of something a lot less fun: nature does not care about our schedules.

Over 850 wildfires have been burning across Canada this month, more than 180 of them in Ontario alone. On July 15th, smoke drifting down from the northwest made Toronto the second most polluted major city on the entire planet for a stretch — worse air quality than cities that top that list on an average day. Skies over the GTA turned that unmistakable orange-grey. If you stepped outside and felt your throat catch, or checked an air quality app and did a double take, you weren’t imagining it.

It’s a strange kind of whiplash — the joy of a World Cup summer and the quiet menace of smoke a thousand kilometres away finding its way into our lungs. Two very different reminders, in the same few weeks, of how connected we all are to things well outside our control.

Alright. On to what we’re actually here for this week.

This week, we’re kicking off a new thread in the Compass, AI Under the Hood — one that goes underneath the AI tools most of us already use every day, into how they actually work.

Why We’re Starting This Series

Here’s something we’ve noticed across almost every SaugaTech meetup, every WhatsApp thread, every builder conversation over pizza: most of us are genuinely fluent in what LLMs can do. We know how to prompt them, chain them, wrap them in an app, ship a demo in a weekend. That fluency is real, and it’s valuable — it’s exactly what’s let this community build the things it’s built.

But if you ask the room how a large language model actually works — not the API call, not the prompt, but the mechanism underneath — the room tends to go quiet.

And that’s not a criticism. Most of us came into this AI wave through the product layer, not the research layer. There was never really a reason to go deeper. The tools worked. You didn’t need to know how an engine combusts fuel to drive a car.

Except increasingly, that’s not quite true anymore.

The gap between “I can use an LLM” and “I understand why an LLM behaves the way it does” is becoming the gap between a good AI feature and a great one.

It’s the difference between a chatbot that hallucinates unpredictably and one you’ve engineered to fail gracefully. Between a RAG pipeline that returns garbage half the time and one that’s actually reliable. Between prompting by vibes and prompting with a real model of what’s happening inside the black box.

As more of us build serious products on top of this technology that underlying understanding stops being optional.

So we’re starting a new thread in the Compass: a slow walk through the Foundational research papers of modern AI, explained in plain English, with an eye specifically toward what each one means for something you might actually build. No paper is off limits because it looks intimidating. That’s the point of doing this together.

🚀 First Things First

We are already planning for the August meetup, you’ll hear the details in next week’s Compass. And if you’re not in the WhatsApp community yet, that’s where the follow-up conversations to editions like this one tend to happen: Click here to request to join the group

⚓ The Paper: “Attention Is All You Need” (2017)

Every large language model you’ve ever used — GPT, Claude, Gemini, Llama — is a descendant of one 2017 research paper out of Google Brain, co-authored by Aidan Gomez, a 21-year-old University of Toronto student at the time. We told his story in our Canada Day edition. This week, we’re telling the story of the paper itself. If you want to read the detailed paper, here is the link to it

The Problem It Solved

Before this paper, AI language models — RNNs, LSTMs — read sentences one word at a time, left to right, like a finger tracing under each word. The model carried a “memory” forward as it read, but that memory was limited. By the time it reached the end of a long sentence, it had often half-forgotten the beginning.

Take the sentence: “The trophy didn’t fit in the suitcase because it was too big.”

What does “it” refer to — the trophy or the suitcase? A human resolves this instantly. An RNN, reading word by word, often struggled — by the time it reached “it,” the earlier context had faded.

There was a second, more practical problem: speed. Reading word by word means you can’t parallelise. You can’t throw a thousand GPUs at a sentence and process it faster, because word 5 depends on having already processed word 4. You’re stuck in a queue.

The Core Idea: Attention

The paper’s proposal: stop reading sequentially. Look at every word in the sentence simultaneously, and let the model figure out which words matter most to each other. They called this mechanism Attention — and the paper’s confident, almost cheeky title was a direct claim: you don’t need the old sequential machinery at all. Attention alone is sufficient.

Think of it like being at a loud party. You don’t process every sound in the room equally — your brain attends to the conversation that matters and tunes out the rest. When the model processes the word “it” in our trophy sentence, Attention lets it look at every other word at once and ask: which of you am I most connected to? “Trophy” scores high. “Suitcase” scores a bit lower. “Because” barely registers. The model uses those scores to build a representation of “it” that’s genuinely informed by the right context.

How It Works: The Library Analogy

The mechanics use three concepts — Query, Key, Value — that sound abstract until you picture a library.

• Query = what you’re searching for

• Key = the label on each book on the shelf

• Value = the actual content inside the book

You compare your Query against every Key in the library, score how well each matches, then pull the Value from the best matches and combine them into your answer. Every word in a Transformer generates its own Query, Key, and Value. Processing a word means scanning every other word’s Key with your Query, scoring the matches, and blending in their Values — weighted by how relevant they are.

The paper also introduced Multi-Head Attention — doing this eight times in parallel, each “head” learning to notice something different. In the paper’s own visualisations, you can see individual heads specialising: one resolves what a pronoun refers to, another tracks grammatical structure, another tracks semantic similarity. Eight specialists reading the same sentence, each noticing something different, pooling their observations into one richer understanding.

One elegant fix solves a side effect of processing everything at once: if you’re not reading in order, how does the model know “dog bites man” differs from “man bites dog”? The paper adds a mathematical position signal — sine and cosine waves — to each word before processing, so the model always knows where a word sits in the sentence, even while processing it out of order. Like numbering theatre seats before the lights go down.

Encoder and Decoder: The Two Rooms

The full Transformer has two halves. The paper was originally built for translation, so this framing helps:

The Encoder reads the input sentence. It runs the full attention mechanism — every word looking at every other word — and produces a rich, context-aware representation of the whole sentence. Think of it as a very deep understanding of what was said.

The Decoder generates the output — the translation — one word at a time. But crucially, when generating each output word, it can attend to the full encoder output. So when it’s generating the French translation of word 5, it can look back at the full English sentence and ask: which English words are most relevant right now?

The decoder also has a constraint: when generating word 5, it can only look at words 1-4 that it has already generated. It can’t peek ahead. This is called masking — and it’s what makes the model useful in the real world where you’re generating text you don’t know yet.

The Result That Made Everyone Sit Up

The paper tested on English-to-German and English-to-French translation benchmarks. The results:

• Outperformed every previous model including ensembles of multiple models

• Achieved state-of-the-art translation quality

• Did it in 3.5 days of training on 8 GPUs — a fraction of the compute cost of previous best models

That last point is as important as the quality improvement. Not only was it better — it was dramatically cheaper and faster to train. That combination is what made the research world stop and take notice immediately.

Why It Changed Everything

The Transformer architecture the paper is the direct ancestor of:

• GPT-1, 2, 3, 4 — OpenAI took the Decoder half and scaled it massively. That’s ChatGPT.

• BERT — Google took the Encoder half and used it for understanding and classification tasks. It powers Google Search to this day.

• Claude, Gemini, Llama, Mistral — every major LLM running today is a Transformer.

• AlphaFold — DeepMind used Transformers to solve protein folding, one of biology’s hardest problems.

⚓ Why This Actually Matters — Real-World Mechanics You’re Already Using

Here’s where the theory turns into something you can use in a product decision next week.

Every “chat with your documents” feature is a Query/Key/Value problem in disguise. When you build a RAG system — retrieval augmented generation, where a model searches your documents before answering — you are literally implementing a version of the library analogy above. The user’s question is the Query. Your document chunks are the Keys. The actual chunk content is the Value. If your RAG pipeline returns bad results, nine times out of ten it’s because your Keys are badly constructed — poor chunking, weak embeddings, unclear document structure — not because the underlying model is “not smart enough.” Understanding the mechanism tells you exactly where to go fix it.

Context window limits exist because Attention is computationally expensive. In the paper’s own complexity table, self-attention scales as the square of sequence length. Double the input length, quadruple the compute. That’s not an arbitrary product limitation from Anthropic or OpenAI — it’s a direct mathematical consequence of the mechanism we just walked through. When you’re deciding whether to dump your entire codebase into a prompt or be selective about what context you feed a model, you’re making a decision with real cost implications rooted in this exact table.

Multi-head attention is why the same model can write code and write poetry. Different heads specialise in different relationships. This is part of why a single Transformer-based model can flexibly handle wildly different tasks without being explicitly programmed for each one — the architecture itself is built to notice many kinds of patterns simultaneously, and which patterns matter shifts depending on what you’re asking it to do.

Hallucination is partly an Attention problem, not just a “the model doesn’t know” problem. When a model attends weakly or ambiguously to the right context — because your prompt is vague, your retrieved documents are poor, or the relevant information is buried too far back in a long context — it fills gaps with its next most confident guess. That guess is often wrong but fluent. If you’ve ever wondered why a model confidently states something false, part of the answer is sitting in the attention scores you never see.

Encoder-only and decoder-only aren’t just jargon — they’re two different extremely powerful products. That’s clearly demonstrated from the fact that Google and OpenAI used just one of them to build massive platforms. If you’re building a classification tool — sentiment analysis, intent detection, content moderation — an encoder-style model is often the right, cheaper, faster choice. If you’re building something generative, you need the decoder machinery. Knowing this distinction can save you from reaching for a giant generative model when a small classifier would do the job better and cheaper.

The founders who build the most efficient AI products aren’t necessarily the ones with the cleverest prompts. Increasingly, they’re the ones who understand enough of what’s actually happening under the hood to make the right architectural call before they write a single line of code.

✨ SaugaTech Epilogue — Understanding the Engine

As builders, each of us don’t need to be able to implement a Transformer from scratch to create great AI products. Most of the best builders in this community never will, and that’s fine.

But there’s a difference between driving a car and understanding, even loosely, why it moves. The founders who’ll build the next generation of great AI products in this region are the ones willing to spend a little time under the hood — not to become researchers, but to make better decisions the next time something breaks, scales badly, or hallucinates in front of a customer.

That’s what this series is for. We’ll keep going — next up, we’ll look at how models went from predicting the next word to actually following instructions, and what changed to make that possible.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Hi SaugaTech Community,

The FIFA World Cup has taken over. There’s no other way to put it.

If you’ve been anywhere near Celebration Square in Mississauga over the last few weeks, you’ll know exactly what we mean. Thousands of people, flags from every corner of the world, the kind of collective energy that only football at this scale can generate. Strangers becoming friends in ninety minutes. The city feeling like a village. It’s been something else.

And Canada — Canada has been right in the middle of it, on both sides of the touchline.

As a host nation, the country delivered. Toronto and Vancouver proved that Canada can run a world-class sporting event with warmth, organisation, and an atmosphere that surprised even the teams who played here.

BMO Field — the smallest venue of the 16 World Cup stadiums — punched well above its weight. Six matches, six sold-out crowds, and a buzz that spilled far beyond the stadium into Little Portugal, Celebration Square, and every fan zone across the GTA. When Portugal manager Roberto Martínez walked out of BMO Field after his side’s dramatic Round of 32 win over Croatia, he told reporters: “I think it’s a shame that there are no more games here in Toronto, because we really enjoyed that. I thought it was an incredible, incredible spectacle for football.” That’s the highest compliment a visiting manager can pay a host city — and Toronto earned it.

As a participant, Les Rouges wrote a story that will be told for a long time. Canada had appeared at two previous World Cups — 1986 and 2022 — and won nothing, scored nothing, left early both times. This time was different. Cyle Larin scored Canada’s first ever World Cup goal against Bosnia at BMO Field. Jonathan David scored a hat-trick in a 6-0 demolition of Qatar — Canada’s first ever win and the biggest winning margin in CONCACAF World Cup history. Stephen Eustáquio’s stoppage-time strike against South Africa in the Round of 32 sent Canada through to the last 16 for the first time ever. Morocco ended the run 3-0 in Houston, though the scoreline flattered them — Canada dominated the first half before the Atlas Lions found their second-half rhythm and pulled clear.

Les Rouges went home. But Canadian football will never be quite the same.

As a community built in Mississauga — one of the host cities’ neighbourhoods — we watched all of this with a particular kind of pride. And as a tech community, we couldn’t help but notice something else running through this tournament alongside the football: the most sophisticated sports technology deployment in history.

That’s what this week’s Compass is about.

🚀 First Things First

Meetup #10 recap is live — the Gandalf demo, the builders showcase, the conversation that stayed with us. Worth a read if you missed it.

And if you’re not in the WhatsApp community yet, that’s where all of this continues between editions: Click here to request to join the group

⚓ What’s Actually Running Under the Hood at FIFA 2026

Every major decision at this tournament — every disallowed goal, every reversed penalty — is being made or unmade by a tech stack most fans couldn’t name.

Start with the ball. The Adidas Trionda carries an embedded Inertial Measurement Unit — a sensor combining an accelerometer and a gyroscope, developed in partnership with Kinexon — that samples its own motion 500 times per second. That’s not a camera watching the ball. That’s the ball telling the system exactly what it’s doing, in real time, with millisecond precision. Its primary job is to detect the exact moment of contact — the precise frame at which the ball was last touched before an offside position is measured.

That’s how Croatia’s goal came off the board at BMO Field. The sensor detected that Igor Matanovic had brushed the ball before the pass that set up Gvardiol’s finish. That touch — invisible to everyone in the stadium, invisible to the broadcast cameras — set the offside clock. The ball knew. The system flagged it. Luka Modric’s World Cup was over.

Sitting on top of that ball data is Semi-Automated Offside Technology, or SAOT. Every one of the 1,248 players across 48 squads was scanned before the tournament — using Lenovo hardware and AI — to create a personalised 3D body avatar. During the game, those avatars overlay on real-time tracking data from 16 dedicated optical tracking cameras per stadium generating over 150 million data points per match, and the system calculates the offside line at the exact frame the ball is played. For 2026 the precision threshold was tightened to 10 centimetres — down from 50. That’s why you’re seeing goals disallowed for a shoulder, a toe, a backside.

Referees also wear body cameras in all 104 matches for the first time, and VAR decisions are now explained over the stadium PA — a direct response to the criticism that fans in the ground are left in the dark while TV viewers see everything.

But football didn’t pioneer any of this. Cricket did.

If you follow cricket, the FIFA tech stack will feel familiar — because cricket was solving the problem of machine-assisted officiating two decades before football got there.

Hawk-Eye was invented in 2001, originally for cricket. Six cameras placed around the ground triangulate the ball’s trajectory after pitching, letting the third umpire predict with high accuracy whether a delivery would have hit the stumps. That same Hawk-Eye system now powers FIFA’s goal-line technology, MLB’s automated ball-strike system, and — since 2025 — every line call at Wimbledon, where human line judges have been eliminated entirely. SAOT is a direct descendant of what cricket built first.

Hot Spot uses infrared imaging to detect the heat generated by friction when ball meets bat or pad. Invisible to the naked eye, visible to infrared. The physics that cricket used to adjudicate edge catches in Test matches are now embedded in wearable sports tech measuring muscle exertion and contact impact across multiple sports.

Ultra Edge — Hawk-Eye’s upgrade to the original Snickometer — uses a sensitive microphone in the stump and oscilloscope technology to detect the acoustic signature of ball-on-bat contact. A sharp spike on the waveform confirms the nick. Sound-based contact detection, refined in cricket, is now standard technique across sports officiating.

Cricket, a sport most of North America barely watches, was solving the hard problem of sensor-based officiating long before the World Cup came to Toronto. Sports tech rarely invents from scratch. It inherits, adapts, and scales.

⚓ Beyond Football — Tech That Silently Runs behind Every Major Sport

The World Cup makes sports tech visible. The more interesting work has been running quietly across every major league for years.

Hockey — Every puck in NHL play carries a sensor. Every player wears a tag in their shoulder pads. Cameras feed data to a central system tracking puck position, player speed, shot release time, and zone entries multiple times per second. That data feeds coaching dashboards and powers everything from line-change timing to power play deployment. Catapult and STATSports provide the wearable GPS layer used by teams across multiple sports.

NBA — The league committed to optical player tracking at scale before anyone else. Second Spectrum, the platform that powers NBA.com player tracking, uses computer vision to track every player’s position on the court multiple times per second. The advanced metrics you see on NBA broadcasts — Expected Possession Value, defensive impact scores, passing network analysis — all flow from this system.

Baseball — The Automated Ball-Strike system, built on Hawk-Eye cameras, has been running in the minor leagues for years and is slated for the 2026 ALCS and NLCS. The era of “the umpire’s zone” as a variable that batters and pitchers manage is ending.

Injury prevention — Zone7, an AI injury prediction platform, has been adopted by Premier League clubs, NBA teams, and MLB franchises because the cost of a torn ACL — in contract value and competitive impact — now vastly exceeds the cost of a predictive analytics subscription.

The sports technology market was worth around $3 billion in 2025. It’s projected to reach $10 billion by 2030, growing at over 25% annually. Every sprint, every heartbeat, every puck bounce is data now.

⚓ The GTA Companies Building It

Here’s where it gets local.

Toronto Metropolitan University’s Future of Sport Lab, in partnership with Maple Leaf Sports and Entertainment, is one of the most significant sports tech incubators in Canada. Startups from the FSL program have collectively raised over $100 million.

A few worth knowing:

Pando Partner (Toronto) — a self-serve platform that gives sports and entertainment organisations the tools to manage and measure sponsorship performance. Sponsorship is a multi-billion dollar market that has historically run on relationships and gut feel. Pando is building the analytics layer that makes it data-driven.

TritonWear (Toronto) — wearable performance technology specifically for competitive swimmers. Swimming is over a decade behind other sports in data collection. TritonWear is closing that gap with sensors worn during training that capture stroke mechanics, turn times, and exertion data that coaches have never had access to before.

theScore (Toronto) — now owned by PENN Entertainment, theScore is one of the most widely used sports media apps in North America, sitting at the convergence of sports media and fan engagement — one of the defining commercial trends of the decade.

Sportlogiq (Montreal — close enough) — trusted by all 31 NHL teams, Sportlogiq uses AI-powered computer vision to turn broadcast video into structured data. Teams that can’t afford dedicated sensor infrastructure can still get player tracking-quality insights from a camera feed.

✨ SaugaTech Epilogue — The Referee You Can’t Argue With

There’s a version of the Argentina-Egypt controversy that has nothing to do with Messi’s star power. It’s a version where the technology worked exactly as designed. A foul was detected. The goal was disallowed. Argentina scored three times and went through.

The technology didn’t care about any of that. It measured. It flagged. It rendered a verdict.

What the World Cup has made visible — through every disallowed goal, every reversed penalty, every 3D avatar graphic on the stadium screen — is that the marriage of sport and technology is now permanent and irreversible. The question is no longer whether to use it. The question is who builds the next layer.

Sports is a $500 billion global industry running on data that most organisations don’t fully understand yet. The infrastructure exists. The talent to build on top of it exists right here in the GTA. And the problems — better fan engagement, smarter coaching tools, more accurate officiating, deeper athlete monitoring — are concrete, specific, and unsolved.

If you’re working in data, computer vision, machine learning, or wearable hardware and you’ve ever thought about applying those skills to sport, now is the time. And if you’re building something in this space and want a room that will take it seriously — reach out. The next Builders Showcase slot might have your name on it.

We’ll see you at the next meetup very soon.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Hi SaugaTech Community,

It’s July 1st. Wish everyone a Happy Canada Day.

Most of us will spend some part of today with family, by a barbecue, watching fireworks over the lake. And somewhere between the burgers and the red-and-white, we’ll feel that quiet pride in being Canadian. The politeness, the multiculturalism, the sense that this country genuinely tries to be something decent in the world.

But here’s the thing that doesn’t get talked about enough at the Canada Day barbecue: Canada also builds.

Canadian entrepreneurs, engineers, innovators and researchers have quietly produced some of the most consequential technology of the last twenty-five years. We built the foundation that modern AI runs on. We produced a platform that powers millions of the world’s online businesses. We launched a company that put the idea of enterprise mobile communication on the map - years before anyone had heard of the iPhone.

And they did most of this without the flags and the fanfare.

That’s the Canadian way.

This Canada Day, we wanted to do something a little different with the Compass. No global tech drama this week. Just ten people and companies from this country who built something worth knowing about - in this millennium, in tech, in a way that matters.

This is your Canada Day reading list. Pour something cold. Let’s get into it.

⚓ Canada Built This — Ten Innovators Worth Knowing

1. Geoffrey Hinton — The Godfather of AI, Based in Toronto

If you’re reading this on a smartphone, using a voice assistant, or working with any AI tool, you are using technology that Geoffrey Hinton helped make possible.

Hinton came to the University of Toronto in 1987, at a time when neural networks were widely dismissed as a dead end. He kept working anyway. In 2012, his team — including his graduate students Alex Krizhevsky and Ilya Sutskever — built a neural network that won the ImageNet competition and cut the error rate in half. Overnight. That moment is now considered the starting gun of the modern AI era.

In 2018, he won the Turing Award, computing’s highest honour. In 2024, he won the Nobel Prize in Physics. He is known, simply, as the Godfather of AI.

He did it from Toronto.

2. Yoshua Bengio — The Most-Cited Computer Scientist in the World

Montreal-based Yoshua Bengio is the most-cited computer scientist on earth. Not in Canada. On earth.

Alongside Hinton, he pioneered the deep learning techniques that underpin virtually every major AI system in use today. Asked if he could draw a direct line from his work to ChatGPT, he said, simply: “Yeah, definitely.”

He won the 2018 Turing Award with Hinton and Yann LeCun. He founded Mila, the Quebec AI Institute, which has become one of the world’s leading AI research hubs. He now runs LawZero, focused on building AI that is safe by design.

Two of the three people most responsible for the AI revolution are Canadian. That fact deserves more barbecue conversation than it gets.

3. Tobi Lütke — Shopify

Tobi Lütke came to Canada from Germany, settled in Ottawa, and wanted to sell snowboards online. He couldn’t find a platform that worked well enough, so he built one.

That platform became Shopify — now one of the most valuable companies in Canada, powering over two million businesses across 175 countries. If you’ve ever bought something from an independent brand online, there’s a reasonable chance Shopify was the infrastructure underneath it.

Lütke is also one of the most thoughtful public voices on AI and the future of work, famously announcing that Shopify employees are expected to treat AI as a core tool rather than an optional add-on. He’s not waiting for the future. He’s building it.

4. Mike Lazaridis — BlackBerry / Research In Motion

Before the iPhone, there was the BlackBerry. Before the BlackBerry, there was Mike Lazaridis.

Starting with a $15,000 student grant in Waterloo, Ontario, Lazaridis co-founded Research In Motion and built the device that made enterprise mobile communication mainstream. At its peak, BlackBerry controlled more than 50% of the US smartphone market. Barack Obama refused to give his up.

BlackBerry’s moment at the top didn’t last. But its impact on Waterloo did. Lazaridis invested $100 million in the Perimeter Institute for Theoretical Physics and poured resources into quantum computing research. The tech ecosystem that now defines the Waterloo-Toronto corridor exists, in no small part, because of what he built and what he chose to put back.

5. Garrett Camp — Uber Co-Founder

Garrett Camp was born in Calgary. He studied at the University of Calgary. And then he co-founded two companies that changed how the world moves through cities and discovers things online.

The first was StumbleUpon, one of the earliest web content recommendation systems. The second — after a night of frustrating cab rides in Paris — was Uber. Camp conceptualised the idea, brought in Travis Kalanick, and helped transform urban transportation globally.

There’s a version of this story where Garrett Camp’s name is as well known as any other major Silicon Valley tech entrepreneur. It isn’t, partly because he’s Canadian. He keeps building and investing, mostly without the noise.

6. Aidan Gomez, Nick Frosst & Ivan Zhang — Cohere

In 2017, a team at Google Brain published a research paper that would become one of the most important documents in the history of computing. It was called “Attention Is All You Need.” It introduced the Transformer architecture — the foundation on which GPT, Claude, Gemini, and virtually every large language model is built. One of the co-authors was Aidan Gomez. He was 21 years old and studying at the University of Toronto.

Gomez went on to co-found Cohere with Nick Frosst — a fellow U of T graduate, former Google Brain researcher, and in a different life, a musician — and Ivan Zhang, who built the technical foundations of Cohere’s language model infrastructure. Together, they built the company Canada’s AI ambitions are being bet on: a Toronto-based enterprise AI platform competing at the highest level globally. Born right here in the 416.

7. Richard Sutton — The Godfather of Reinforcement Learning

If Hinton and Bengio built the foundation of deep learning, Richard Sutton built the foundation of how AI systems learn to make decisions. His 1998 textbook on reinforcement learning — co-authored with Andrew Barto — is still the definitive reference in the field. Every AI system that learns by doing, from AlphaGo to the models that train modern robotics, traces its lineage to Sutton’s work.

He spent decades at the University of Alberta in Edmonton, turning down opportunities abroad to build one of the world’s great AI research programs right here in Canada. Amii — the Alberta Machine Intelligence Institute — grew directly from that commitment. Sutton is proof that world-defining research doesn’t require a prestigious American zip code. It requires someone who believes deeply enough to stay and build.

8. Ryan Holmes — Hootsuite

Ryan Holmes didn’t set out to build a global software company. He built a tool to manage his own social media accounts more efficiently — and then realised the problem was universal.

That tool became Hootsuite, launched in Vancouver in 2008, and it essentially invented the social media management category before most businesses had a social media strategy at all. At its peak, Hootsuite served over 18 million users across 175 countries and became one of Canada’s most recognisable B2B SaaS brands globally. It proved that enterprise software with worldwide reach could be built from Vancouver just as easily as anywhere else — and that Canadian founders could define an entire product category, not just compete in one.

9. Shopify (as a company) — The Blueprint

Shopify deserves its own entry as a company, separate from Lütke, because of what it represents as a template.

It was built in Canada, for the world, headquartered in Ottawa, serving two million businesses across 175 countries. It proved that when Canadians back themselves, they can build at the highest level globally — and do it on their own terms.

The infrastructure is here. The talent is here. The next Shopify can absolutely be built right here — in Mississauga, in Brampton, in Oakville, anywhere a founder decides to start.

10. The Canadian AI Ecosystem — Mila, Vector, Amii

The final entry isn’t a person. It’s a bet Canada made before anyone else did.

In 2017, Canada became the first country in the world to launch a national AI strategy. The Pan-Canadian AI Strategy invested $125 million across three institutes: Mila in Montreal (Bengio’s house), the Vector Institute in Toronto (Hinton’s house), and Amii in Edmonton (Sutton’s house).

These institutes created the conditions for talent to stay, build, and compound right here in Canada. They created the conditions in which Cohere, Ada Support, and dozens of other AI companies were born. They were the room that the talent needed.

SaugaTech is not a national AI strategy. But the instinct is the same.

✨ SaugaTech Epilogue — The Only Thing Standing Between Us and Greatness

One of our team visited the RCAF Museum in Trenton this week. Walking through those hangars, past the artifacts of Canada’s Cold War aviation program, one story kept coming back.

In 1957, Canada unveiled the Avro Arrow — a supersonic interceptor that was faster, more technically advanced, and more capable than anything else flying at the time. Commercial supersonic jets like the Concorde were still a decade away. Canada had already built one for the military. The engineers who made it were among the best aerospace minds in the world.

When Canadians put their minds and belief behind something, they built the best technology on the planet. Full stop.

The Arrow proved it beyond any doubt.

The program was eventually cancelled — a decision that belongs to history. But the proof of what Canada can do when it backs itself? That’s permanent. No one can cancel that.

That same proof has been written again and again since.

Hinton stayed in Toronto and kept working on neural networks when the rest of the world said it was a dead end — and rewrote the future of computing. Lütke bet on himself in Ottawa and built one of the most important companies in the world. Lazaridis built BlackBerry in Waterloo and put enterprise mobile communication on the map. Gomez co-authored the paper the entire AI industry runs on, at 21, at U of T.

The talent has never been the problem. It never was.

The only thing standing between us and the next chapter of Canadian innovation is belief. Every great Canadian tech story started with someone who decided their idea was worth it, their city was enough, and their moment was now. That decision is available to all of us.

We have the talent. We have the track record. We have each other.

Let’s build like it’s Canada Day every day - with pride in our legacy, with inspiration from the Canadian trailblazers who have done it, time and again.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Share

That’s what made Meetup #10 feel different before it even started. The last few sessions — Vibe Coding, Agentic AI — drew people riding the wave of something new and shiny. This one drew people who wanted to understand something real. Security isn’t a trend. It’s the thing that keeps your product standing once the hype settles. The room knew that. You could feel it in the conversations before we even called things to order.

But it wasn’t only the topic that pulled people in. A lot of folks showed up as much for each other as for the content — techies with curious side projects and no shortage of opinions, finally in a room full of people like them. That social energy, the easy back-and-forth between strangers who clicked fast, is becoming one of SaugaTech’s signatures. And on a warm Saturday afternoon in the first meetup of the summer, it was especially alive.

Meetup #10 also happened to be a milestone. And we treated it like one.

Meetup #10 — A Milestone That Means More Than the Number

Ten meetups. It sounds like a number. But for those of us who were in that first room at RiverGrove Community Centre in April 2025 — a handful of strangers who found each other through a Reddit post — it feels like something much more than that.

We started with a simple frustration: Mississauga had the talent, the ambition, and the energy. What it lacked was a place to come together. Fourteen months later, we have 250+ members on WhatsApp, 500+ subscribers on the Compass newsletter, and a community that keeps showing up — not because they have to, but because they want to.

Meetup #10 felt like the right moment to pause, look back at how far we’ve come, and — more importantly — be explicit about where we’re going.

Why are we telling you all of this?

Because we believe a community deserves to know what it’s building toward. Transparency isn’t just a value — it’s a practice. If our vision resonates with you, we want you to lean in. And if you think we’re missing something, getting something wrong, or need a course correction — we genuinely want to hear it. This community is not ours to steer alone. It belongs to everyone in that room, and everyone who reads this.

Defending the Prompt — The Room Gets Honest About AI Security

We’ve talked about AI at almost every SaugaTech session in some form. But Meetup #10 was the first time we went deep on something most builders quietly worry about but rarely discuss openly: what happens when your AI-powered product gets attacked?

Aries Youssefian and Samiul Islam opened the session — Aries from a systems architecture perspective, Samiul from a prompt engineering and model interaction lens. Between them, they’ve spent hundreds of hours building with, breaking, and securing AI systems in real environments. And true to the SaugaTech spirit, this wasn’t a lecture. It was a conversation.

We talked about how AI systems introduce attack surfaces that traditional security frameworks simply weren’t designed for. We talked about the gap between AI safety as a research field and AI security as a day-to-day engineering problem. We talked about what it actually means to secure a product that depends on a third-party model you don’t fully control.

And then we actually tried simulating an attack - with Gandalf.

For those who haven’t tried it — Gandalf by Lakera is a game that simulates prompt injection attempts. The premise is simple: try to trick an AI into revealing a secret password by manipulating its instructions. What sounds like a fun party trick quickly becomes a masterclass in how fragile AI guardrails can be — and how creative attackers can get.

Watching a room full of tech professionals try to outwit an AI in real time was equal parts hilarious and sobering. The laughter was genuine. So was the realisation that followed: if we can do this in a game, someone is doing this to production systems right now.

That’s the kind of learning that sticks.

Builders Showcase — Three Products, Three Stories

Post-pizza, we handed the floor to three builders from our community. Ten minutes each. Real things being built — right here in the GTA.

🎁 Shreshtha Agarwal — Gift4u.ca

Shreshtha is building Gift4u.ca — Canada’s own AI-powered gift registry. The idea is elegant: create a gift registry, get AI-generated suggestions based on the recipient’s interests, and use direct links to buy the selected products. No more awkward “what do you want for your birthday?” conversations. No more gift cards that never get used.

What stood out was the clarity of the problem being solved. Gift-giving is universal, the friction is real, and the AI layer adds genuine value rather than being bolted on for the sake of it. The room had questions — and Shreshtha had answers.

🍽️ Ofir David — Simpa

Ofir is building Simpa — a community-powered platform that helps restaurant-goers find restaurants that match their dietary restrictions. Anyone who has navigated a dinner out with a mixed group of dietary needs knows exactly how painful this problem is.

What made Simpa interesting is the community angle. This isn’t just a filter on a restaurant app. It’s a platform where the community itself contributes to the knowledge base — making it richer and more reliable over time. The GTA, with its extraordinary diversity of cuisines and dietary communities, is exactly the right place to build this.

📈 Mahesh Krishnamurthy — VantagePier

Mahesh talked about his experience of building VantagePier, a vibe-coded platform for stock and portfolio analysis built for equity investors. Importantly, Mahesh built this product using Antigravity after getting inspired from our last meetup on Vibe Coding. That in a nutshell fortified our belief why spaces like SaugaTech are important, to see and get inspired to do things that we otherwise we would have never attempted.

The Conversation That Stayed With Us

After the builders wrapped up, the conversations spilled into the room — the kind that don’t end when the session does.

One exchange in particular stuck.

We were talking about how the barrier to building has never been lower. Everyone is a tech person now. With the right AI tools, you can put together a great demo — even a working application — faster than ever before. The three builders we saw in this meetup are proof of that.

But here’s the thing: a Working app and a Scalable app are two very different things.

The real differentiator — the thing that separates a demo from a product that can handle thousands of simultaneous users without breaking, leaking data, or being manipulated — is architecture. It’s how you think about security from day one. It’s how you design for scale before you need it.

And that, in a way, is exactly what this meetup was about. The security session in the first half and the builders showcase in the second half weren’t two separate conversations. They were the same conversation — just approached from different angles.

Build fast. But build right.

Thank You

To Aries and Samiul — for an honest, practical, and genuinely fun session. The Gandalf demo alone was worth the trip.

To Shreshtha, Ofir, and Mahesh — for having the courage to stand up in front of a room and show what you’re building. That takes guts. This community sees you.

To everyone who showed up on a Saturday afternoon — thank you. Forty-five people choosing to spend their weekend learning and connecting is not something we take for granted.

And if you’re building something and want a room that will take it seriously — reach out. The next Builders Showcase slot might have your name on it.

Who knows, maybe you’ll run into someone building the next Shopify, right here in Mississauga.

See You at the next one.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

P.S. Missed this session? Join our WhatsApp community to stay in the loop on upcoming events (message us on +14377992290 or saugtechproduct@gmail.com). Follow us on LinkedIn.

Github Link to Session Materials

Hi SaugaTech Community,

If you’re in Mississauga this week, Celebration Square is buzzing. The patios are full. The big screens are going up. The FIFA World Cup kicks off on June 11 — and Canada plays its opening match on Friday right here in Toronto at BMO Field. For a community as diverse as ours, spread across every corner of the western GTA, this one feels personal. The biggest sports tournament of the planet, hosted right in our backyard for the first time in history.

It’s a good week to be in Mississauga.

And while everyone’s watching football, something equally significant is happening in the world of finance and technology — three events in the same week that together paint a fascinating and slightly contradictory picture of where AI actually stands right now.

Here’s the question sitting underneath all of it, one that’s been floating around our WhatsApp group for a while: if AI is genuinely the most transformative technology in a generation, why are so many of the people actually using it still struggling to put a number on what it’s given them? And if enterprises are burning through AI budgets faster than they planned, if ROI is frustratingly hard to pin down — then why are the biggest AI companies on the planet rushing to go public at trillion-dollar valuations right now?

Both of those things are true at the same time. And the tension between them is the most interesting question in tech this week.

Three things happened in the last ten days that each answer the “what is AI worth?” question differently. A historic IPO priced. An enterprise cost crisis deepened. And Canada placed a $2.3 billion national bet. Same question. Three very different answers.

Grab a coffee. Let’s get into it.

🚀 First Things First: June 20 Meetup

We’re back at IDEA Mississauga, our theme this time is Building Secure Tech Products in the AI Era. Aries Youssefian and Samiul Islam open the session from architecture and prompt engineering perspectives, then we open it up to the room.

Post-pizza, we have a Builders Showcase — three 10-minute slots for community members to share what they’re building. Startup, side project, personal tool — anything real goes.

📅 Saturday, June 20, 2026

📍 IDEA Square One, Mississauga

🕐 1:00 PM – 3:00 PM (Doors open 12.30 pm)

Grab a spot if you haven’t already, we are almost at capacity.

More details & RSVP Here

Now coming back to the AI RoI question that we are trying to find answers to.

What Markets Say: The AI IPO Queue Is Open For Business

Something remarkable happened this week that got a little lost in the World Cup noise.

On June 1, Anthropic filed confidentially with the SEC for an IPO, targeting an October 2026 listing at a valuation of approximately $965 billion — just days after closing a $65 billion funding round that pushed it ahead of OpenAI in private market value for the first time. Days later, on June 8, OpenAI followed — filing its own confidential IPO paperwork, reportedly targeting September at a valuation approaching $1 trillion. And this Thursday, June 12 — the same day Canada plays its World Cup opener — SpaceX debuts on the Nasdaq at $135 a share, a market cap of $1.75 trillion, raising $75 billion in the largest IPO in history.

Three of the most powerful AI entities on the planet. Three IPOs. All within months of each other.

Now here is the part nobody is saying loudly enough.

OpenAI currently projects a $14 billion operating loss in 2026 alone, with no profitability expected until around 2029 or 2030. For every dollar it earns, it spends $2.22. Anthropic’s revenue trajectory is striking — annualised revenue grew from $10 billion last year to $47 billion by May 2026 — but with a $65 billion funding round just closed and billions more committed to AWS infrastructure, its burn rate is not modest either.

So why are they all going public now?

The honest answer is: because they have to. Private funding rounds at these scales have a ceiling. The infrastructure commitments these companies have made are staggering — OpenAI alone has committed to the $500 billion Stargate program, a $300 billion Oracle capacity agreement, and a $250 billion Azure services commitment through 2032. You cannot fund that from venture capital indefinitely. Public markets are the next pool of capital large enough to keep the machines running.

This is not cynicism — building frontier AI genuinely requires this kind of investment. But it is worth being clear-eyed about what these IPOs actually are. They are not exits for founders who have built profitable businesses. They are capital raises for companies spending enormously in the belief that the returns will eventually justify it. The public market investor is being asked to fund the gap between where these companies are today and where they need to be by 2030.

Markets will almost certainly say yes on Thursday. The question that connects directly to what enterprises are experiencing on the ground is whether the revenue to justify those valuations actually materialises. And that depends on something the IPO roadshows are spending surprisingly little time on: whether enterprises can figure out how to generate real, measurable ROI from AI before the patience of public market investors runs out.

What Enterprises Are Paying: The Token Bill Nobody Budgeted For

Here’s something that sparked a real debate recently in the tech circles.

Uber burned through its entire 2026 AI coding budget by April.

Not December. April.

The company had rolled out Claude Code to 5,000 engineers in December 2025. By April, 70% of code commits were AI-driven. And yet Uber’s COO admitted he couldn’t draw a direct line between rising token consumption and the consumer features actually being shipped.

“If you’re not actually able to draw a direct line to how many useful features and functionality you’re shipping to your users, that trade becomes harder to justify.”

This is the enterprise AI cost crisis that nobody is writing about in the same breath as trillion-dollar IPOs — but probably should be. Per-token prices have actually fallen 98% since 2022. Yet enterprise AI bills have risen by an estimated 320%. The cheapest production models cost around $0.04 per million tokens. The most expensive frontier reasoning models cost upward of $180 per million tokens. That is a 4,500x pricing spread. Most enterprises are defaulting to the most capable model for every task with minimal governance, routing logic, and no cost visibility.

The core problem is that tokens are nothing like the software seats enterprises have been buying for decades. Tokens scale nonlinearly. In agentic workflows — where AI is calling other AI, looping through tasks, generating intermediate outputs — a simple linear workflow that cost $0.04 per interaction in 2023 now costs roughly $1.20 in an orchestrated agentic system — about 30 times more. The pilot that looked affordable at 10 users becomes unrecognizable at 1,000.

Now connect this back to the IPO story. OpenAI and Anthropic are asking public markets to fund years of losses on the belief that enterprise AI spending will compound massively. But the enterprises that would generate that revenue are currently discovering their AI budgets don’t stretch as far as they thought. A Gartner study found that even as inference costs fall 90% by 2030, cheaper tokens won’t translate to cheaper enterprise AI — because agentic models require far more tokens per task, and providers won’t fully pass through lower costs. Something has to give — either enterprise ROI becomes demonstrable, or the trillion-dollar valuation story becomes very hard to sustain.

What Canada Just Decided: A $2.3 Billion National Bet

Last Thursday, June 4, Prime Minister Carney stood in Toronto and launched AI for All — Canada’s national AI strategy. The strategy contains $2.3 billion in spending and sets a goal of increasing Canadian business AI adoption from 12% to 60% by 2034.

Read that again. Twelve percent.

Canada — home to Geoffrey Hinton, Yoshua Bengio, and the foundational research that made modern deep learning possible — currently has only 12% of its businesses using AI tools. While Canada has world-class talent and one of the fastest-growing digital sectors in the G7, it is among the slowest countries to adopt AI at scale. This gap risks driving Canadian talent and startups abroad, and leaving critical parts of the ecosystem under foreign control.

The strategy’s most striking element isn’t the spending — it’s the structure. Canada will take equity stakes in its most promising AI firms, accelerating the creation of national champions. The $500 million Canadian Tech Growth Fund will provide growth capital and occasional federal equity investment in Canadian AI companies — with the explicit goal of keeping Canada’s best companies Canadian rather than watching them scale abroad.

The mixed reviews are fair. Critics note the strategy is light on concrete safety timelines and enforcement architecture, and the 12% to 60% adoption target spans eight years — mostly beyond any current government’s mandate. But the direction is clear. Canada looked at the SpaceX IPO, looked at the enterprise AI spending numbers, and decided: we are not going to be spectators in this transition.

For GTA founders, that is not nothing.

What This Means for Builders Across the GTA

Three answers to the same question. Taken together they point at something concrete for those of us building in Mississauga, Brampton, Oakville, and across the western GTA.

The valuation gap is a product opportunity. Markets pricing AI at $1.75 trillion while enterprises struggle with ROI creates a window. The companies that figure out how to deliver measurable, predictable AI value — not impressive demos but actual business outcomes with defensible unit economics — will be the ones that survive the inevitable valuation reckoning. That is a product problem. Product problems are solvable.

Token economics are the new unit economics. Your cost model is no longer seats times price. It is tokens times usage pattern times model choice. Founders who design for efficiency from day one — right model for the right task, not defaulting to the most expensive option — will have meaningfully better businesses than those who don’t.

Canada’s strategy is a door worth knocking on. The $500 million Canadian Tech Growth Fund will be looking for companies to back. A government mandate to take Canada from 12% to 60% AI adoption means there is real demand for companies that help Canadian businesses use AI effectively. For GTA founders with real enterprise traction, this is worth paying attention to.

The ROI conversation is ours to lead. Most Canadian businesses haven’t made their AI decision yet. The 12% adoption number tells us the room is still mostly empty. The people who show up with honest, grounded thinking about what AI delivers and what it actually costs will shape that conversation. Being informed in a room where most people aren’t is a genuine advantage — and it is exactly what this community is built to provide.

✨ SaugaTech Epilogue

On Thursday, when SpaceX starts trading on the Nasdaq and Canada runs out for its World Cup opener at BMO Field, two very different kinds of bets will be placed at almost the same moment.

One bet is on AI transforming everything, at $1.75 trillion. The other is on a team playing at home for the first time.

Both carry the same underlying belief: that showing up prepared, on the right stage, at the right moment matters. That the gap between where you are and where you could be is real — but crossable.

OpenAI is projecting a $14 billion loss this year. Anthropic is burning through billions while racing to go public. Neither is profitable. Both are betting that the enterprise ROI story gets written in their favour before public market patience runs out. They might be right. The infrastructure they’re building is genuinely extraordinary.

But the enterprises paying the token bills, the Canadian businesses sitting at 12% adoption, and the GTA founders figuring out their unit economics — they are the ones who will actually write that ROI story. Not in S-1 filings or on Nasdaq roadshows. In real workflows, real products, and real decisions made in real organizations.

That’s the ground level. And that’s where this community operates.

See you on June 20. And go Canada. 🍁⚽

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Share

Summer has finally arrived in Mississauga!

The splash pads are open, people are swimming in the lake, the vibe has finally shifted from chilly to chilled out. Before you completely lose yourself to patio season and

We’re gathering the community for Defending the Prompt - Building Secure Tech Products in the AI era. We’ll be diving into new threat landscapes, prompt injections, and how to secure those third-party AI integrations.

Click here to RSVP

The Details:
📅 Date: Saturday, June 20th
🕐 Time: 1:00 PM – 3:00 PM
📍 Location: IDEA Square One (100 City Centre Dr, 2nd Floor)
🥤 Vibe: Casual. Light refreshments will be provided!

We’re also hosting a Builders Showcase! Three of our community members will be doing quick 10-minute demos of what they’ve been quietly building behind the scenes. (Got a side project or startup idea? Write to us on saugatechproduct@gmail.com to claim a slot!)

As always, the session is designed for all skill levels. Its also a unique opportunity to:

• Network with local founders, senior engineers, PMs and tech leaders.

• Learn from community members by hearing their real world experiences

• Contribute to the growth of the Mississauga and GTA tech ecosystem.

Stay cool and see you there,

Cheers

Team SaugaTech

Hi SaugaTech Community,

It’s been a little while since the last Compass landed in your inbox.

Life happened — for a lot of us in the volunteer group at once. Work sprints, family commitments, a hospital trip or two. The WhatsApp group kept buzzing the whole time, which tells you everything about this community. But sitting down to write something that does this community justice takes time and a certain kind of headspace. We needed both.

We’re back. And we’re starting with a question that’s been sitting quietly underneath every AI conversation we’ve had this year — and recently surfaced in the WhatsApp group with some real heat.

How much is all of this actually costing us?

Not in dollars.

In water.

In electricity.

In the quality of life of communities that never asked to host the infrastructure powering the AI revolution — and are now finding out what that means for their energy bills, their water supply, and their neighbourhoods.

Here’s what pushed this to the top. Last week, Pope Leo XIV released Magnifica Humanitas — his first major theological document, devoted entirely to AI. The leader of 1.3 billion Catholics warned that control of AI must not remain in the hands of “a few,” called for rigorous ethical constraints on its use in warfare, and argued that the technology is actively fuelling global conflict. He called for the “disarming” of AI. Warned that “new forms of slavery” are tied to its rise. Cautioned against a race for ever more powerful algorithms driven by “the desire to secure geopolitical or commercial dominance.”

You don’t have to be Catholic — or religious at all — to find this significant. The moral leader of the world’s largest religious institution didn’t choose climate change, poverty, or conflict as the defining issue of his papacy. He chose AI. Because the question he’s actually asking isn’t a theological one. It’s a human one.

What are we willing to pay for this? And who decides?

Grab a coffee. Let’s get into it.

🚀 First Things First: June 20 Meetup

We’re back with our next meetup on Saturday, June 20 at IDEA Mississauga, Square One.

We’re still shaping the topic and format — and we want to hear from you. What do you want to learn, debate, or build? If you have something to share — a project, a perspective, hands-on experience with AI tools — we’d love to hear from potential facilitators too.

Reach out at saugatechproduct@gmail.com or drop a message in the SaugaTech WhatsApp Group (Note please email or message us on +14377992290 if you want to be added to the whatsapp group)

This community grows when you bring your ideas into the room.

The Infrastructure Underneath Everything

Before we get into costs, it helps to have a sense of what we're actually talking about.

Every time you run a prompt, generate an image, ask an AI to review your code — somewhere, a data centre processes that request. Servers, GPUs, cooling systems, power infrastructure. None of the things we talk about in this newsletter exist without them.

And they are growing at a pace most people haven’t reckoned with.

According to the IEA, global data centre electricity consumption in 2024 was roughly 415 terawatt hours — about 1.5% of all the electricity consumed on earth. That number is growing at four times the rate of global electricity consumption overall. By some projections, data centres could consume more electricity by 2026 than Japan does today.

In the U.S., data centre energy demand is expected to nearly double between 2025 and 2028. That’s the energy equivalent of adding an entire Spain to the grid in three years.

Closer to home: Ontario’s grid operator projects commercial data centre demand in the province will grow more than 400% by 2050. There are already 94 data centre locations across Mississauga and the broader GTA — one of Canada’s densest clusters — with more development spreading through Markham, Etobicoke, and Cambridge.

The community most of us call home is already at the centre of this story.

Three Costs That Don’t Show Up anywhere

Here’s the thing about how data centres get discussed: the conversation almost always stays at the level of the company building them. Their capital investment. Their energy contracts. Their sustainability pledges. What gets far less attention is what happens to everyone else in the vicinity — the costs that fall outside the transaction entirely.

There are three that are worth understanding clearly.

The grid is shared — and upgrades aren’t free. When a hyperscale data centre connects to the provincial grid, it draws on transmission infrastructure built over decades by ratepayers. The new capacity required to serve that demand has to be built — and in most jurisdictions, those upgrade costs get distributed across all electricity users, not billed to the company that triggered the need. The result is visible in the data: electricity prices in U.S. regions with high data centre density jumped 267% over five years, well above the national average. Nearly three-quarters of voters in Virginia — where data centres are heavily concentrated — now blame the facilities directly for their rising energy bills. The companies paid their contracts. Their neighbours paid for the grid to grow.

Water isn’t a utility you can scale. You can build more transmission lines. You can’t manufacture more groundwater. Data centres use enormous volumes of water to cool their servers — and in many regions, that consumption competes directly with agricultural use, municipal supply, and local ecosystems. The numbers are hard to pin down precisely because most tech companies don’t disclose water use at a facility level, but researchers estimate that a single AI chatbot conversation — 10 to 50 responses — uses roughly half a litre of water. At the volume AI now operates, that stops being a rounding error. Amazon’s Project Rainier in Indiana — one of the largest AI facilities ever built — drew so much water during construction that nearby residents reported their wells running dry. Indiana state officials opened a formal investigation. Amazon paid its water bills. The neighbours lost access to their water supply.

Community disruption doesn’t appear on a balance sheet. Land gets acquired, permits get filed, construction begins. The communities living closest to these facilities — often rural, often lower-income, often communities of colour or Indigenous communities — rarely have meaningful input into whether they want to host this infrastructure, or on what terms. No invoice captures that cost. No sustainability report accounts for it. It simply falls on the people with the least power to object.

These three costs — shared grid burden, water as a commons, and the absence of consent — are the actual mechanism behind Pope Leo’s concern. Not that AI companies are villains. But that a system optimised for private benefit can generate public costs that no one explicitly agreed to bear.

The Canadian Picture

Canada is moving aggressively to build AI infrastructure, and the rationale is genuine. Ontario, Quebec, Alberta, and BC are treating data centres as core national infrastructure — turning AI strategy from policy documents into physical construction.

But the Canadian story has its own version of the same tension.

Kevin O’Leary’s proposed data centre in drought-stricken Alberta was put forward without the consent of an affected First Nation. In Alberta more broadly, data centre growth risks driving up power costs if new gas plants or transmission upgrades run over budget — with tech companies potentially locking in favourable rates that shift the cost of grid expansion to general ratepayers. Calgary restricted water use in both 2025 and 2026. The Alberta Utilities Commission dismissed one data centre proposal in March 2026, citing unresolved questions about whether its requirements had been met.

Ontario responded with Bill 40 — requiring data centres to obtain approval before connecting to the grid, ending automatic sign-off for large energy-intensive loads, and prioritising projects that deliver measurable community benefits, create quality jobs, and support data sovereignty. It’s a meaningful shift in who gets to ask questions before the concrete is poured.

Microsoft, which has committed $19 billion in Canadian data centre investment between 2023 and 2027, has explicitly promised to cover the full cost of electricity including new generation and grid upgrades — and to minimise water use — in direct response to community pressure. That’s a real commitment. It’s also a signal of how hard communities had to push to get it.

The Question Underneath All of This

Here’s the tension Pope Leo was actually pointing at.

AI is delivering real benefits. Drug discovery is accelerating. Climate modelling is improving. Diagnostic tools are reaching patients who previously had no access to specialist care. The productivity gains for engineers, writers, researchers, and builders are real and compounding. None of that should be dismissed.

But the infrastructure powering those benefits is generating costs that fall on communities who didn’t choose to bear them — through shared grids, shared water tables, and shared land. The mechanism isn’t malice. It’s a governance gap: the rules haven’t kept pace with the scale of the buildout.

“It is not enough to invoke ethics in the abstract; robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility are required.”

Strip the theological framing and what remains is a governance question: who decides how this technology gets built, and who gets a say when the costs land somewhere they didn’t plan for?

This isn’t an argument against AI. It’s an argument for honesty about what it costs — and for building the governance structures that make the costs fair.

What This Means for Us — Builders, Technologists, Citizens

This is the section where most tech newsletters pivot to “here are five frameworks for optimising your AI spend.” We’re taking a different direction, purely from a humanity point of view.

1. Efficiency isn’t just a P&L number. It’s a contribution. To the future of our communities.

   Every AI system that does more with less compute is not only cheaper to run — it draws less from a shared grid, consumes less water, and reduces the cumulative pressure on infrastructure that communities depend on. When you write a tighter prompt, choose a smaller model for a task that doesn’t need a large one, or architect a system to avoid redundant inference calls, you’re not just saving money. You’re returning something to the pool.

   The IEA projects that roughly half of power demand growth in the U.S. and Japan over the next five years will come from data centres. Every decision that bends that curve even slightly is meaningful.

2. We are the people who can hold the balanced conversation — and that’s a responsibility.

   The public debate about AI’s environmental and social costs tends to collapse into two camps: uncritical boosterism (”the benefits outweigh everything”) and reactive alarm (”shut it all down”). Neither camp is particularly useful. And the people most capable of holding a more honest, nuanced position are the ones who understand both sides — the technology and the community it operates inside.

   That’s people like us, reading this newsletter.

   Most people in our families, our neighbourhoods, our city councils don’t have the context to evaluate what they’re hearing. They’re getting fragments — a headline about dry wells, a talking point about Canadian AI sovereignty, a papal encyclical — without the connective tissue to form a real view. We can provide that connective tissue. In conversations at work, at the dinner table, in WhatsApp threads that aren’t this one. Not as advocates for a position, but as people who’ve actually thought about it. That capacity to translate complexity into honest conversation is genuinely scarce. We should use it.

3. Technologists are now ESG actors. Whether we’ve accepted that role or not.

   Environmental, Social, and Governance criteria used to feel like a concern for oil companies and mining executives — industries with visible, measurable environmental footprints. That era is over. The infrastructure decisions we make as builders, the vendors we choose, the compute platforms we deploy on, the energy sources those platforms run on — all of it is now part of an emerging accountability framework that regulators, enterprise customers, and institutional investors are starting to enforce.

   Ontario’s Bill 40 is an early signal. The EU’s AI Act and emerging sustainability disclosure requirements for tech companies are others. Enterprise buyers in healthcare, financial services, and insurance are already beginning to ask their AI vendors where compute runs, what the energy source is, and what the environmental disclosure looks like. Getting ahead of that question is product strategy — and for founders building in the GTA right now, it’s also an opportunity to differentiate on values that are becoming regulatory requirements.

✨ SaugaTech Epilogue

Every major technological shift has eventually required a reckoning. The industrial revolution brought extraordinary productivity — and polluted rivers, destroyed lungs, and reshaped communities without their consent. The reckoning came. Laws got written. Standards got set. The benefits became more broadly shared.

AI is moving faster. Its costs are less visible. Its benefits are less evenly distributed. The dry wells in Indiana and the rising electricity bills in Virginia are early signals — not of a technology gone wrong, but of a transition being managed without enough honesty about what it demands.

Pope Leo called for AI to be “disarmed.” What he meant, stripped of the theological register, is simpler: a technology this powerful shouldn’t be deployed at the expense of the most vulnerable, and the people bearing the costs should have a voice in how it gets built.

That’s not an unreasonable ask. And communities like ours — informed, engaged, sitting at the intersection of technology and real life in a real city — are exactly the ones to make it.

The question isn’t whether AI is worth building. It is. The question is whether we’re building it in a way we’d be proud of in twenty years.

That conversation starts here.

See you on June 20.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Share

The tech world woke up to quite a surprise on April 22nd, 2026.

A space company just valued a three-year-old coding tool at more than the GDP of most countries. More than Ford, General Motors, and Air Canada combined.

SpaceX announced it had secured an option to acquire Cursor — an AI-powered code editor — for $60 billion. Not $60 million. Billion. With a B.

We’ve been watching the AI coding tools space closely in our WhatsApp group over the past year. The conversations have been fascinating — developers sharing their experiences with Cursor, Claude, Copilot, debating which tool actually makes them faster, which one truly “gets it.” But this deal, if it materializes, changes the conversation entirely. This isn’t just about which tool is better anymore. This is about who owns the future of software development — and what that means for those of us building across the GTA.

Grab a coffee, let’s dig into what happened — and why it matters.

But before we get started….

🚀First Things First - Join the Community

SaugaTech turned one last week. 🎂

If you haven’t read the anniversary note yet — go read it. It’s personal, honest, and says everything about why this community exists and where we’re headed over the next 10 years. You can find it here.

And if you’re not already part of our WhatsApp group — that’s where the real conversations happen. Every week, builders from across the GTA are sharing what they’re working on, what’s breaking, what’s working, and what they’re reading. From Openclaw to Career Mentoring to this Cursor story? It’s been all over the group. Join us there.

We are already planning for our May meetup, will announce the date and topic soon. If you have a topic you’re passionate about or want to speak at a future session, please reach out to us on saugatechproduct@gmail.com.

Anyway, back to Cursor now.

What Is Cursor — And How Did It Get Here?

If you’re not a developer, Cursor might not be on your radar yet. But it should be.

Cursor is an AI-first code editor — think of it as a supercharged version of Visual Studio Code, but with AI baked into every part of the development workflow. It doesn’t just autocomplete your code. It writes entire functions, debugs your errors, refactors multi-file projects, and even reviews pull requests autonomously.

And it grew fast. Really fast.

Here’s what Cursor achieved in roughly two years:

• 1 million daily active users by early 2026

• $1 billion in annualized recurring revenue (ARR) — crossing that threshold in late 2025

• 67% of Fortune 500 companies using it as of April 2026

• 150 million lines of enterprise code generated per day

Cursor didn’t win by outspending competitors. They won by building a better product and letting developers spread the word. Classic product-led growth. And now, they have the deepest pockets in tech backing them.

Why SpaceX? Why Now?

This is where the story gets interesting.

SpaceX isn’t just a space company anymore. In February 2026, SpaceX merged with xAI — Elon Musk’s AI venture — in a $1.25 trillion all-stock deal. That merger created a vertically integrated AI powerhouse: SpaceX provides the compute infrastructure (including the Colossus AI supercomputer), xAI provides the models and research, and now Cursor provides the enterprise software layer.

The timing of the Cursor deal is deliberate. SpaceX confidentially filed for its IPO on April 1, 2026, targeting a $1.75 trillion valuation for a June listing. Adding Cursor to the portfolio before the IPO gives investors something they desperately want to see: enterprise software revenue.

Space launches are impressive. AI research is exciting. But $1 billion in ARR from Fortune 500 companies? That’s a story Wall Street understands.

And here’s the kicker: Musk isn’t just looking to buy Cursor for the revenue. He’s buying it because he believes AI-written code is foundational infrastructure for the future. If software is eating the world, and AI is writing the software, then owning the platform where that happens is a trillion-dollar bet.

The Deep Pockets Advantage

Let’s talk about what makes this deal different from a typical acquisition: xAI has functionally unlimited capital.

xAI has raised approximately $45 billion over nine rounds. Investors include Valor Equity Partners, Fidelity, Qatar Investment Authority, NVIDIA, and Cisco. But more importantly, xAI is now part of the Musk empire — which means it has access to SpaceX’s balance sheet, Tesla’s AI infrastructure, and X’s distribution.

Compare that to OpenAI, which is burning cash, facing internal talent exodus, and scrambling to monetize ChatGPT beyond consumer subscriptions. OpenAI is under immense pressure to IPO, but they don’t have a clear enterprise revenue story in the coding tools space. GitHub Copilot generates revenue, but it’s not theirs — it flows to Microsoft.

Meanwhile, Musk just bought $1 billion in ARR and 67% Fortune 500 penetration overnight.

That’s the advantage of deep pockets and vertical integration. While competitors are raising rounds and managing burn rates, xAI can simply acquire the market leader and integrate it into a broader AI ecosystem. This is a different game — and it’s one that OpenAI, for all its talent and brand recognition, is struggling to play.

The E Factor

There is one variable in this story that no spreadsheet can fully price in: Elon Musk himself.

Love him or loathe him — and the tech world is firmly split on this — Musk is a category-defining force. The moment his name attaches to something, the rules change. Media attention, investor interest, developer curiosity, regulatory scrutiny — all of it arrives at once and at scale. That’s a force multiplier that no competitor can simply buy.

Consider what the Musk effect has already done for xAI. A $20 billion Series E round — the largest in AI history — closed in January 2026. Investors including NVIDIA, Cisco, and the Qatar Investment Authority didn’t just bet on the technology. They bet on the man behind it. And now, with Cursor in the portfolio and a SpaceX IPO on the horizon, that bet is looking sharper by the day.

The AI Coding Tools Landscape — Who’s Winning Where?

The Cursor acquisition doesn’t just change xAI’s position. It potentially reshapes the entire AI coding tools market. And if you’re a builder, you need to understand where each player stands — because the tool you choose matters.

Here’s how the market is segmenting:

Cursor (potentially xAI/SpaceX in near future) — The Enterprise Play

• Target: Professional developers, Fortune 500 enterprises

• Positioning: AI-first IDE with multi-file editing, autonomous agents, enterprise deployments

• Strengths: Product-led growth, proven enterprise adoption, $1B ARR, now backed by Musk’s deep pockets and xAI’s compute infrastructure

• Weaknesses: Still needs to prove integration with xAI’s broader ecosystem; acquisition is very recent

Claude (Anthropic) — The Core Coder’s Choice

• Target: Technical power users, engineers who care about reasoning quality

• Positioning: Best-in-class reasoning, long context windows, trusted by developers for complex, nuanced tasks

• Strengths: Developer trust, technical depth, strong performance on hard coding problems

• Weaknesses: Smaller distribution than Microsoft or Google; less enterprise infrastructure

Gemini (Google) — The Non-Coder Ecosystem

• Target: Non-coders, low-code/no-code users, enterprise integrations

• Positioning: Integrated into Google Workspace, Android, Chrome — democratizing coding for non-technical users

• Strengths: Massive distribution, enterprise relationships, ecosystem lock-in

• Weaknesses: Not the tool of choice for serious developers yet; seen as a “good enough” option rather than best-in-class

GitHub Copilot (OpenAI/Microsoft) — The Default That’s Losing Ground

• Target: Developers already in the VS Code/GitHub ecosystem

• Positioning: Default choice for many devs, integrated into GitHub workflow

• Strengths: Distribution advantage (built into VS Code), Microsoft’s enterprise sales machine

• Weaknesses: Neither here nor there — not the best tool for core coders (Claude wins on reasoning), not the best for non-coders (Gemini wins on ecosystem), and now losing enterprise share to Cursor

The “Neither Here Nor There” Problem

Here’s the uncomfortable truth for OpenAI: they’re being outflanked in the coding tools market from every direction.

• Technical depth? Claude owns that.

• Distribution and ecosystem integration? Gemini (Google) owns that.

• Enterprise developer adoption? Cursor (now xAI) owns that.

GitHub Copilot is still widely used, but it’s not the best at anything. It’s the default. And defaults get disrupted when better alternatives emerge and the market matures.

OpenAI’s valuation — $157 billion as of late 2024 — is predicated on being the AI leader across all categories. But in coding tools specifically, they’re being squeezed from every side. And with IPO pressure mounting, they need to show sustainable enterprise revenue beyond ChatGPT subscriptions.

Meanwhile, xAI just bought their way into the enterprise coding market. Anthropic is carving out the technical power user segment. Google is locking in the non-coder ecosystem. OpenAI is stuck in the middle — and the middle is a dangerous place to be.

What This Means for GTA Builders

So what does a $60 billion acquisition of an AI code editor mean for those of us building in across the GTA?

A few things.

AI coding tools are no longer optional. If 67% of Fortune 500 companies are using AI code editors, fluency in these tools is table stakes. The developers and founders who know how to work with AI coding tools will outpace those who don’t — not by a little, but by a lot.

Deep pockets + strategic focus beats first-mover advantage. Cursor wasn’t the first AI code editor. GitHub Copilot was. But Cursor executed better, grew faster, and now has Elon Musk’s war chest behind them. We’ve said it before in this newsletter — first-mover advantage is overrated. Execution is everything.

Pick your lane — and own it. Claude picked core coders. Gemini picked non-coders. Cursor picked enterprises. Each has a clear wedge. OpenAI tried to be everything to everyone and is losing ground. If you’re building a product, the lesson is the same: pick your lane, go deep, and don’t let the fear of missing out pull you in every direction.

Vertical integration is back. Musk’s strategy — own the compute (Colossus), the models (xAI), the tools (Cursor), and the distribution (SpaceX, Tesla, X) — is working. It’s the competitive moat of the 2020s. For GTA builders, the lesson is simpler: think about what you own end-to-end, not just what you build.

And finally — choose your tool based on your use case. Core coder working on complex problems? Use Claude. Building for non-coders or integrating into Google Workspace? Use Gemini. Working in an enterprise or need multi-file AI editing? Use Cursor. Don’t default to GitHub Copilot just because it’s familiar. The market has moved.

SaugaTech Epilogue

We’re watching this space closely. The AI coding wars are just getting started, and the next 12 months will determine who wins.

Will xAI successfully integrate Cursor into their ecosystem? Will OpenAI find a wedge in the coding tools market before their IPO window closes? Will Anthropic’s technical depth translate into enterprise adoption at scale? Will Google’s ecosystem play pay off with non-coders?

The tools are moving fast, the ownership is concentrating, and the window to build deep expertise is narrowing — which is exactly why the conversations happening in this community matter.

If you’re using Cursor, Claude, Copilot, Antigravity or any other AI coding tool — we’d love to hear about your experience. What’s working? What’s not? What are you building with it? Bring it to the WhatsApp group. That’s what it’s there for.

Share this with a builder who needs to read it. And keep building. 🧭

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Subscribe now

Share

Our very first meetup was at the RiverGrove Community Centre on the 12th of April 2025 - a room full of people who had never met, brought together by nothing more than a Reddit post and a shared frustration.

And I’ve been sitting with that for a while, trying to figure out what to say. I kept coming back to the same place - the beginning. So that’s where I’ll start.

For those of you who have been a part of the community for some time now, some of what follows will be familiar. But bear with me - because I’ve come to believe that a community’s origin story isn’t just history. It’s a compass. It tells you who you are when you’ve forgotten, and where to return when you’ve drifted.

It was a Thursday evening in March 2025.

I was at my workplace in Mississauga, when a notification buzzed up on my phone - a reminder for a ProductTO meetup happening that evening in Downtown Toronto.

I wanted to go. I genuinely did.

But I looked at the time, thought about the commute, thought about getting back, and I did what most of us in Mississauga do on a weekday evening.

I put my phone down.

And then I sat with that quiet frustration for a while.

The Frustration That Started Everything

Here’s the thing about Mississauga that not enough people say out loud - this city is full of incredibly talented, ambitious, curious people. Founders, operators, engineers, product thinkers, designers. People who have built things, broken things, learned things. People who want to connect, want to learn, want to grow.

But Mississauga has always had a geography problem. The meetups are in Toronto. The events are in Toronto. The “scene” is in Toronto. And for those of us with families, with jobs that don’t end at 5pm, with a commute that turns a 7pm event into a 10pm return - Toronto might as well be a different country on a Tuesday night.

And it’s not just Mississauga. The same story plays out in Brampton, in Oakville, in Burlington, in Hamilton. Talented, ambitious people - spread across the most diverse stretch of urban Canada - with no local home for their curiosity and no easy way to find each other.

I moved to Canada in 2023, coming from India, where I had spent years in the trenches of early-stage fintech startups - a small business neobank and a retail lending company. I knew what it felt like to be on a 0-to-1 journey. I knew the specific kind of loneliness that comes with building something from nothing. And when I joined a large bank here in Canada, I found stability - but I also found myself missing something I couldn’t quite name.

I missed the builders. I missed the people who loved the adrenaline rush of the daily hustle and figuring things out in real time. I missed the conversations that happen at 9pm when the day’s work is done and someone says “okay but what if we tried this?”

I was new to Mississauga. I didn’t have a network here. And I had no idea how to build one.

The Reddit Post

So on that Thursday evening, instead of going to a meetup I couldn’t attend, I did something that felt equal parts obvious and terrifying.

I posted on Reddit.

I asked a simple question - something along the lines of: “Are there any tech folks in Mississauga who’d be interested in meeting up locally?”

And then I closed the app and tried not to think about it.

I want to be honest with you - I was anxious. I am a product person. I have spent my career thinking about users, building roadmaps, working in startups. I had never built a community before. I didn’t know if anyone would respond. I didn’t know if the idea was even good. There is a particular kind of vulnerability in putting something out into the world and not knowing if anyone will care.

People cared.

What This Region Was Waiting For

The responses came in. Not in the hundreds - but enough. Enough to know that the frustration I felt wasn’t mine alone. Enough to know that there were other people - in Mississauga, in Brampton, across the GTA’s western suburbs - sitting on Thursday evenings, getting meetup reminders they couldn’t act on, wishing there was something closer to home.

That’s when I understood something important: this region didn’t have a talent problem. It didn’t have an ambition problem. It had a connective tissue problem. The people were always here. The energy was always here. It just needed somewhere to go.

SaugaTech was born in Mississauga - that’s where the first post went up, where the first meetup happened, where the community found its feet. But it was never only for Mississauga. It was always for everyone who felt the same pull - the builder in Brampton, the founder in Oakville, the engineer in Burlington, the operator in Hamilton who has been quietly doing remarkable things with no one to share it with.

SaugaTech was never meant to be a grand vision. It started as a Reddit thread. And then, slowly, it became something I genuinely didn’t expect.

The Moment We Knew This Was Real

I’ve been asked what the moment was when I knew SaugaTech had become something genuine.

It wasn’t the subscriber count. It wasn’t when we hit 250 members. It wasn’t even when the Compass newsletter started getting traction.

It was at one of our meetups in August 2025.

A mid-career executive - someone in his 40s, clearly accomplished, clearly experienced - stood up and said something that I think about often. He said he was feeling lost in this wave of AI. That he didn’t know where to start. That he was struggling to make sense of it all.

In a room full of tech people, that takes courage.

And what happened next is what made me proud. The room didn’t judge him. The room met him. People nodded. People shared their own confusion. People offered their own experiences - not as experts talking down, but as fellow travellers comparing notes.

That’s when I understood what we had actually built.

Not a tech meetup. Not a newsletter. A safe space.

A place where a 40-year-old executive can say “I don’t know” and be met with warmth instead of silence. A place where vulnerability is not a weakness but a starting point.

That is not something we could have planned for. That is something the community created for itself. We just held the door open.

The People Who Made This Real

I want to take a moment - and I mean this sincerely - to name the people without whom SaugaTech would have remained a Reddit post.

Ibrahim has been SaugaTech’s foundation stone since Day 1. When I had the idea but had no idea how to execute it - no venue, no logistics, no real understanding of how Mississauga works as a city - Ibrahim stepped in. As someone who grew up and studied here, he knew the lay of the land in a way I simply didn’t. He singlehandedly set up the logistics for our meetups when we were, genuinely, a nobody. No track record, no audience, no guarantees. He showed up anyway.

Fernanda at TouchDown Coworking, Oakville gave us a home when we were still figuring out who we were. Lending your space to a community that’s just getting started is an act of generosity that we will always be grateful for.

As we grew, Salman and Bimal brought something we badly needed - structure. In the last six to eight months, they have been the driving force behind turning SaugaTech from a passionate but loosely organised group into something that can actually scale. Growth without structure is just noise. They helped us turn the noise into signal.

Maheen brought a creative sharpness that changed how we present ourselves to the world. Her marketing and branding instincts have helped us sharpen our narrative - to say clearly who we are, who we’re for, and why it matters. That clarity doesn’t happen by accident.

Tej and the team at Invest Mississauga came in and gave us something that early communities desperately need - institutional support. Venue, logistics, marketing. The kind of backing that signals to the world that what you’re building is worth paying attention to.

And to our early supporters and facilitators, Aries Youssefian, Nilay Lad and many more - thank you for standing up in front of our community and sharing your knowledge, your experience, and your time. The sessions you facilitated gave our members something that no newsletter or LinkedIn post can replicate: the lived experience of someone who has actually done the thing.

To every single person who showed up to a meetup, joined the WhatsApp group, shared a post, suggested a topic, or simply told a friend - thank you. You are not just members of this community. You are this community.

More Than We Ever Imagined

When I posted on Reddit that Thursday evening, I was hoping maybe ten or fifteen people would show up to a meetup someday.

Today, we are 450+ members strong on Substack, almost 200 on Whatsapp community. We have a newsletter - the Compass - that goes out regularly and covers everything from B2B SaaS strategy to AI agents to the stories of Canada’s quiet builders. We have had multiple meetups, hands-on workshops, and sessions that have genuinely changed how people think about their work and their careers.

I’ll be honest - there were moments along the way where we weren’t sure we’d get here. Moments of self-doubt. Moments of wondering if the next post would land, if anyone would show up, if the energy would sustain. I am not a community builder by training, none of us are in the SaugaTech volunteers group. We are figuring this out as we move forward, learning from every meetup, every conversation, every piece of feedback.

But this community kept showing up. And that made all the difference.

This Is a 10-Year Journey - And It Belongs to All of Us

SaugaTech is not my community. It never was.

I just posted on Reddit. I just held the first door open.

But every person who walked through that door built this alongside us. And what we want to say now, on this first anniversary, is that what we are trying to build is a 10-year journey - a genuine, lasting tech ecosystem rooted in Mississauga, open to every corner of the GTA, and ambitious enough to matter on a national stage.

The best version of SaugaTech hasn’t been imagined yet. And we strongly believe it will be imagined by this room - by the 450+ of you already here, and the thousands more yet to find us across Brampton, Oakville, Burlington, Hamilton, and beyond. Imagine the conversations that haven’t happened, the collaborations that haven’t started, the companies that haven’t been built - all because the right people hadn’t met each other yet.

That’s what this can be. And the most beautiful thing about a community is that it grows in the direction its people take it. We don’t have a fixed destination - we have a shared direction. Every person who leans in, even a little, shapes where we end up. And in doing so, grows themselves - new connections, new perspectives, new confidence. When you invest in the room, the room invests back in you.

I’d love nothing more than to look back in 10 years and see a SaugaTech that none of us could have predicted today. Because that’s exactly what happened in year one.

The North Star

Someday, not so far in future - and I genuinely believe this - there will be a large-scale tech summit, right here in Mississauga. Not a satellite event of something happening in Toronto. Something that belongs to this region. Where the speakers are people from our community - from Mississauga, Brampton, Oakville, Burlington, Hamilton and everywhere in between. Where the founders on stage first met each other at one of our meetups.

We don’t know exactly when. We don’t know exactly how. But We know it’s possible.

We are one year in. Nine to go.

Thank You - For Everything That’s Coming

To everyone who has been part of this first year - Thank You. Genuinely, deeply, thank you.

This region always had the people. It always had the spirit. It just needed a place to come together.

Now that we have it, hopefully we will be able to nurture this, together.

There is a couplet by Urdu poet, Majrooh Sultanpuri that I keep coming back to, one that feels like it was written for exactly this kind of journey:

“Main akela hi chala tha janib-e-manzil magar, Log saath aate gaye aur karwaan banta gaya.”

That literally translates to: I set out alone towards my destination - but people kept joining me, and a caravan was formed.

That is the only way I know how to describe what happened here.

We still have a long road ahead. But I can’t think of a better group of people to walk it with. 🧭

Let’s keep building, Let’s keep learning, Together.

Cheers

Robin (Team SaugaTech)

CONNECT | COLLABORATE | INNOVATE

P.S. If this resonated with you - share it with someone who belongs in this room. And if you aren’t on the whatsapp group, Click here to Join, that’s where the daily action is.

Subscribe now

Share

Hi SaugaTech Community,

This week, Anthropic did something the AI industry has never done before. They announced their most powerful model ever built — Claude Mythos — and in the same breath said the public cannot have it. Not yet. Possibly not for a long time.

The reason? The model is too capable. Mythos can autonomously discover zero-day vulnerabilities across every major operating system and browser, chain exploits together, and produce proof-of-concept attacks with a depth that outpaces human security teams. Anthropic decided that releasing it publicly before defenders had a chance to use it would hand attackers an advantage that couldn’t be taken back. So instead, they launched Project Glasswing — giving access exclusively to a small group of enterprise partners including Amazon, Apple, Microsoft, Google, JPMorgan Chase, and Nvidia — to use Mythos specifically for defensive cybersecurity work. 40 organizations total. No one else.

It is a striking moment. The most powerful AI model ever documented — 93.9% on SWE-bench Verified, finding a bug in OpenBSD that had gone undetected for 27 years — is being handed exclusively to large enterprises.

But here is the tension at the heart of the enterprise AI story right now. The most powerful tools are flowing to organizations at the top of the pyramid — and yet most of those organizations are still struggling to get a basic AI pilot from sandbox to production. The capability gap between what AI can do and what enterprises are actually doing with it has never been wider.

The boardroom is getting Mythos. The product team down the hall is still debugging why the model that worked perfectly in staging behaves differently in the real workflow. The data isn’t clean enough. The vendor’s ROI slide deck didn’t mention the twelve weeks it takes to integrate with the legacy system.

For those of us building in the GTA — engineers, PMs, architects working inside these organizations or selling to them — understanding that gap is the difference between a product that actually gets used and one that quietly dies in a proof of concept.

The good news is that some organizations are genuinely cracking it. And the playbook they’re using has less to do with access to frontier models and more to do with how they think about their own people.

Grab a coffee. Let’s get into it.

🚀 First Things First

Details for our next meetup are dropping soon in the SaugaTech WhatsApp Group. If you’re not in there yet, join now — that’s where the real conversation is already happening between meetups. We also recently reorganized the group as a Community with specific forums for Builders, Job Openings, Vibe Coding, Systems Design, Career Evolution & Coaching etc. So the conversations are only going to get richer from here.

What’s Actually Working at Enterprise AI rollouts

The organizations seeing real returns share a few things in common. None of them start with the headline use case.

Democratize access, then let the use cases surface. The most instructive enterprise AI story right now isn’t about a technology breakthrough — it’s about a large global bank that made a counterintuitive decision. Instead of restricting AI to a select group working on approved initiatives, they gave tools to over 140,000 employees across dozens of countries and told them to figure out what was useful. The result was a 70% tool adoption rate, with employees interacting with AI over 21 million times — and internal reporting describing work that once took hours turning into tasks done in minutes. The use cases didn’t come from a strategy team. They came from the people doing the work, discovering what helped them, and sharing it across teams. Give people tools and permission to experiment. The insights follow.

Go where the pain is obvious and the measurement is simple. In healthcare, two categories are delivering clear, measurable ROI right now: ambient clinical documentation — which reduces physician burnout by eliminating post-appointment write-ups — and coding and billing automation, which recovers revenue lost to manual errors and claim denials. Neither is glamorous. Both solve a specific, chronic pain that everyone in the room immediately recognizes. The pattern holds across industries. Cost benefits from AI are most commonly showing up in software engineering, manufacturing, and IT — functions where the work is repetitive, the output is measurable, and the baseline is easy to compare against. Transformation is hard to measure. Time saved on a specific task every Tuesday is not.

Solve small, then cross-pollinate. Companies deploying the majority of their AI initiatives to production — versus laggards stuck in pilot mode — share one characteristic: tight alignment between where AI is deployed and where it actually delivers impact. They don’t try to transform everything at once. They find one workflow that AI makes meaningfully faster or cheaper, prove it works, document it, and ask which other teams have the same problem. The compounding comes not from one big initiative but from dozens of small ones spreading across the organization once the first one proves out.

Put AI in the hands of your tech teams first. This is the one that works across every industry with a technology footprint — which is essentially every industry. 84% of developers now use or plan to use AI tools, with 51% using them daily, saving an average of 3.6 hours per week. 41% of all code written today is AI-generated. Developers are shipping faster, spending less time on boilerplate, and freeing up cognitive bandwidth for harder problems. The honest nuance worth keeping: developers say they’re working faster, but companies are not always seeing measurable improvement in overall delivery velocity. The gains are real at the individual level. Capturing them at the organizational level requires pairing the tools with better review practices and clearer measurement. The teams getting this right aren’t just giving developers access to AI coding tools — they’re building the habits and governance around those tools that turn individual speed into team throughput.

Where Things Get Stuck

Now for the part that doesn’t make the press releases.

42% of companies have abandoned the majority of their AI initiatives before they ever reach production — a dramatic rise from just 17% in 2024. More than half of CEOs have not realized either revenue or cost benefits from AI. Among the projects that did reach production, only 38% delivered their projected ROI within the first two years.

Five things keep showing up in the wreckage.

The data problem nobody wants to admit. Every organization has data. Almost none of them have the specific, clean, well-labelled data that AI systems actually need. Data quality remediation runs $12.9 million annually per organization. Each vendor integration adds five to twelve weeks. None of those numbers appear in the vendor’s ROI slide deck. Organizations discover this six months into a project, after a successful sandbox demo, when they try to move to production and the foundations aren’t there.

Bolting AI onto broken processes. Most failures occur because companies try to attach AI onto existing, broken processes. A customer service bot that works technically but contradicts the company’s culture of high-touch service isn’t an AI failure — it’s a strategy failure. The technology did what it was asked. Nobody asked the right question before building it.

Keeping AI locked up. A small team gets access to AI tools. They build something impressive in a sandbox. The demo goes well. Then nothing happens, because the people whose daily work would actually benefit never got access in the first place. While only 40% of companies say they purchased an official AI subscription, workers from over 90% of companies surveyed reported regular use of personal AI tools for work tasks. The unofficial adoption is already happening — it’s just invisible, uncoordinated, and not compounding into anything. The organizations treating this shadow usage as a signal rather than a risk are the ones finding their best use cases.

The probabilistic problem in regulated industries. Traditional enterprise software is deterministic: the same input produces the same output every time. Modern AI is probabilistic: outputs vary based on statistical inference. Because regulations require organizations to explain adverse decisions — credit decisions, insurance outcomes — if they can’t explain AI outputs, they face significant liability. In credit underwriting, mortgage approvals, and medical diagnoses, “probably right” isn’t good enough. The result is that some of the highest-value use cases in the GTA’s biggest industries stay stuck at the pilot stage — not because the technology isn’t capable, but because the governance infrastructure isn’t yet mature enough. What actually works in these environments is a hybrid model — AI proposes, rules constrain, humans approve, with a full audit trail at every step. Organizations that build for that from day one will unlock what everyone else is still piloting.

The human problem nobody puts in the slide deck. Technology is rarely the real blocker. The AI skills gap is cited as the biggest barrier to enterprise AI integration — and the number one organizational response has been education, not workflow redesign. But the deeper issue sits in middle management. These are the people who decide which tools their teams use, which processes get changed, and which pilots actually move forward. Many of them are experienced, capable leaders who have built their careers on deep expertise — and who now find themselves unsure where AI fits, or quietly skeptical of a technology they haven’t had time to properly explore. When that layer of the organization isn’t brought along — when AI is handed down from the top without giving mid-managers the space to understand it on their own terms — rollouts stall. Not because of bad data or broken processes. Because the person approving the workflow change doesn’t believe in it yet. The organizations getting this right are treating middle management as a deployment problem, not an afterthought.

The Opportunity Nobody Is Pitching For

Here’s what gets lost in the coverage of Mythos and frontier models and enterprise transformation.

Every organization has people spending their Tuesday afternoon doing exactly what they did last Tuesday. Running the same MIS reports. Doing the same vlookups. Emailing the same numbers to the same distribution list. Nobody calls it a problem because it’s always been done this way.

With a few hours and the free AI tools available today, that entire workflow becomes a single button click. A simple HTML tool that pulls the data, formats the output, and produces the report — built by someone who understands what the report needs to do.

This scales further than people realize. Alberta’s GovLab — a public sector AI lab built through a partnership between the Government of Alberta, AltaML, and Mitacs — has been automating document classification across government departments and freeing staff for higher-value work. The lab has helped public sector organizations uncover over $32 million in potential value through applied AI — not through moonshot transformation projects, but through targeted automation of the repetitive, high-volume work that exists in every department.

And the GTA’s working professionals — who know exactly which Friday afternoon process nobody has ever bothered to fix — are better positioned to find that problem than any external consultant.

That’s your wedge. Not an enterprise transformation. Not an AI strategy. Just: what if this thing that takes four hours took four minutes instead?

Build the tool. Show the business head what Friday afternoon looks like when it’s free. That’s efficiency that every stakeholder can see and feel — no ROI model required, no vendor negotiation, no change management programme.

This Is Also How You Build Your Career

The Mythos story is about the frontier. What’s happening in the middle of most organizations is a different story — and it’s where the real career opportunity lives.

The people who are going to stand out over the next few years aren’t necessarily the ones with access to the most powerful models. They’re the ones who combined domain knowledge with the curiosity to experiment with the tools available today — and then showed others what was possible.

Every organization right now has a gap between the people who understand the business problems and the people who understand the AI tools. The person who sits at that intersection and starts solving real things — even small things, even just for their own team — becomes the person everyone wants to talk to. The person who gets pulled into the next initiative. The person who shapes how the organization thinks about AI rather than just reacting to decisions made above them.

This is how influence gets built during periods of technological change. Not by waiting for a mandate from the top. By building something useful for the people around you and letting the results speak.

The MIS report that used to take four hours and now takes four minutes gets noticed. The compliance summary that used to require two analysts and now takes one person twenty minutes gets noticed. The onboarding document that used to take a week and now gets generated in an afternoon gets noticed.

And this connects directly to the middle management problem. The mid-manager who rolls up their sleeves, experiments with the tools on a real problem, and shows their team what’s possible — that person doesn’t just move the rollout forward. They become the internal champion that every large-scale AI implementation actually needs but rarely has. You don’t wait to be convinced. You convince yourself first, with something real.

Start with your own team. Solve one problem. Show people what’s possible. Then let them pull you toward the next one.

What This Means for GTA Builders — Start Small, Start Now

Whether you’re building AI products for enterprise customers or navigating AI adoption inside a large organization, the same principles separate what works from what doesn’t.

Give people tools before you ask for use cases. The bank that gave 140,000 employees access and said “show us what you find useful” discovered more valuable use cases faster than any internal innovation programme would have. If you’re leading AI adoption inside an organization, democratize access early. If you’re building AI products, design for the person doing the work, not the executive signing the cheque.

Start with the boring problems. Nobody writes case studies about automating the monthly MIS report or classifying government documents. But it compounds quietly and reliably in a way that AI chatbots rarely do. The back-office workflow that nobody notices until it breaks is often where the most durable ROI lives.

Cross-pollinate ruthlessly. When someone in your organization solves a problem with AI — even a small one — make it visible. Document what they built, share how they built it, and ask which other teams have the same problem. That’s how individual wins become organizational capability.

Understand the deterministic boundary. If you’re building AI products for regulated industries — credit, insurance, healthcare — the question your customer’s compliance team will eventually ask is not “does it work?” but “can you explain every decision, reproduce it exactly, and audit it completely?” Build for that question from day one. The organizations that crack the hybrid model — AI inside deterministic guardrails — will unlock the use cases that everyone else is still stuck piloting. That’s a real product opportunity for GTA builders right now.

Bring the middle layer along. Whether you’re an internal champion or an external vendor, your AI initiative is only as strong as the mid-managers who have to change how their teams work. The most underinvested part of every enterprise AI rollout is the time spent helping that layer understand, experiment, and develop conviction. The slide deck won’t do it. Sitting down with someone and showing them what their specific job looks like with AI — that’s what moves it.

Slow down on the data before speeding up on the model. The organizations achieving durable AI advantage in 2026 aren’t the ones who moved fastest — they’re the ones who slowed down long enough to build the data and governance foundations that everything else depends on. If you’re building for enterprise customers, the companies most ready to buy are the ones who’ve already done this foundational work. Find them first.

✨ SaugaTech Epilogue

Anthropic built the most powerful AI model ever documented and decided not to release it. That restraint, whatever you think of the reasoning behind it, tells you something about where we are. The capability is accelerating faster than most organizations can absorb it.

That’s not cause for alarm. It’s cause for clarity.

The organizations winning right now didn’t get there by waiting for the most powerful model. They got there by giving their people tools they could actually use, letting them find problems worth solving, and spreading what worked. The returns came from thousands of small efficiency gains — discovered by the people closest to the work, not the transformation team — and compounded over time.

For those of us in the western GTA — working in financial services, manufacturing, healthcare, logistics — the same opportunity exists at every scale. The probabilistic vs deterministic question will shape regulated industries for years. The mid-manager who experiments first will shape how their organization uses AI. And the person who solves the Friday afternoon report problem this weekend is already ahead of most enterprise AI strategies.

You don’t need Mythos to start. You need a problem you understand, a few hours, and the willingness to show someone what’s possible.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Subscribe now

Share

Before we talk about what happened this weekend, let’s go back to May 2025 for a moment.

SaugaTech Meetup #2 at a small room at Mississauga Valley Community Centre. About 15 people in the room. A projector that couldn’t quite decide which angle it wanted to sit at. Aries Youssefian at the front, talking to a humble group of tech enthusiasts through lean tech stacks for startups. Here is how it looked like

Our founding members, Robin and Ibrahim, who were in the room that day, still share a laugh about that projector. And feel a deep sense of gratitude about how far we’ve come in less than twelve months.

Fast forward to this Saturday, 28th March at IDEA Mississauga, Square One. Around 60 people packed into the room — and then kept coming. People were pulling up floor space in the hallway just to be part of it. Aries was back at the front, this time leading a hands-on Vibe Coding workshop with Google Antigravity and Gemini CLI. And at the end of the session, 7 people walked up on stage to demo what they had built. From scratch. In under two hours. Here is what it looked like this time

That’s the SaugaTech story in two scenes.

What Happened on Saturday

The session was called Think It, Build It: Vibe Coding with AI Agents and it did exactly what it said. Aries opened with a walkthrough of how the coding landscape has shifted — from simple autocompletes in 2021 to agentic workflows today, where you’re no longer writing every line but guiding intent and letting agents plan, build, and test alongside you.

Then everyone got to work.

Using free tools — Gemini CLI and Google Antigravity — participants started building real applications on the spot. The room had developers, product managers, architects, marketers, and people with no coding background at all. That mix turned out to be one of the best parts of the day.

A UofT grad who describes herself as mostly a non-coder built a working website in about an hour and a half. It pulled live news headlines, summarised them, and then — after a few headlines — triggered a jumping game inspired by Google’s offline dinosaur. She then iterated and built two more versions in the same afternoon. Her take: the tools felt genuinely magical.

A Senior IT Solution Architect showcased a decentralised CRM he vibecoded. His partner, who had no prior coding experience, built a personalised planner app tracking tasks, reminders, meals, and notes — and had it done before the session wrapped. His reaction: “I was surprised at how quickly she went from not knowing the tools to actually building a working application.”

A Senior AI Product Manager built Mississauga View — a situational awareness dashboard that visualises real-time municipal telemetry across critical infrastructure and transit layers, giving you a live pulse on what’s happening across the city. From idea to working application in two hours.

A Front End Developer built LinkedYet, a real-time network graph where you can add connections and watch the social graph grow live — inspired by the energy of people leaving an event with new connections, and making that experience something you can actually see and interact with.

These were not just UI demos. These were working applications, built by real people, in an afternoon, using tools that were free and publicly available. You can actually try these applications using below links

• Mississauga View

• LinkedYet

Mississauga Tech Ecosystem - Becoming More Than the Sum of Its Parts

Looking out at that crowd on Saturday, Aries reflected on something worth sitting with: “Just last April, I led a session on lean tech stacks for startups to a humble group of about a dozen people in a small room at the Mississauga Valley Community Centre. To see this community grow from a handful of tech enthusiasts to a packed house with standing-room-only energy is incredible.”

That growth isn’t just about numbers. It’s about what this community is becoming.

Mississauga has always had the talent. The engineers, the architects, the product thinkers, the founders — they’ve always been here. What’s been missing is a shared space where that talent finds each other, builds together, and compounds. Saturday felt like proof that the space now exists.

This Is What Connect, Collaborate, Innovate Actually Looks Like

Those three words sit at the heart of everything we do. Saturday was a good reminder of what they mean in practice.

CONNECT: A global marketing leader, a senior solution architect, a UofT grad, an AI product manager, a front-end developer, and dozens more — none of them in the same room by accident. They came because there’s a community worth showing up for in Mississauga. Conversations moved across org structures, go-to-market strategy, product management, and careers — not as a side effect of the workshop, but as the natural result of the right people finally being in the same room.

COLLABORATE: Participants who didnt know each other before the event helping each other debug prompts. Seven people brave enough to walk on stage and share what they made, knowing the room would cheer them on regardless of the result. And after the session wrapped, conversations seamlessly shifting to the SaugaTech WhatsApp Group, with people organically offering to collaborate and explore how to take the ideas built during the workshop further.

INNOVATE: A self-described non-coder building three versions of a website in an afternoon. A city data idea turned into a live dashboard. Networking itself turned into something you can see and interact with. People who came in saying “I don’t really code” leaving with something they built.

That’s the virtuous flywheel in action. And it’s getting stronger with every meetup.

Epilogue

Thanks Aries once again for facilitating the workshop and making all the workshop materials open-source. Starter code, prompts, and project resources are all available at github.com/ariesyous/vibecodingworkshop. If you missed Saturday or want to go deeper, everything you need is there.

And if you could not make it to the meetup this time — join the conversation in the SaugaTech WhatsApp Group. That’s where the next idea is probably already forming. Who knows maybe you might run into someone out there building the next Shopify, right here in Mississauga.

And importantly, spread the word. Tell your colleagues/friends/peers, share this with them, invite them to the WhatsApp group — this is your community, help us grow and shape it.

See you at the next one.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

A few more glimpses from the session

Share

Subscribe now

P.S. If you have ideas on what topics to you would want us to cover in future meetups or would like to share your expertise as a facilitator, please drop your thoughts in the SaugaTech WhatsApp Group or email us at saugatechproduct@gmail.com.

Hi SaugaTech Community,

If you stepped outside in Mississauga this morning, you know exactly what kind of year we’re having. Last week it was 15°C and we were all convinced spring had arrived. Today it’s -8°C with a feels-like of -20°C. The weather genuinely cannot make up its mind.

Feels familiar, honestly. Neither can the tech industry.

But while everyone argues about whether we’re in an AI bubble or the beginning of something transformative, something quieter is happening across Canada. Companies are building things that work. Solving real problems. Making real revenue. No hype cycles, no moonshot obsession. Just good products that people need and are willing to pay for.

Google just announced its 2026 Accelerator Canada cohort: 14 startups that collectively raised $580 million and created 1,400 jobs since 2020. The funding numbers are impressive, sure. But what caught our attention was what these companies actually do and who they’re building for.

Grab a coffee. Let’s get into it.

🚀 First Things First: March 28 Meetup

Our next meetup is March 28 at IDEA Mississauga, Square One and we are almost out of spots. The session is called “Think It, Build It: Vibe Coding with AI Agents” and Aries will walk us through how he actually codes with AI tools before we build something together as a group. Really looking forward to getting the community in the same room for this one.If you haven’t registered yet, do it soon.

And join the SaugaTech WhatsApp Group to keep up with updates and get the conversation going before the session. Some of the best discussions from our meetups have started in that group days before anyone showed up.

Getting back to the Google cohort now.

What Canadian Startups Are Actually Building

Here’s a look at some of what Google’s 2026 cohort is working on.

CropMind out of Calgary helps farmers predict crop yields months before harvest. Planting decisions, irrigation, resource planning — farmers have always made these calls based on experience and instinct. CropMind gives them data to work with instead. Less waste, better planning, more stable supply chains.

EyeCareX in Vancouver is automating comprehensive eye exams. Canada has a real optometrist shortage, particularly in rural areas. People wait six months for appointments and by the time they’re seen, early signs of glaucoma or diabetes-related eye disease have been missed. EyeCareX speeds up screening so the doctors who are available can spend their time where it actually counts.

Innovate-Ops from Kitchener monitors industrial equipment and flags failures before they happen. Manufacturers lose thousands of dollars an hour when a production line goes down unexpectedly. These folks are selling the ability to keep things running, which is a very different conversation than selling software.

LandLogic in Toronto helps construction companies track land acquisitions and navigate regulatory compliance across jurisdictions. Nobody gets excited about permit tracking, but developers lose millions to delays caused by exactly that. More efficient development means more housing. That matters here.

Banting AI, also Toronto, gives physicians real-time clinical decision support. Doctors are swimming in information: latest research, drug interactions, treatment protocols. Banting helps surface what’s relevant in the moment. Better information, better decisions.

Every single one of these companies started by picking a specific problem and going deep on it. The technology came later. The customer never really cared about the technology anyway — they just wanted the problem gone.

What’s Happening Right Here in Mississauga

While Google celebrates its cohort, something just as meaningful is happening at IDEA Square One — which, if you were at our last SaugaTech meetup, you know well.

Xatoms is working on the global water crisis. Two billion people don’t have reliable access to clean water. They’ve developed photocatalysts using AI and quantum chemistry that can eliminate bacteria, viruses, pesticides, and heavy metals from water at a lower cost than what’s out there today. They’re already running pilots in Texas, Kenya, and South Africa.

DIA Labs Canada is building a continuous asthma monitoring device using microneedle technology. The problem with managing asthma today is that monitoring is reactive: you find out something is wrong when breathing becomes difficult. DIA Labs lets patients track inflammation in real time so they can get ahead of it rather than ending up in an emergency room.

Haami Lingerie is solving something the lingerie industry has largely ignored: plus-size women need bras that actually function comfortably. They’ve developed patentable technology for wireless, strapless, cupless bras designed specifically for this. Real pain point, real solution, real customers waiting for it.

Tea Time Therapy offers mental health services built specifically for the South Asian community. Culturally competent mental health care is genuinely hard to find. Thisanya built a space where South Asian clients feel actually understood, not just accommodated. No platform play. No AI angle. Just knowing your community and showing up for them properly.

Showstoppers Garage does car detailing, ceramic coatings, and tints. The innovation here is craft and customer experience. Saudia built a sustainable local business by being really good at something people care about. That counts.

All of these companies came through IDEA Mississauga’s Step-Up and Starter Company Plus programs. That support matters more than people realize. The difference between a good idea that goes nowhere and a real business serving real customers is often just access to the right mentorship, funding, and connections at the right time.

The Pattern Worth Paying Attention To

Look across everything above and a clear picture emerges.

These founders started by finding a problem, not by picking a technology. They spent time with the people living that problem. They understood it well enough that the solution became fairly obvious. Then they built the simplest version that could actually help.

Sometimes that involved sophisticated AI. Sometimes it was a service business with a great customer experience. The technology was always secondary. A farmer buying CropMind doesn’t think about the algorithm. An asthma patient using DIA Labs doesn’t think about microneedles. A woman buying Haami doesn’t think about the patent. They think about whether the problem is better now. That’s the only thing that matters.

AI can be an Enabler — Domain Expertise Is the Unlock

AI is genuinely powerful right now. Things that weren’t buildable three years ago are buildable today. That part is real.

But the founders in this piece aren’t winning because they know how to use AI. They’re winning because they know their problems inside out.

CropMind’s team has spent real time on farms. They’ve talked to dozens of growers. They know which decisions keep farmers up at night and which data points actually predict yields. The AI amplifies what they already know. Without that knowledge, they’d just be building another ag-tech product that farmers look at once and forget.

EyeCareX understands the clinical world from the inside: patient flow, regulatory requirements, the specific workflow bottlenecks that create backlogs. Their automation works because they know exactly where to apply it.

Innovate-Ops has felt the cost of unplanned downtime firsthand. That experience shapes every design decision: what to monitor, when to alert, how to communicate urgency to an operator in the middle of a shift.

This is the part that actually can’t be shortcut. Domain knowledge doesn’t come from reading reports about an industry. It comes from being in it. Sitting in the meetings. Living with the workarounds. Noticing the thing that everyone else stopped questioning because they assumed that’s just how it is.

Here’s the thing though: if you’ve spent five years in healthcare, finance, manufacturing, logistics, education, real estate — you already have that. You know where the friction is. You know the questions nobody bothers asking anymore. You know the processes that are held together with spreadsheets and hope.

That knowledge is genuinely hard to replicate. AI can make you move faster and build more efficiently, but it can’t tell you which problem is worth solving or why the existing solutions keep missing the mark. That comes from you.

What This Means for the Aspiring Builders in This Community — Start Small, Start Now

Here’s the honest reality of our community. Most of us are not 25 with two years of runway and no dependents. We’re working professionals with mortgages and families and responsibilities that don’t pause for a good idea. The stability we’ve built is real, and risking it feels irresponsible.

So the idea sits in the back of your head. You think about it on the commute. You mention it to a friend once. And then life keeps moving.

But the companies in this edition didn’t all start with a dramatic leap. They started by picking a specific problem and doing something small about it.

Start a side project. Keep your job. Find ten people who deal with the problem you’re thinking about and have honest conversations with them. Build the smallest version that could actually be useful. If you can charge even a little for it, do that early, because someone paying for something tells you more than a hundred people saying they like the idea.

ApplyBoard started by helping students fill out university applications. Clio started with helping individual lawyers track billable hours. None of these are revolutionary concepts. They’re just well-executed solutions to frustrations that a lot of people share, built by people who understood those frustrations well.

A few things that are worth keeping in mind:

• Your industry experience is the advantage. The years you’ve spent in a particular field aren’t holding you back. They’re the reason you can see problems that someone coming from outside would completely miss. Talk to enough people in that world that you understand their day better than they can articulate it themselves.

• Someone paying you is the real signal. A pitch deck, a waitlist, positive feedback from friends — none of that tells you much. Someone handing over money, even a small amount, tells you the problem is real enough to solve.

• Start with one thing. Don’t try to build the whole platform. Solve one specific problem for one specific group of people and do it well enough that they tell others. Everything else can come later.

• Use whatever technology actually helps. If AI makes the solution better, use it. If a spreadsheet is enough, use a spreadsheet. Customers don’t care how it works behind the scenes. They care that their problem is solved.

• The problems are here. Manufacturing in the 905. Financial services downtown. Healthcare across the province. Agriculture in the areas surrounding the GTA. You don’t need to move somewhere else to find a problem worth solving. You’re probably already sitting next to one.

If what you build is genuinely useful to real people, things will happen. Maybe slowly, maybe differently than you expected. But real problems have a way of finding solutions, and the people closest to those problems are usually the best positioned to build them.

You don’t need to quit your job to start. You just need to actually start.

✨ SaugaTech Epilogue

The weather can’t decide if it’s spring or winter. The tech industry can’t decide if it’s a bubble or a revolution. Honestly, neither matters that much. What matters is whether you’re solving something real for someone who needs it. The rest is noise.

Every company we have mentioned in this edition started the same way: someone understood a problem deeply, found the people living with it, and built something that helped. Some used cutting-edge technology. Some did not. The throughline was never the tool. It was the problem.

So if you're sitting in a coffee shop in Mississauga or Brampton, squeezing in thirty minutes between school drop-off and your nine o'clock - this is for you. You don't need a co-working space and a pitch deck to get started. You don’t need a big launch. You don’t need investors. You just need to find ten people who share that problem and build something small enough to ship this weekend.

The problem you can’t stop thinking about is probably the problem you understand better than most people ever will.

Start there. Stay consistent. If the problem is real, things will happen.

See you on March 28.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Subscribe now

Share

Hope you are doing well and enjoyed the fake spring for the couple of days it lasted. With another bout of snow coming in, think we should focus on enjoying indoor activities for some more time, while the weather figures itself out.

Speaking of which, we’re back with another in-person meetup, another hands-on session coupled with a networking opportunity with fellow techies.

Think It, Build It: Vibe Coding with AI Agents

📅 Saturday, March 28, 2026, 1-3 pm (Doors open at 12.30 pm)
📍 IDEA Mississauga, Square One

🎟️ More Details and RSVP Here

What You Need

Just bring a curious mind and maybe an idea you’d like to build. No advanced coding experience required. Laptops (full charged) recommended but not mandatory.

While the session is designed for all skill levels, its also a unique opportunity to:

• Network with local founders, senior engineers, PMs and tech leaders.

• Collaborate on the practical applications of vibe coding in your day to day work.

• Contribute to the growth of the Mississauga and GTA tech ecosystem.

RSVP & Secure Your Spot

Looking forward to see everyone on 28th March.

Best,

The SaugaTech Team

Supported by Invest Mississauga

P.S. Join the WhatsApp Group for early announcements and community discussions.

Share

Hi SaugaTech Community,

In the last edition of SaugaTech Compass, we wrote about how OpenAI hired Peter Steinberger after Anthropic’s legal team pushed him away. Strategic opportunism, we called it.

Little did we know that the same pattern is going to play out again in just a matter of days, but with way higher stakes.

Last Friday, Trump banned every federal agency from using Anthropic’s technology. The Defense Secretary labeled them a “supply chain risk”- a language usually reserved for Chinese companies. Hours later, OpenAI announced a deal to put their models on Pentagon networks.

Same concerns about autonomous weapons and mass surveillance. Same stated principles. Completely opposite outcomes.

Here’s what makes this interesting: Dario Amodei, who runs Anthropic, used to work for Sam Altman at OpenAI. They built GPT-2 and GPT-3 together. Then they had a falling out over how fast to move versus how careful to be. Dario left in 2021 to start Anthropic as the “safety-first” alternative.

Now they’re running the two most valuable AI companies in the world, locked in a very public fight about principles versus pragmatism. For those of us building in the GTA, this matters—because eventually, it’s a story of choices, choices that we all have to make one way or the other, about how our work gets used.

🚀 First Things First: SaugaTech March Meetup

SaugaTech March meetup is scheduled for the 28th March at IDEA Mississauga, Square One. We will be announcing the topic and open early registrations this Friday on SaugaTech WhatsApp Group . Join the group to register early, lest we run out of spots like last time. We will also post a formal communication about the event early next week on this Substack group as well.

Now coming back to our main story.

Anthropic X Open AI | How We Got Here

Dario Amodei joined OpenAI in 2016 as VP of Research. He helped build GPT-2 and GPT-3, and co-invented the technique that made ChatGPT possible—getting AI to learn from human feedback.

By 2020, tensions were building. Dario and other senior researchers believed these models would become extraordinarily powerful, but that power needed serious safety work. According to people who were there, disagreements over strategy and commercialization got personal.

In early 2021, Dario and his sister Daniela left with a dozen other OpenAI researchers to found Anthropic as the explicit “safety-first” alternative.

Fast forward to 2026: Anthropic is worth $380 billion. OpenAI is worth $500 billion. And they’re fighting over Pentagon contracts.

The Pentagon Standoff

Last July, the Department of Defense offered $200 million contracts to four AI companies—Anthropic, OpenAI, Google DeepMind, and xAI—to deploy models on classified military networks.

The Pentagon wanted access for “all lawful purposes.” No exceptions.

Anthropic pushed back. They wanted two explicit prohibitions: no fully autonomous weapons, and no mass domestic surveillance of Americans. Everything else—intelligence analysis, logistics, cybersecurity—was fine.

The Pentagon refused. The Under Secretary called Dario a “liar” with a “God complex.” Trump posted that Anthropic was run by “Leftwing nut jobs.”

On February 27, negotiations collapsed. By 5:02 PM, Anthropic was blacklisted.

Hours later, OpenAI announced their deal. Sam said OpenAI shared the same red lines but had reached agreement with “strong contractual protections.”

What was different? Anthropic wanted contractual prohibitions preventing those uses even if laws changed. OpenAI agreed to “all lawful purposes” and trusts current laws and internal monitoring.

Critics point out the loophole: if law allows mass surveillance—and existing rules let agencies buy personal data without warrants—then OpenAI’s models can be used for exactly what Anthropic tried to prevent.

Sam admitted the deal was “definitely rushed” and acknowledged: “it clearly exposes us to some risk.”

Why OpenAI Couldn’t Walk Away

Here’s the story behind the story that’s not being talked about: OpenAI most likely couldn’t afford to walk away.

The company is burning cash. ChatGPT traffic is declining. Enterprise market share dropped from 50% to 27%. Google’s Gemini has been leapfrogging them. DeepSeek proved you don’t need billions in compute to compete.

OpenAI introduced ads in ChatGPT—something Sam said he “hated.” They’re planning a late-2026 IPO at a $750 billion valuation.

This is OpenAI’s existential year. They can’t afford to be on the wrong side of government. Not when they need regulatory support for IPO. Not when defense contracts provide stable revenue. Not when getting blacklisted would crater their valuation.

Anthropic is profitable with $1.16 billion in monthly revenue. They could afford principle.

OpenAI couldn’t.

The Monday Scramble

By Monday, the backlash hit.

Claude overtook ChatGPT in the App Store. Chalk graffiti covered OpenAI’s offices. QuitGPT launched. OpenAI employees questioned whether the deal had real protections.

Sam admitted they “shouldn’t have rushed” and it “looked opportunistic and sloppy.” He called the Pentagon to rework the contract.

The amended deal now prohibits using OpenAI technology on “commercially acquired” data—closing the loophole where agencies buy personal data from brokers. It states: “the AI system shall not be intentionally used for domestic surveillance of U.S. persons.”

But legal experts remain skeptical. The contract “does not give OpenAI an Anthropic-style right to prohibit otherwise-lawful government use.” It just says the Pentagon can’t break current laws—and laws change.

As MIT Technology Review put it: OpenAI “settled for softer legal boundaries” while Anthropic pushed for moral ones.

The Big Question : Who Decides

Lets strip away the drama for a minute. The underlying question is very simple: Who decides how powerful technology gets used?

Anthropic says builders have moral responsibility. If you know your technology could violate human rights, you have an obligation to prevent it—even if it costs money.

OpenAI says government, not companies, should decide how lawful tools are deployed. If laws need changing, that’s democracy—not corporate veto power.

Both positions are coherent. Both have risks.

Anthropic’s risk: Companies wield outsized influence over national security. If every AI lab demands veto power, government loses autonomy.

OpenAI’s risk: “All lawful purposes” becomes a blank check. Surveillance laws from the 1970s don’t contemplate AI-powered monitoring. Trusting future administrations won’t exploit gaps feels optimistic.

What This Means for GTA Builders

Most of us aren’t building AI for the military. But we’ll face similar choices.

Do you let advertisers microtarget vulnerable populations? Do you build addictive features or dark patterns which drive users to make sub-optimal purchase decisions? Do you sell to customers whose use cases make you uncomfortable?

Anthropic drew red lines and lost $200 million. OpenAI signed and faced revolt. Both paid costs.

1. Principles are expensive. Dario walked away from $200 million over restrictions that wouldn’t have changed current military operations. That’s either integrity or virtue signaling, depending on perspective.

2. Pragmatism has costs. OpenAI got the deal but took reputational damage. By Monday they were scrambling to amend and admit mistakes.

3. Clarity matters most. Both claimed the same values but reached opposite conclusions. The worst outcome is claiming principles you won’t enforce or accepting terms you don’t believe.

You can’t avoid ethical questions by building neutral tools. Every technology has uses its creators never intended.

The question isn’t whether you’ll face these choices. It’s whether you’ll make them deliberately.

✨SaugaTech Epilogue

For Dario and Sam—former colleagues who built modern AI together before their bitter split—the rivalry is now total. They compete for customers, talent, funding, and control over how their technology shapes warfare and surveillance. One chose principles and lost $200 million. One chose pragmatism and faced employee revolt. Both paid different costs, but here’s what neither had to pay: the cost of not knowing where they stood.

That clarity matters more than most people realize. When you’re honest about your boundaries—what you will and won’t build, who you will and won’t serve—you sleep better. Not because you made the “right” choice by some universal standard, but because you made your choice deliberately.

For those of us building in the GTA, the lesson isn’t to pick Team Dario or Team Sam. It’s to know your own boundaries before someone else tests them. Build fast, ship often, take risks—but know the line you won’t cross.

The companies that thrive long-term won’t be the ones that avoided tough decisions. They’ll be the ones that made those decisions with clear eyes and lived with the consequences without losing themselves in the process.

Technology will always have unintended uses. Platforms will always face choices about what they enable. You can’t avoid that by building “neutral” tools. You can only decide ahead of time what you stand for, so when the pressure comes—and it will come—you already know your answer.

Keep building. Keep learning. But keep your compass close.

Looking forward to seeing you on 28th March at our next meetup.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Subscribe now

Share

Hi SaugaTech Community,

Last Thursday, someone in our WhatsApp group asked a simple question: “Has anyone tried ClawdBot? Just connected it to my Telegram and it seems quite interesting”. Within an hour, the thread had 40+ messages. People were debating whether to install it, sharing concerns about the security warnings, discussing how to sandbox it properly, basically how to get it working with minimum risk.

The conversation perfectly captured the tension: this thing looked incredible, but can we trust it?

Even if you missed that discussion, chances are someone in your techie circle mentioned OpenClaw. Maybe you saw the GitHub star chart that broke the internet. Maybe you read about the Austrian developer who joined OpenAI on Valentine’s Day. Maybe you just wondered why everyone keeps talking about lobsters.

Here’s what actually happened: A solo developer built a weekend project that gained 200,000+ GitHub stars in 10 weeks—faster than Linux, Kubernetes, or any open-source project in history. Anthropic tried to shut it down with trademark complaints. Crypto scammers hijacked the name and launched a fake $16 million token. Security researchers called it a nightmare. And OpenAI swooped in, hired the developer, and turned what could have been a legal mess into their strategic advantage.

This is the story of ClawdBot, Moltbot, and OpenClaw. It’s about how Anthropic missed a trick, how OpenAI capitalized on that mistake, and what it all means for those of us building in the GTA.

🚀 First Things First: Join the Community

Before we dive in, quick reminder:

• We’re planning our next meetup, write to us at saugatechproduct@gmail.com if you want any specific topic to be covered or want to volunteer as a speaker

• Follow our journey on Linkedin, we just posted about the great event we had at IDEA Mississauga about Building your first AI Agent.

• If you’re on whatsapp, join the SaugaTech WhatsApp Group to stay in the loop

Back to ClawdBot now.

The Beginning: A Weekend Hack in Vienna

Peter Steinberger is not your typical developer. He bootstrapped PSPDFKit—a PDF SDK company—for 13 years without venture capital and sold it for over €100 million in 2021. Then he took three years off and got bored.

In November 2025, he started tinkering with AI agents. Not the conversational chatbots everyone was building. Actual agents that could do things—read your emails, manage your calendar, execute terminal commands, control your computer.

He called it “WhatsApp Relay.” The idea was simple: message an AI through WhatsApp, and it would perform tasks on your home computer. He named it ClawdBot—a playful pun on Anthropic’s Claude, which powered the reasoning layer.

It was project number 44. The previous 43 had gone nowhere.

This one was different.

The Viral Explosion

By late January 2026, ClawdBot had crossed 9,000 GitHub stars. Then something wild happened. Matt Schlicht launched “Moltbook”—a satirical social network exclusively for AI agents. OpenClaw agents started creating profiles, friending each other, and generating content. The internet lost its mind.

Within days, ClawdBot was gaining 710 stars per hour. Over one weekend, it added 34,168 stars. The project drew 2 million visitors in a single week. By mid-February, it had crossed 200,000 stars and 36,000 forks.

Users created 1.5 million AI agents on the platform. Developers were running OpenClaw instances that could book flights, manage spreadsheets, debug code, and coordinate across WhatsApp, Telegram, Slack, and Discord.

What started as a passion project was suddenly turning into a global infrastructure.

Anthropic’s Legal Move (And the Backfire)

On January 27, 2026, Anthropic’s legal team sent Steinberger a trademark complaint. “ClawdBot” was too close to “Claude.” They asked him to rename it.

Steinberger didn’t fight. He understood trademark law. At 5 AM, in a chaotic Discord brainstorm with the community, they settled on “Moltbot”—a reference to how lobsters shed their shells to grow.

The rebrand to Moltbot lasted three days. The community hated the name. “It never quite rolled off the tongue,” Steinberger later wrote. On January 30, he renamed again—this time to OpenClaw. Domains secured. Trademark searches cleared. MIT license formalized. The project was now positioned as “model-agnostic agentic infrastructure.”

Here’s where Anthropic’s legal move backfired spectacularly.

By forcing the rebrand, they severed the association between OpenClaw and Claude. What started as a project built on top of Anthropic’s API became a platform-agnostic tool that worked with OpenAI, DeepSeek, Google, and anyone else. Anthropic turned a brand ambassador into a competitor.

And OpenAI was paying attention.

OpenAI’s Strategic Play

While Anthropic was sending cease-and-desist letters, OpenAI was watching OpenClaw become the fastest-growing repo in GitHub history.

They saw what Steinberger had built: a fully autonomous agent platform that ran locally, worked across messaging apps, and had already created 1.5 million agent instances. This wasn’t vaporware. This was production infrastructure with real users solving real problems.

On February 14, 2026—Valentine’s Day—Peter Steinberger announced he was joining OpenAI. The deal was elegant. Steinberger would lead OpenAI’s personal agent development. OpenClaw would transition to an independent open-source foundation. OpenAI would sponsor the project and commit to keeping it open. Steinberger retained creative control over the community.

OpenAI turned Anthropic’s legal enforcement into a hiring opportunity. They got the developer, the community, the momentum, and the positive PR of supporting open source. All while Anthropic’s lawyers were still drafting trademark complaints.

The Open Source Win

Strip away the corporate drama for a second. The real story here is what open source just proved.

One developer, working alone, using AI agents to write code, built infrastructure that rivaled proprietary products from billion-dollar companies. He did it in 10 weeks. He had no team, no VC funding, no marketing budget.

And he gained 200,000+ GitHub stars—more than Linux, Kubernetes, or TensorFlow gained in their first year.

OpenClaw now runs on local hardware. It’s privacy-conscious. It’s extensible. And it’s community-driven. Over 1,000 developers have contributed code. Thousands more have built “skills”—modular plugins that extend what the agent can do.

Yes, there were security issues. Researchers found vulnerabilities. Cisco called it a nightmare. Gartner issued warnings. China’s Ministry of Industry issued a public alert.

But none of that slowed adoption. People kept installing it. Because the value was real.

The market spoke: developers want agents that do things, and they want them now. They’re willing to trade some security risk for functionality, especially when they control the infrastructure.

This is what winning looks like for open source in 2026.

What This Means for Builders in the GTA

The OpenClaw saga offers three lessons that matter for those of us building in Mississauga, Toronto, Brampton, and across the GTA.

1. If Steinberger Can, We Can Too

Peter Steinberger is not superhuman. He built 43 projects that went nowhere before project 44 took off. He didn’t have insider access or special compute. He just used publicly available LLMs and built something people wanted.

There’s nothing stopping someone in Mississauga from doing the same thing. The tools are identical. The APIs work the same. The open-source ecosystem is global.

Steinberger isn’t alone. Look at what solo developers have already done:

Pieter Levels (Dutch) built Nomad List, Photo AI, and RemoteOK—generating $3M+ annually—completely solo. No team. No VC funding. Just PHP, a laptop, and relentless shipping. He launched 40+ projects before finding what worked. His competitive advantage? Shipping before things are ready and iterating based on real usage.

Maor Shlomo built Base44—an AI-powered no-code app builder—in six months and sold it to Wix for $80 million. Solo founder. No VC. He shipped 13 updates per day in the early weeks, got to 10,000 users in month one through virality, and reached profitability while scaling.

The pattern is clear: execution and taste beat resources. Steinberger understood what developers needed because he was solving his own problem first. He iterated fast. He built in public and let the community shape the product.

That playbook works here. You don’t need to be in Silicon Valley. You don’t need $100 million in funding. You need to understand a problem deeply, build a solution fast, and ship it where your users already are.

2. Open Source Is a Moat

When you open-source a project that solves a real problem, you’re not just sharing code—you’re creating a movement. Contributors fork it, build plugins, evangelize it, and defend it.

That’s what happened with OpenClaw. When Anthropic forced the rebrand, the community rallied. When scammers launched the fake token, the community exposed it. When security researchers issued warnings, the community hardened the code.

Open source done right creates defensibility that proprietary companies can’t match.

3. Platforms Will Try to Co-Opt You

OpenAI didn’t acquire OpenClaw. They hired Steinberger and committed to keeping the project open. They get the talent, the goodwill, and the association with the fastest-growing repo in GitHub history—without shutting down a beloved open-source project.

For builders: if you build something valuable on someone else’s platform, they will eventually try to co-opt you through acquisition, copying, or hiring. The question is whether you retain enough control to negotiate from strength.

Steinberger did. OpenClaw remains open. The foundation is independent. The community continues. That’s the goal.

The Epilogue: What Happens Next

OpenClaw is now transitioning to an independent foundation. Steinberger is at OpenAI working on personal agents. The community is building faster than ever.

Anthropic, meanwhile, is facing questions about why they enforced a trademark claim that ultimately benefited their biggest competitor. The legal team was probably just doing their job. But the strategic outcome was disastrous.

OpenAI, on the other hand, just hired one of the most productive developers in the AI space and secured goodwill with the open-source community. All while their competitor handed them the opportunity.

For those of us in the GTA watching this unfold, the takeaway is simple: the barriers to building world-class software are lower than ever. One person with taste, execution speed, and a deep understanding of their users can create infrastructure that competes with billion-dollar companies.

Peter Steinberger did it from Vienna. Someone in Mississauga, Toronto, or Brampton could do it next.

The tools are here. The community is here. The only question is who’s building.

Let’s keep building, Let’s keep learning, Together.

Team SaugaTech

CONNECT | COLLABORATE | INNOVATE

Share

Subscribe now