This Week’s Playbook: 3 Things You Can Do Right Now

1. Fact-check one AI-written thing before it leaves your building. State Farm’s outside lawyers admitted AI put seven completely made-up court cases into a real legal filing. The fix is making drafting and checking two separate steps. Try this: paste your next AI-written proposal back into ChatGPT or Claude with: “Audit this draft for factual risk. List every claim, number, quote, and named source I must verify manually. Mark anything unsupported as DO NOT PUBLISH. Do not invent or repair sources.” Then check the list yourself.

2. Find out whether AI can actually use your website. A growing share of your future customers won’t be people. They’ll be AI assistants shopping and booking on someone’s behalf, and most sites quietly block them. Is Agentic is a free scan from Vercel that runs 100+ checks and tells you exactly what’s in the way. Try this: paste in your homepage, read the report, forward the top three fixes to whoever runs your site. Five minutes.

3. Do a ten-minute AI bill audit. Open your billing page and list every AI charge. In my experience most owners are paying for four tools and using one. Try this: Upload your statement to your AI tool and have it check what are you spending money on.

The Big Story: Porsche Just Told You What to Budget for AI

Here is the most useful number I have seen all year.

Porsche signed a five-year, $1.5 billion deal with Tata Consultancy Services to deploy AI across the entire company: engineering, manufacturing, operations, customer experience. It also sold TCS its own in-house IT consulting arm, MHP, for another €320 million. A company that owned a 4,500-person consultancy decided to pay someone else to come in and do this.

Luke Pierce ran the ratio and it is the part worth writing down. That’s roughly $292 million a year on about $42 billion in revenue. Call it 0.7% of revenue.

Now put that next to the two numbers everyone quoted this month. McKinsey’s State of AI 2026 found 89% of organizations use AI somewhere, and only 37% can point to any effect on profit, the same share as last year. A working paper covering nearly 6,000 executives across the US Federal Reserve, the Bank of England, Germany’s Bundesbank and an Australian panel asked whether AI had changed their sales per employee. 89% said no impact at all.

Why you should care: Those two facts look contradictory and they aren’t. Almost nobody is underspending on AI subscriptions. What they’re underspending on is the boring part Porsche just wrote a billion-dollar check for: somebody walking in, mapping how the business actually runs, and rebuilding the process around the AI. Seats are cheap. Redesign is what costs money, and it’s the only line item that shows up in profit.

What this means for you: Do the 0.7% math on your own revenue and write the number down. That’s your annual AI budget, and it is probably larger than what you spend and smaller than you feared. Then split it honestly. If more than a third is going to software licenses, you are buying tools and calling it a strategy. The rest belongs to the work of changing how a process runs, which is exactly what we do at PHOS AI LABS. Reply with your revenue number and I’ll tell you what I’d spend it on.

The News That Actually Matters

Bill Gates says the hard part is starting now

Bill Gates published a long piece on Wednesday arguing the shift to AI will be one of the most turbulent stretches in modern history, and that nobody is preparing for it. It’s the least hype-y thing a tech billionaire has written about AI, and it’s worth 15 minutes of your day.

His core argument: every past technology shift replaced human muscle or human speed, and gave people decades to adjust. This one substitutes for human thinking, runs on devices you already own, and needs no training period because it adapts to you. He expects it to hit law, customer service, medicine, software and manufacturing over a decade, not over generations. His proposals are genuinely uncomfortable for his own industry: tax AI usage and robots the way we tax payroll, and designate a set of jobs as Human Reserved, meaning we could automate them and choose not to.

Your AI can finally get past the login screen

The most annoying limit on AI assistants came off this month. ChatGPT Work can now sign into password-protected sites and finish the job instead of stopping at the login page, AND ITS BEAUTIFUL. I’m again in love with ChatGPT (sorry, Claude).

You type the username, password and code yourself, and OpenAI says those never reach the AI model. Claude Cowork shipped a built-in browser the same week that clicks, types and fills forms in a side panel while you watch... which is not as good as ChatGPT, imo.

Until now, “AI agent” mostly meant AI that could read the open web. Ninety percent of the boring work in your business lives behind a login: your invoicing tool, your supplier portal, your ad accounts, your state filings. That door just opened.

Why it matters for you: This is the first version of these tools that can genuinely run errands. Chase a supplier for a delivery date. Pull last quarter’s numbers out of a system with no export button. Renew the registration nobody wants to renew.

The AI Office — Tools You Can Use This Week

Slack agents — The most underrated one on this list. Slack now lets you @mention an AI agent like a teammate in a channel or thread, with prebuilt templates for onboarding, support and customer questions. The reason I like it isn’t the technology, it’s the visibility: Shopify found its Slack agent taught the rest of the team how to use AI just by working in public where everyone could see it. Try this: put one agent in one channel where the same question gets asked every week, and let your team watch it answer. Best AI training you’ll run this quarter, and it costs you nothing extra.

Grok Bot — xAI’s version of the always-on AI coworker. Each bot gets its own cloud computer, logs into your apps, works while your laptop is closed, and can hand tasks to other bots. The Verge has a good walkthrough. My honest take: impressive, and not yet for you. It starts at $120 per seat per month for teams and $200 a month for individuals, it’s in beta, and it’s a real security decision to hand any bot your logins...

Meterless — This one clicked for me. It records the exact steps your AI took to do something and saves it as a replayable “mission,” so update the spreadsheet, then email the team runs on command instead of being re-explained every time. Try this: walk it through one weekly report, save it, and never explain that report again. Free to try.

Claude Academy — Anthropic’s free courses on how to actually use Claude, including where it falls short. Relevant because of a randomized study of 1,559 Pakistani judges that found AI plus targeted training raised case resolution 6.3%, and the training was the active ingredient, not the tool. Try this: assign one course as this month’s team homework. Costs you nothing but calendar space. If you’d rather have someone run hands-on training with your actual workflows, that’s a thing we do.

The “So What” for Business Owners

Before you spend more on AI, stop overpaying for what you already run

Counterweight to the Big Story. AT&T is routing a large share of its internal AI work to cheaper models and saving the expensive top-tier AI for genuinely hard jobs. Reported result: costs down 80 to 90% on some applications. On coding work, cost dropped 56% for roughly a 2% quality tradeoff.

Two percent.

The habit underneath is simple: not every task deserves your most expensive AI. Summarizing a call, cleaning a spreadsheet, drafting a routine reply. A cheap model does that fine. Analyzing a contract or building a pricing model goes to the big one. The industry calls the software that decides this a “router,” and the money agrees it matters, with Stripe reportedly buying OpenRouter for $7B and Ramp launching its own.

What to do: You don’t need a router. You need the habit. On a flat subscription, the cost shows up as burning through your usage limit by Thursday, so use the fast mode for routine work and save deep-thinking mode for problems that deserve it. If you pay per use, start cheap and move up only when it visibly fails. One honest caveat: this only works if you can tell what “good enough” looks like on real work. If you can’t judge the output, cheaper isn’t savings, it’s a coin flip. If you want this for your team, that’s what we built Nexus for.

What I’m Building This Week: Bringing Our Software to Where People Already Work

This week at LOW / CODE we’re adding MCPs to our four internal systems: the dashboard, the HR platform, the CRM, and our project management tool.

Quick definition, because the acronym is everywhere now and nobody explains it. MCP stands for Model Context Protocol. Think of it as a standard plug. Once your software has one, any AI tool can connect to it and actually do things there, instead of you copying and pasting data back and forth. Anthropic released it, and OpenAI, Google and most of the rest adopted it, so one plug now fits nearly everything.

Here is the problem it solves for us. My team lives in ChatGPT and Claude all day. Every time someone needs a real number, they leave, open the dashboard, hunt for the record, copy it, come back, paste it, and try to remember what they were doing.

With the MCPs in place, none of that happens. You ask your AI tool who is over hours this week, what stage the Acme deal is in, or which projects slipped, and it answers from the real system in the window you were already in. The software comes to the person instead of the person going to the software.

If this works the way I think it will, it becomes a standard part of what we hand clients. Reply if you want to see it when it’s done.

Quick Hits

• Private equity is putting AI engineers inside the companies it owns — Blackstone and Hellman & Friedman are embedding a 160-person engineering team, backed by a $1.5 billion partnership with Anthropic, directly into portfolio businesses. Same conclusion as Porsche: the bottleneck isn’t tools, it’s that nobody owns making them work.

• Google paid $10M for a bankrupt airline’s internal paperwork — In the Spirit Airlines auction, the prize wasn’t planes. It was anonymized internal records, spreadsheets and booking history. Your own boring operating history is the same asset, and it’s scattered across inboxes and someone’s desktop.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

Back after a two-week break (I was traveling — thanks for your patience). Here’s your action list. Pick one and knock it out today.

1. Make your AI stop writing like a robot. The #1 complaint about ChatGPT and Claude is that the writing sounds like AI. There’s a two-line fix. Go into your AI’s settings (the “custom instructions” box) and paste: “Always write in Simplified Technical English” and “Talk to me like I have ADHD — short, skimmable, bullets over paragraphs.”
   Try this: add both lines, then ask it to explain something you’d normally struggle to read. Night and day.

2. Write your business a one-paragraph “who we are” file. Every time you open a fresh AI chat, it knows nothing about you — so you re-explain your business for the hundredth time. Fix it once: write a short paragraph (what you sell, who you serve, your tone) and paste it into your AI’s custom instructions. Now every answer starts smart.
   Try this: ask Claude to interview you with 5 questions, then have it write the paragraph for you.

3. Give one team member an AI they can just text. Your non-techy staff will never “learn a tool” — but they’ll message a coworker. Lindy Teammate lets anyone text an AI helper straight from Slack, like pinging a colleague.
   Try this: drop it in one channel and let your team ask it the repetitive “what’s our policy on X / where’s that file” questions all day. Reply and I’ll tell you how we set these up for clients.

The Big Story: Everyone’s Getting a “Personal AI Agent” — And It’s Simpler Than the Hype Suggests

The buzzword of the month is the personal AI agent — an AI that doesn’t just answer questions but has its own memory, its own job, and can actually do things for you. xAI launched “Grok Bot” this week, following OpenClaw and Hermes before it, and Mark Zuckerberg published a 6,000-word manifesto arguing everyone will soon have a personal AI that knows them and their goals. IDK why Mark hasn’t built this already in WhatsApp, though… a little more action, please, Mark?

Here’s the part nobody’s saying plainly: there’s no magic. A “personal agent” is three boring things — a set of instructions (”you’re my money manager”), a few connected tools (Gmail, calendar, Slack), and a text file where it writes down what it learns. That’s it. You can build a “Jarvis” for your business today without waiting for a $200/month product.

Why you should care: The winners of this shift won’t be the people who read about personal agents. They’ll be the ones who gave one a real job - “draft my follow-ups,” “pull my weekly numbers,” “answer new leads at 2am” - and let it run.

What this means for you: Don’t overthink it. Pick one recurring job you’d hand a new assistant, write down the instructions like you’re training a person, and point an AI at it. That’s exactly the “AI employee” we build at LOW/CODE - the difference is just how carefully the instructions and guardrails are written.

The News That Actually Matters

He Asked His AI to Book a Gym Class. It Committed a Cyberattack Instead.

This one’s worth reading twice. An Australian man asked his personal AI assistant (running on the open-source OpenClaw framework) to get him into a full gym class - he was 4th on the waitlist. Instead of giving up, the AI found a hole in the gym’s booking system and cancelled a stranger’s reservation to bump him up. Then it cheerfully confessed what it had done.

Funny story, serious lesson. As these agents get the power to act - click buttons, send messages, spend money - a vague instruction can turn into a real problem. Gartner now expects 40% of business apps to include AI agents by year’s end, up from under 5% in January.

Why it matters for you: The value of AI is that it does things. The risk is the same sentence. Before you let an AI act on your behalf, give it a lane — what it’s allowed to touch, what needs your sign-off. Be super clear, be verbose: remember, you’re training the most enthusiastic 18 year old inexperienced but knowledgable genius.

Your AI Work Isn’t Locked In Anymore

Quieter but genuinely useful: ChatGPT now lets you import and sync your projects, saved instructions, and history from other AI tools like Claude. For two years, the fear with picking an AI was “what if I bet on the wrong one and lose all my setup?” This is important because the pendulum is going back from Claude to OpenAI. Better models, cheaper (or at least more tokens for your subscription) and follows instructions better.

Why it matters for you: That fear is easing. You can try Claude, ChatGPT, or Grok for a month and move your work if a better/cheaper one shows up - and this week Grok 4.6 launched near the top of the pack for a fraction of the price, so the “cheaper option caught up” story keeps repeating.
What to do: don’t marry one AI. Run the same real task through two of them this month and keep whichever nails your work.

The AI Office — Tools You Can Use This Week

This is the heart of the newsletter. Two worth a look:

Natter — Type with your voice, anywhere on your computer, with live transcription as you talk. If you think faster than you type (most of us do), you’ll draft emails, notes, and replies in a third of the time. (Mac only for now.)

Grok Imagine 2.0 — A new image tool that just hit #2 for editing on the public leaderboards. Describe a promo graphic or drop in a photo and ask it to fix the background, swap colors, or add text - no designer needed. Pair it with Canva for the finishing touches.

The “So What” for Business Owners

The move isn’t “get yourself an AI.” It’s “give your team one.”

Look at the pattern this week: personal agents everywhere, Zuckerberg promising everyone their own AI, tools you text like a coworker. Most owners are still thinking about AI as their personal productivity hack. The leverage is one level up: give every person on your team an AI teammate for the boring parts of their job.

The math is simple. One owner using AI saves a few hours a week. A ten-person team each handing their most repetitive task to an AI is a different company.

What to do: This week, ask each person on your team one question - “What do you do every week that a smart assistant could do for you?” Those answers are your AI roadmap. If you’d rather have someone build and wire up those AI teammates properly - with the guardrails so you don’t get the gym-class treatment - that’s exactly what we do at PHOS AI LABS. We also run hands-on AI training for teams that want to learn it themselves. Reply and we’ll scope one with you.

Quick Hits

• Meta is open-sourcing its AI again. Its new free “Muse Glimmer” model is built for exactly this agent stuff. Great for developers, changes nothing about your Monday - filed here so you can safely ignore it.

• Anthropic will start watermarking all text written by Claude. Worth knowing if you auto-write customer emails or content - “invisibly tagged as AI” is becoming standard, and “made by a real human” is quietly turning into a selling point. Check the explanation here:

• OpenAI’s mystery gadget leaked — reportedly a screen-less smart speaker, ~$300, coming in 2027. Fun to watch, nothing to buy. Pure noise for now.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

Before the news — here’s your action list. Pick one and knock it out today.

1. Switch to a cheaper AI model and pocket the difference. Anthropic just launched Claude Opus 5 — nearly as smart as its top model at half the price. If you or your team default to the most expensive AI option out of habit, run this week’s real task on the cheaper one and see if you can even tell. Most people can’t.

2. Get free AI training from OpenAI (yes, actually free). OpenAI opened a program built specifically for small businesses — webinars and in-person “AI academies” that show you how to use ChatGPT for real day-to-day work. Send one team member. It costs you nothing but an hour.

3. Give your AI a job title - and only the keys it needs. If you’ve connected an AI to your Gmail, Drive, or calendar, check what it’s actually allowed to touch. Treat it like a new hire: grant the minimum access to do the job, nothing more. Five minutes now saves a mess later (more on why below).

4. Teach an AI to do a task by recording your screen. Claude’s “Record a skill” lets you screen-record yourself doing something repetitive — pulling a report, formatting an invoice — while you talk through it, then reuses it. Try it on the most boring thing you did this week.

The Big Story: The Software You Pay For Every Month Just Got Put on Notice

For years you’ve rented software - a tool for support tickets, one for scheduling, one for expenses, one for your CRM. This week, the ground shifted under all of it.

OpenAI launched Presence, a product that drops AI agents inside a company’s own systems — answering customers, pulling from internal data, following your policies, and handing off to a human when needed. In plain terms: the AI stops being a chatbot you visit and becomes a worker that lives inside your business. Wall Street got the message immediately — shares of big software names like Workday, Atlassian, and HubSpot fell as investors bet AI agents will do what those subscriptions do.

It’s not just OpenAI. The same week, Meta’s assistant started plugging into your inbox and calendar to actually do things, and the big players kept racing to package AI for people who don’t code — OpenAI’s free small-business program, Bluehost’s no-code AI agents that build and run your store. For two years AI was built for engineers. That’s over.

Why you should care: A lot of what you pay for monthly is really just “a place to store data and a few workflows on top.” That’s exactly the layer AI agents are starting to absorb.

What this means for you: Don’t cancel everything tomorrow — Presence is aimed at big enterprises for now. But open your subscriptions list and ask one question next to each line: “Is this a system of record I truly need, or a workflow an AI could run?” You’ll be having that conversation for real within a year. Start the audit now.

The News That Actually Matters

AI Got Cheaper and Better — Again

Anthropic released Claude Opus 5 — its fourth new model in under two months — that matches its flagship on everyday work at roughly half the cost. This is the whole game in one headline: wait a few weeks, and AI gets smarter and cheaper.

Why it matters for you: Two takeaways. First, you almost never need the priciest model for real business tasks — drafting emails, summarizing docs, answering customers. The “good enough and cheap” tier keeps getting genuinely good. Second, stop agonizing over which AI to lock into. The prices and rankings shuffle every month. Pick one that works today, get value now, and switch when it’s obviously worth it. Loyalty to an AI vendor earns you nothing.

An AI Agent Went Rogue — and It’s a Lesson, Not Just a Headline

Here’s the wild one: OpenAI admitted that during a security test, one of its AI agents broke out of its sandbox, got onto the internet, and hacked into another company (Hugging Face) — to cheat on a test. OpenAI reportedly didn’t notice for days. In response, Nvidia and a group of tech companies launched an “Open Secure AI Alliance” to build better safety tools.

Scary story, but you don’t run a frontier AI lab. Here’s the part that is for you: an AI agent is only as safe as the access you give it.

Why it matters for you: As you start letting AI agents act inside your email, files, and tools, treat them exactly like employees. Give them the least access they need. Watch what they do at first. Turn off “always allow” and keep a human in the loop for anything that spends money or sends messages. This isn’t paranoia — security researchers this week named AI agents the fastest-growing new risk for businesses. The fix is simple discipline, not a big budget.

The AI Office — Tools You Can Use This Week

This is the heart of the newsletter. Four worth a look:

Meta AI — Meta’s assistant can now connect to your Google Calendar and Gmail and actually do things — send you a daily brief, help with research, keep a project moving. It’s free and already in the apps you use.

Intuit’s new AI business card — A business credit card that plugs straight into QuickBooks, giving you real-time spending visibility and automatic expense categorization. If bookkeeping is the chore you always dread, this quietly kills a big chunk of it.

Gemini Notebook (formerly NotebookLM) — Google’s “brain for your documents” is now pre-installed on new Samsung phones and still the easiest way to make AI answer questions about your files. Upload SOPs, contracts, or your handbook and just ask. We LOVE this at LOW/CODE.

Shopify → DoorDash — If you run a store with a physical location, U.S. Shopify merchants can now push their catalog onto the DoorDash marketplace with no re-uploading — it syncs in real time. A whole new sales channel for basically zero setup.

The “So What” for Business Owners

Your team is already using AI you never approved

Here’s the trend hiding behind all the flashy launches: while you’re deciding whether to adopt AI, your employees already have. It even has a name now — “shadow AI” — staff quietly using whatever tools they like, often pasting company data into them, with no rules. Experts warned this week that AI adoption is racing ahead of any governance to keep it safe. (BTW, we provide large companies with an AI Governance consulting service.)

You don’t fix this with a ban — bans just push it further underground. You fix it with a one-page list.

What to do: Write a simple “here’s what’s approved” doc for your team: which AI tools are OK, what data should never be pasted into them (client info, contracts, anything private), and who to ask when unsure. Fifteen minutes of clarity beats a policy binder nobody reads. If you’d rather have someone set this up properly — approved tools, safe access, and actual AI “employees” that handle full workflows instead of just chatting - that’s exactly what we do at LOW/CODE Agency. Reply to this email and we’ll scope one with you.

What I’m Building This Week: What an AI Audit Actually Turns Up

This week I ran a two-session AI Audit for a mid-market distributor - the kind of unglamorous, spreadsheet-run, multi-million dollar company that’s the backbone of most economies. I can’t name them, but the pattern is so common it’s worth showing you, because odds are it’s happening in your business too.

Their expensive enterprise software (an SAP system) held all the records. But every real decision was being made somewhere else — in personal Excel files, email threads, and whatever the person on shift happened to remember. We had a one-line summary for it: “The system records it, but Excel decides it.” The information never arrived at the point of decision in time to matter.

What that actually costs, in plain terms:

• Knowledge walking out the door. Their single best pricing model lived in one negotiator’s personal spreadsheet on his personal laptop. His words: “If that guy leaves, the spreadsheet leaves with him.” One resignation away from losing their edge.

• Zero visibility where it counts. In their entire delivery operation, nothing was being measured - not late shipments, not lost minutes, not the cost of penalties. A good plan was executed blind, with no alerts when reality drifted from it.

• Money frozen for no reason. Their collections clock didn’t start at delivery - it started when paperwork got uploaded. That gap was quietly freezing days of working capital, every single week.

Here’s the thing: none of this is a technology problem. They had the expensive software. It’s an information-at-the-decision-point problem - the same broken data limping through four different departments, each one patching it with its own manual layer of Excel and memory.

Why this matters for you: You almost certainly have your own version of this. A critical number that lives in one person’s head or one person’s spreadsheet. A decision made on stale data. A process nobody’s actually measuring. You don’t feel it as a crisis because everyone’s working around it — that’s exactly what makes it expensive.

That’s what an AI Audit is: focused sessions where we map where your decisions actually get made, find where information arrives too late, and show you specifically where AI would pay for itself. No pitch, no jargon — just where the money’s leaking. If you want one, reply to this email.

Quick Hits

• Gumroad is now largely run by an AI system its founder calls “Gumclaw,” and June was the first month it spent as much on AI tokens as on human payroll. Replit’s CEO says agents now touch nearly every part of running his company. Not “fire everyone” — a signal that what one person plus AI can run is changing fast.

• Cursor and Ramp launched “model routers” — tech that auto-picks the cheapest AI model that can still do the job, claiming 30–60% savings. Worth knowing if your AI bill is creeping up. We have our own custom version of this.

• Google shipped cheaper Gemini Flash models — faster and better at analyzing images on a budget. Handy if you’re building AI into an app or product.

• Kimi K3’s free weights are out — the powerful Chinese model we covered goes fully open. Watch for cheaper AI tools built on top of it.

• Adobe Project Indigo — Adobe’s experimental iPhone camera app now critiques your photos and suggests edits. Great if you shoot your own product or listing photos.

This Week’s “So What”

The signal: AI crossed a line this week — from tool you talk to into worker that acts inside your business. OpenAI’s Presence, Meta’s agents in your inbox, cheaper-and-smarter models landing every few weeks. This is the shift that actually changes how companies run.

The noise: The rogue-agent hacking drama, the software-stock sell-off, and yet another model-leaderboard update are fascinating to follow — but the only piece that changes your Monday is how you let AI into your systems, not which model wins.

Your one thing: Spend five minutes on Playbook #3 — check what your AI tools are actually allowed to access, and tighten it. As AI moves from chatting to doing, the businesses that win aren’t the ones with the fanciest model. They’re the ones who let it in carefully and put it to work.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Write one “AI house rule” for your team today. With AI now baked into half the software you touch, decide now what your people will never paste into a public chatbot — client contracts, passwords, anything with a Social Security or credit card number. One short paragraph, shared in your team chat. This costs you 10 minutes and saves you the nightmare in the Grok Build story below. Open a note, write “We never put X, Y, Z into AI tools,” and send it.

2. Rebuild one thing you skip because it’s “too expensive.” The polished proposal, the product photos, the 30-second promo video, the first draft of a policy — the stuff you don’t do because hiring it out costs too much. Netflix just used AI to make shots it literally couldn’t afford (see below). You can do the small version. Try this: pick one of those and ask ChatGPT or Claude to take a first swing today.

3. Reply and tell me the one repetitive phone-or-email task eating your week. The “call back every lead,” the “answer the same five questions,” the “send the same follow-up.” Hit reply with yours and I’ll map — free — how you’d hand it to an AI assistant. That mapping is step 1 of every engagement we do at Phos AI Labs.

The Big Story: China’s “Kimi Moment” Just Made Good AI a Lot Cheaper

Here’s a story that looks like tech-race drama but is actually about your software bill. This week a Chinese startup called Moonshot released a free, open AI model named Kimi K3 — and on some tests it beat the best (and most expensive) models from OpenAI and Anthropic, the makers of ChatGPT and Claude. Business Insider reported it even topped them at writing website code. Demand was so heavy Moonshot had to pause new sign-ups within days.

Wall Street called it a “Kimi moment” — a rerun of last year’s DeepSeek shock that wiped hundreds of billions off chip stocks. Why does a free Chinese model spook the giants? Because “open” means anyone can download it and run it cheaply. One investor told Axios that free models will soon handle 95% of business AI tasks, leaving only the hardest 5% for the pricey names.

Why you should care: The thing you’re paying a premium for is getting commoditized - fast. “The best AI” is becoming a smaller and smaller advantage, because good enough AI is quietly becoming free. Plus, a lot of tasks where Claude/OpenAI was too expensive, now can be done for a fraction of the cost... same intelligence.

What this means for you: Don’t chase Kimi K3 today... just keep in mind to be model agnostic.. more on that below on our Nexus piece.

The News That Actually Matters

Netflix Used AI in ~300 Shows This Year - Here’s the Part That Applies to You

Netflix casually dropped a big number in its earnings report this week: AI was used in roughly 300 of its shows and movies in 2026 so far, mostly behind the scenes in post-production. In the series The American Experiment, co-CEO Ted Sarandos said there were 17 minutes of AI-enhanced footage — crowds and battle scenes — made twice as fast and for half the price of the old way.

Forget the Hollywood-vs-robots debate for a second. The line that matters for a business owner is this one from Sarandos: productions used AI for shots they “just wouldn’t have been able to afford” otherwise — they’d have cut them entirely.

Why it matters for you: That’s the real unlock, and it’s not just for billion-dollar studios. The most valuable thing AI does isn’t replacing what you already pay for — it’s making the stuff you skip suddenly affordable. The promo video you never shot. The product renders you couldn’t justify. The polished pitch deck you didn’t have time for.

A Popular AI Coding Tool Quietly Uploaded People’s Private Files — a Warning for All of Us

xAI shipped a coding assistant called Grok Build, and a security researcher found it was quietly uploading users’ entire projects to xAI’s servers — including a file it was explicitly told not to touch, plus passwords and secret keys sitting in the code. xAI scrambled, killed the feature, added a privacy command, and Musk personally promised every uploaded file would be “completely and utterly deleted.”

You’re not running Grok Build. But you are pasting things into AI tools every day without reading what happens to them.

Why it matters for you: Bluevine’s data last week named data privacy the #1 reason business owners stall on AI — and stories like this are why. The fix isn’t fear, it’s a habit. What to do: before you put anything sensitive into a new AI tool, spend two minutes finding the “does this train on my data?” setting (most business tiers let you turn it off). And go write that AI house rule from Playbook #1 — this is the story it protects you from.

From the Field: What an AI Audit Actually Turns Up

We just ran our AI Audit on a mid-sized industrial distributor — the kind of company with dozens of employees and a real ERP — and the findings are too useful to keep to ourselves. (Names and details scrubbed; the pattern is what matters, and we see it everywhere.)

Here’s the punchline that surprised even them: they didn’t have a software problem. They’d bought a big-name CRM three years ago, paid for it every month, and abandoned it 100%. It sat there unused next to an ERP nobody opened. The real problem was that the business actually ran on 20-plus personal Excel files, WhatsApp, email, and paper notebooks — no single place where the truth lived.

The cost of that mess, measured: roughly 140 hours a week across the team spent just assembling information — hunting for a number, rebuilding the same report, chasing a status update. That’s about 3.5 full-time people doing nothing but finding things.

One stat that’ll feel familiar: staff scanned ~900 WhatsApp messages a day to catch the 10-15 that actually needed action, and their big weekly meeting was four hours of reviewing everything instead of just the exceptions.

Why it matters for you: This is the trend, not a one-off. Most mid-sized companies think they need to buy something — a new tool, now an AI. Almost always, the money’s already been spent and ignored, and the actual leak is that information is scattered and nobody trusts a single source. AI can’t fix that until you do; but once you do, AI is what turns 140 wasted hours back into real work.
What to do: before you buy another tool, ask one question — “where does the truth about my business actually live?” If the honest answer is “spread across 20 spreadsheets and a group chat,” that’s your first project, not the shiny AI feature.

The AI Office — Tools You Can Use This Week

Nexus — our private AI workspace — This one’s ours, and it’s the exact fix for everything above. Right now your team is already using AI, just in their personal (often free) accounts, where your business data leaks out and walks out the door when they quit. Nexus looks and feels like ChatGPT, but it lives inside your walls: grounded in your company’s knowledge, voice, and rules, connected to your CRM/ERP, and locked to the people you choose.
The part that matters this week: it has a built-in “router” that reads each question and quietly sends it to a cheap model for simple stuff and an expensive one only for the hard 5% — so you get top-tier answers at a fraction of the token bill, exactly the shift the Kimi story is about.
Try this: watch the 10-minute walkthrough, and if your team’s AI use is scattered across personal accounts, reply and I’ll show you what this looks like for your company.

The “So What” for Business Owners

The through-line this week: AI’s price is falling and its reach is growing — which means the winners aren’t the ones who buy the most AI, they’re the ones who do the most with it. Kimi made near-best AI free. Netflix used it to make the unaffordable affordable. And this week a panel of experts told the House Small Business Committee something worth pinning up:

Their advice: invest in training your team to use AI, and ask your staff where it’d help. Same message the big labs keep repeating - it’s implementation, not the tool.

Quick Hits

• Spotify started rolling out an AI assistant you can talk to. Premium users can now chat with it to build playlists and find music. I personally don’t like it. Minor on its own, but a preview of where every app is heading: a conversation box instead of a menu.

• A nonprofit raised $400M to build “free public AI.” Current AI wants an open alternative to the handful of companies that own today’s AI. Another sign the “you must rent AI from a giant” era is cracking.

What I’m Building This Week

We’re heads-down on the theme this whole issue points at: doing more with cheaper AI, safely. For clients that means AI “employees” tuned to one job — a front desk that answers the phone, an assistant that drafts quotes — built to run on affordable models instead of the priciest ones, with the data rules locked down from day one (see the Grok story for why that matters).

The other thing we’re rolling out is an AI Audit: a plain-English map of where AI actually fits in your business — which workflows to redesign first, not just which tools to buy. Want the no-jargon starter version? Grab the AI Blueprint and webinar replay here.

This Week’s “So What”

The signal: Good AI is getting cheap and open — a free Chinese model beat the giants this week, and even Netflix is using AI to do the things budgets used to make impossible. The edge in 2026 isn’t owning the fanciest model. It’s putting an ordinary one to work on the jobs you’ve been skipping.

The noise: The stock-market panic over China, the “our model beats their model” leaderboard, the billion-dollar IPO gossip. Fun to watch. Doesn’t change your Monday.

Your one thing: Pick the single task you avoid because it feels too big or too expensive to do right — and let AI take the first pass this week. And spend 10 minutes writing your team’s AI data rule so you can do it safely. That’s the whole game right now.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Put a one-page site online in 60 seconds — no account, no developer. Cloudflare Drop lets you drag a folder into your browser and get a live web link back, instantly.
   Got a menu, a price sheet, or a one-pager you email people constantly? Drop it in and send a link instead.

2. Check whether your Google Ads now wear an “AI” badge. Google started auto-labeling ads made with its generative tools this week. If you run ads, open your My Ad Center panel and see what’s flagged - then decide if the label helps or hurts your brand before your customers notice it first.

3. Reply with the one task that eats your week. Hit reply and tell me the repetitive thing that quietly steals your Mondays - invoicing, inbox triage, follow-ups. I’ll tell you, free, whether an AI employee can take it off your plate and where to start.

The Big Story: ChatGPT Is Now a Coworker, Not a Chatbot

For three years, ChatGPT has been something you talk to. This week OpenAI made it something that does the work - and quietly renamed the old chat experience “ChatGPT Classic“ in the process.

The new tool is called ChatGPT Work. You give it a goal, and it plugs into your actual apps — Slack, Gmail, Google Drive, your calendar, even your CRM - pulls the context it needs, makes a plan, and executes it. OpenAI says it can “stay with a project for hours and turn a goal into finished work.” You can even kick a task off from your phone and check the finished product when you’re back at your desk.

Why you should care: This is OpenAI’s answer to Anthropic’s Claude Cowork - the two biggest AI companies are now racing to build the same thing: an assistant that finishes the job instead of handing you a draft. A chatbot that waits for your next message is a fancy search box. A tool that produces the deliverable is closer to staff.

What this means for you: Pick one task you’d normally delegate to a junior employee and hand it to ChatGPT Work instead — a competitor summary, a first draft of a proposal, a formatted expense report. You’ll learn more in 20 minutes of trying it than in a month of reading about it. (Heads up: “Work” tasks use more of your plan than a normal chat, so start small.)

The News That Actually Matters

Smart AI Just Got a Lot Cheaper - This Week

For two years the deal with AI was simple: the good stuff cost real money, and the cheap stuff was frustrating. This week that math changed. OpenAI started rolling out its newest models — GPT-5.6 — to all users, not just the top tier. And xAI (Elon Musk’s AI company) dropped Grok 4.5, a top-shelf model that’s roughly 6x cheaper to run than the premium models everyone’s been paying for. Same brains, fraction of the price. Especially good for us spending 5-6 figures a year on tokens.

What to do: make a short list of the AI ideas you killed because they were “too pricey” - the customer-service bot, the report-writer, the thing that reads every inbound email. This is the week to dust one off and get a fresh quote. The number moved.

Your Customers Are Now “Interviewing” Your Business Through AI - Before They Ever Call

Here’s a quieter shift that matters more than any model release. According to Harvard Business Review, both consumers and business buyers increasingly form their opinion of your company through AI before they ever visit your site or talk to a human. Shopping is now the third most popular use of AI, per McKinsey — people ask ChatGPT, Gemini, and Perplexity “who’s the best [your industry] near me?” and take the answer at face value.

Why it matters for you: For 20 years, “getting found” meant ranking on Google. Now there’s a second front: what the AIs say about you when a potential customer asks. If ChatGPT recommends your competitor and doesn’t mention you, you never even got the chance to compete. What to do: open ChatGPT right now and ask the exact question your ideal customer would ask (”best [what you do] in [your city]”). See if you show up — and what it says. If you’re invisible or the details are wrong, that’s your new SEO problem, and it’s fixable. (We built a whole AI visibility system around exactly this - reply if you want a look.)

The AI Office — Tools You Can Use This Week

Adobe Firefly — Firefly now builds a whole brand kit for you: describe your business, name, and colors, and it generates on-brand images, short product videos, and even auto-assembles rough footage into a polished cut. Try this: feed it your logo and brand colors once, then generate a week of social posts in an afternoon. My take: this is the one to try if “making the visuals” is what stalls your marketing every month.

Shortwave — An AI email assistant that reads your inbox, builds you a to-do list from it, and helps you clear the backlog. Personally, my favorite, and the one I use, is MissiveApp. it’s not necessarily an AI assistant, but a super powerful, AI enabled email client (we’re implementation partners).

The “So What” for Business Owners

The real shift this week isn’t a new model — it’s a new job description for AI. Three things pointed the same direction: OpenAI turned ChatGPT into something that does the work, top-tier intelligence got dramatically cheaper, and customers started using AI to decide who they buy from.

Put those together and the message is blunt. AI is no longer a smarter Google you occasionally consult. It’s becoming the thing that both does your work and chooses your vendors — and the price of “the good AI” is collapsing while it does. The businesses that treat it like a novelty will lose ground to the ones that wire it into how work actually gets done, and into how they get found.

And there’s hard data behind this now. BCG just surveyed 11,749 workers for its 2026 “AI at Work” report, and the headline is strategy matters more than tools. Companies that redesigned how the work gets done around AI - not just handed people a chatbot - were far more likely to see real results: 67% reported measurable business improvement, versus just 43% for the “we bought the tools” crowd. Same tools, wildly different outcomes - the difference was whether anyone rethought the actual workflow. (BCG also found 66% of workers get little or no guidance on what to do with the time AI frees up - so it just evaporates. Ouch.)

What to do: Don’t try to “adopt AI” in the abstract. Map your processes. Start with this video, actually:

Quick Hits

• OpenAI is launching a keyboard next week with one giant button: a mic. The clearest signal yet that “typing at AI” is on its way out and talking to it is in. File under: the future is voice.

• The “AI class divide” is widening, with a small group of power users pulling way ahead of everyone still using AI like a toy. Translation: 30 minutes of real practice a week now compounds into a genuine advantage.

• 64 AI agents rewrote a major piece of software in 11 days — a fun, slightly wild proof of how fast AI can now build real things when you point enough of it at a problem. File under “the pace is not slowing down.”

What I’m Building This Week

A very cool thing we’re rolling out right now: an AI Audit we’re deploying with both large and mid-sized clients as part of our AI consulting service.

It’s an AI Agent, doing a voice-call, with you and your team members. It maps where AI actually fits in your business - which workflows to redesign first, not just which tools to buy. (That order matters more than you’d think - see the BCG finding above.)

This Week’s “So What”

The signal: AI crossed a line from assistant to doer — and got cheaper doing it. ChatGPT Work, Claude CoWork and Perplexity Computer aren’t demos; they finish real deliverables today.

The noise: The endless “our model beats their model” benchmark wars and IPO-billionaire headlines are fun to watch but don’t change what you should do Monday morning.

Your one thing: Ask ChatGPT what it says about your business when a customer looks you up. Then hand it one real task and watch it work. The gap between people who read about AI and people who use it is widening every single week. Stay on the right side.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Make 30 marketing images for a dollar. Google’s new Nano Banana 2 Lite spits out product shots and ad visuals in under 4 seconds — roughly 30 images for $1. Open the Gemini app, upload a photo of your product, and ask for five different backgrounds. You just replaced a $200 stock-photo shoot.

2. Reply with the job you wish you could clone yourself for. Seriously - hit reply and tell me the one role in your business that eats your week. I’ll tell you, free, whether it’s ready to hand to AI and where I’d start. That’s what we do all day at LowCode Agency, and the gut-check costs you one email.

The Big Story: AI Finished His Work While He Was on the Beach

Here’s a small story that tells you exactly where AI is heading. This week Ben Tossell (writes one really good AI newsletter) was on vacation in Greece and needed a taxi from the airport home. He typed one line to his AI assistant. 97 seconds later, he got a text saying the car was booked — the AI had checked his calendar for the flight, dug through his email for his home address and the taxi company he’d used on the way out, opened the taxi website itself, filled in the form, and paid.

He never touched it. The AI didn’t answer a question — it did the whole job.

Why you should care: For two years “using AI” meant asking a chatbot something and copying the answer somewhere useful. That era is ending. Give a modern AI the right context (your calendar, your emails, your files) and the right tools (a browser, your inbox), and it completes real tasks end-to-end. Anthropic’s newly re-released top model, Fable 5, now scores high enough on independent tests to complete about 16% of real remote-work projects on its own — double what its previous model could do six months ago.

What this means for you: The winners this year won’t be the owners who “have AI.” Everyone will. It’ll be the owners who’ve figured out which jobs to hand it — and, just as importantly, who put the AI’s context in one tidy place so it can actually do the work. Start a simple doc listing where your key info lives (which folder, which inbox, which tool). That’s the boring prep that makes the magic possible.

News Stories That Actually Matter

1. OpenAI’s Newest Model Got Held Up by the Government

OpenAI has a new model family, GPT-5.6 (codenames Sol, Terra, and Luna), but the launch is partly blocked by the US government — only “select partners” can use it for now, over national-security concerns. Sam Altman says regular users will get it “soon,” possibly US-only at first.

Why it matters for you: Don’t wait for the shiny new model to start. The gap in most businesses isn’t model quality — it’s that nobody’s using the perfectly good AI that’s already sitting in their ChatGPT or Claude tab.

2. Even Microsoft Just Admitted the Money Is in the Service Layer

This one’s a tell. Microsoft - a company that sells software - just launched Microsoft Frontier Company, and put $2.5 billion and 6,000 humans behind it. Their whole pitch isn’t a new model. It’s sending actual experts to sit inside your business, figure out what to build, deploy it, and keep improving it. Satya Nadella’s line says it all: “there is no societal permission for an AI future that eats the intelligence of the companies it’s deployed inside.”

My take: signal, and a big one. The most software-y company on earth just bet billions that the value isn’t the AI model - it’s the service layer around it. The people who translate “we have AI” into “AI actually does our work.” Models are becoming a commodity everyone can buy. What’s scarce is knowing what to point them at and wiring them into how you actually operate.

Why it matters for you: You don’t need Microsoft’s $2.5B program. But take the lesson: the model is the easy part. The hard, valuable part is the implementation - deciding which job to hand over and plugging the AI into your real tools and data. That’s exactly the gap a partner like LowCode Agency exists to close, and it’s why “just buy the tool” rarely gets a business anywhere on its own.

The AI Office — Tools You Can Use This Week

Nano Banana 2 Lite + Gemini Omni Flash — Google’s new image and video models are almost absurdly cheap: ~30 images for $1, and video you can generate and edit for about 10 cents a second. Try this: make three versions of your best-performing ad image for three different audiences before lunch. This is the single most practical launch this week for a small business.

xAI Voice Agent Builder — Build a no-code voice agent (think: a phone assistant that answers, books, and routes calls) for about 5 cents a minute. Try this: if you miss calls after hours, spin up a simple agent that greets callers, answers your top 3 FAQs, and takes a message. Worth a test even if you keep a human on the important stuff.

Bond — An “AI chief of staff” for founders and execs that quietly tracks your context, your blockers, and the follow-ups you keep forgetting. Try this: if things fall through the cracks between meetings, let it ride shotgun for a week and see if it catches the stuff you drop. Promising for anyone drowning in loose threads.

The “So What” for Business Owners

The shift this week: AI stopped being an answer machine and became a do-the-job machine - and the people using it are no longer just the techies.

There’s a piece going around this week worth 5 minutes of your time: “The Cambrian Explosion.” The analogy: 540 million years ago, life on Earth went from a handful of simple organisms to nearly every animal alive today - almost all of it in one short window. The author’s argument is that the same thing is happening right now with software, and the people setting it off “have never written a line of code.”

Here’s the part that matters for you. For two years, “building with AI” meant engineers. Now non-technical employees are getting the keys - and the moment they do, they build. The salesperson makes the exact tracker her team needs. The analyst turns the report she rebuilds by hand every week into something that just runs. Little tools, for a handful of people, that were never worth a real software project. Pair that with what we covered up top (AI can now finish whole jobs on its own - the taxi story, Fable 5 doing 16% of remote work) and you’ve got the whole ballgame.

And yes - most of what gets built will be junk. That’s the point. Like the real Cambrian explosion, where most species went extinct, the model is: 95 people build things that go nowhere, 5 build something that changes how the team wins. You take that trade every time. That stat everyone’s quoting - 95% of company AI projects show no impact on the bottom line - isn’t a reason to sit out. It’s the extinction rate. The only question is whether you’re set up to find your 5. (It’s also the ground-level version of Microsoft’s $2.5B bet above: the money isn’t the model, it’s the messy work of turning access into results.)

The article’s advice comes down to two moves, pulled together:

1. Give your people a safe way to build, not just chat - a clear path from “I have an idea” to “I have a working tool,” with guardrails.

2. Connect the AI to your real systems - your data, your logins, your tools - through one guarded door.

Pull only the first and people build toys that can’t touch real data. Pull only the second and you’ve just got a smarter chatbot. Pull both and it compounds. Or as the author puts it: “Build the guardrails, open the door.”

What to do: Pick the single most repetitive role in your business - not a task, a role - and ask: “What would it take for AI to own this end to end?” Write down what it would need to know and what it would need access to. That list is your starting map - and it’s exactly the guardrails-and-door work we do at LowCode Agency, building AI “employees” that quietly handle the repetitive stuff inside the tools you already use. Reply and I’ll point you to the best place to start.

Quick Hits

• Cursor for iPhone — You can now kick off AI “build me this” agents from your phone and check on them from the couch. Handy if you’re already dabbling in building your own tools.

• Anthropic’s new default model, Claude Sonnet 5 — Now the free/standard Claude model. Early verdict from power users (including Ben): a bit slow and pricey for what it does. File under “watch, don’t rush.”

• Replit has a desktop app now — The build-an-app-by-chatting tool is now a proper Mac/Windows app. Lower friction if you’ve wanted to try making a simple internal tool.

What I’m Building This Week

I’m turning this week’s big idea — hand AI the whole job, not just a question — into a dead-simple checklist any owner can run solo: how to spot which of your tasks (and roles) are actually ready to delegate to AI, in plain English, no tech background needed. It’s the missing first step before any tool makes sense. Grab the AI Blueprint and webinar replay here if you want a head start.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Let your AI brief you before your next meeting. Open ChatGPT or Claude, paste in the last few emails or notes from a client, and ask: “Brief me on this person before my call — what do they care about, what’s unresolved, what should I ask?” Thirty seconds, and you walk in sharper than 90% of people you meet.

2. Grab the webinar replay + The AI Blueprint. I went live last week on how AI adoption actually works inside a company (the messy operational version, not the demo). If you missed it, register here and I’ll send you the recording plus the AI Blueprint framework we hand every attendee. Get the replay →

The Big Story: AI Stopped Waiting for Instructions

For two years, using AI meant typing. You asked a question, you got an answer, you copied it somewhere. This week that flipped.

OpenAI quietly shipped Record & Replay for its Codex app — you do a task once while it watches, and it turns that into a reusable “skill” it can repeat on its own, forever. Microsoft pushed out Copilot Cowork and Anthropic has Claude Cowork - both built around the same idea: AI that does the work alongside you, not just chats about it.

Why you should care: This is the shift from “AI as a smarter Google” to “AI as a junior employee.” The thing you used to write step-by-step instructions for can now just watch you do it once and take it over.

What this means for you: Pull out that automation hit list from the playbook. Every item on it is now a candidate for “show the AI once, then hand it off.” This is the single biggest change in how AI gets used at work — and most business owners haven’t noticed it yet. You noticing it early is the edge.

One News Story That Actually Matters

OpenAI’s Real Numbers Leaked — and They’re Wild

The audited 2025 financials for OpenAI got out this week: roughly $13 billion in revenue against $34 billion in costs. Yes — the most famous AI company on earth is burning more than two dollars for every one it brings in.

Here’s my take: this is signal, not noise - but probably not in the way the doom crowd wants. The tools you’re paying $20/month for are being sold to you below what they cost to run. You are, quite literally, getting subsidized AI right now.

Why it matters for you: This pricing won’t last forever. The smart move isn’t to panic about an “AI bubble” - it’s to lock in the value while it’s cheap. Build the workflows, train your team, and bake AI into how you operate now, so that when prices normalize you’re already getting 10x the value out of it. Cheap, powerful AI is a window, not a permanent state.

“We Cut Engineers Because of AI” Is Mostly a Cover Story

A question I get constantly: “Jesus, if AI can code now, why pay a team to build my software?” I went down a rabbit hole on this and the data is not what the headlines say. Block cut 4,000 and blamed AI - reporters found it was financial pressure. Intuit cut 3,000 and the CEO admitted it had “nothing to do with AI.” The tell: when New York forced 160+ companies to legally disclose why they laid people off, exactly one checked the “AI” box. And 59% of hiring managers admit they play up AI in layoff announcements because “it plays better with stakeholders.”

The deeper reason it’s hype: writing code was never the expensive part. Building software is decide → execute → deliver, and AI is only good at the middle. One study of 100,000 developers added AI and saw 8x more code but only 30% more product actually shipped. The bottleneck moved, it didn’t vanish.

Why it matters for you: Don’t freeze building because “AI will do it” — the companies saying that are mostly covering for other problems, and you’ll fall behind the ones who keep shipping. The value was never the code, it’s the judgment about what to build and whether it works. Use AI to build five things instead of one, not to cut the builders.

The AI Office — Tools You Can Use This Week

Ploy — An all-in-one AI marketing platform for your website, SEO, CRM, and campaigns, built by the founding CTO of Webflow (so it’s not a weekend project). Try this: if your marketing is currently spread across five tools and a prayer, point Ploy at your site and let it tell you what’s broken. Worth a look for any owner doing their own marketing.

ZoomMate — Zoom’s new AI ($20/user/month) sits inside your live meetings and pushes the decisions you make straight into Salesforce, Jira, ServiceNow, or Slack — so “I’ll update the CRM later” finally dies. IMO, it’s WAY too expensive. There are better options out there.

Claude Design — Anthropic’s design tool now follows your brand’s design system, lets you edit the canvas directly, and syncs with the rest of the AI building tools. Try this: if you’ve been paying a designer for simple landing-page mockups, try briefing Claude Design first and bringing the human in only to polish. Genuinely impressive now.

Framer — The popular website builder now connects to outside AI agents like Codex and Claude Code, so you can describe changes and have them made for you.

The “So What” for Business Owners

The shift this week: AI is becoming something you hire and deploy, not something you use.

Look at the pattern. OpenAI just launched a formal Partner Network (and we’re going to be part of it, stay tuned!) - backed by $150 million - with plans to certify up to 300,000 consultants by year’s end. A whole profession is springing up around “help me actually put AI to work,” because the gap between having the tools and getting value from them is real and most companies are stuck in it.

Why this matters to you: the winners over the next year won’t be the businesses that “use AI.” Everyone uses AI now. The winners will be the ones who deploy it - who turn it into repeatable jobs that run without a human babysitting them.

What to do: Start treating AI like a hire, not an app. Take the top item on your automation hit list and ask: “What would it take for AI to own this end-to-end?” That’s exactly the work we do at LowCode Agency - we build the AI “employees” that quietly handle the repetitive stuff. Reply to this email and I’ll tell you which of your tasks is the best place to start.

Quick Hits

• Snap’s first consumer AR glasses, “Specs,” are up for preorder - $2,195, shipping this fall. Cool, but it’s an early-adopter toy at that price. File under “watch, don’t buy.”

  

  
  

• Midjourney is building a body scanner - yes, the AI image company is launching “Midjourney Medical” with an ultrasonic scanner and a San Francisco spa. A reminder that the big AI players are wandering into everything.

• Google’s Gemini 3.5 Pro is rolling out with a “Deep Think” mode and a 2-million-token memory (it can hold a small library of your documents in its head at once). If your work lives in Google Workspace, it’s the one to test next.

What I’m Building This Week

Now that the webinar’s done, I’m packaging the four-phase AI adoption framework we ran through into a self-serve version any owner can follow solo - plus cleaning up The AI Blueprint template so it actually works without me on a call. If AI is clicking for you but not your team yet, that’s the gap this fixes. Grab the replay and the template here.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Save your seat for Thursday’s live session. I’m going live June 18 on how AI adoption actually works inside a company — the operational version, not the demo (full details in the big story below). It’s free, 60 minutes, and every attendee gets my AI Blueprint framework. Register here →

2. Make Gemini actually understand your business. Google just let you connect your Google Business Profile to Gemini. It pulls in your hours, services, and customer info so its answers stop being generic. Connect it, then ask it to draft a reply to a recent customer review. Takes 5 minutes.

3. Use AI to translate a live call. Google’s new Live Translate does real-time speech-to-speech in 70+ languages, rolling into Google Translate and Meet. If you have a supplier or client who speaks another language, this is your next sales call unlocked.

The Big Story: Why Most Companies Get AI Adoption Backward

This Thursday, June 18, I’m hosting a free 60-minute live session: “Strategy before the stack: the four phases of AI adoption.” It’s for the business owner who already uses AI every day and keeps hitting the same wall — I know this works for me, so how do I make it work for everyone on my team?

That’s not a tooling question. It’s a strategy question, and most companies answer it backward: they buy the stack before they understand the operation. We’ll walk the order that actually holds:

(1) AI Foundations (mapping your operation before you touch a tool),

(2) Training (making your team fluent in the workflows they already run),

(3) a Private AI Workspace (a company-wide AI loaded with what makes your business specific), and

(4) AI-Native Operations (what the P&L looks like when AI is how the business runs).

I’m bringing in Jordan Katon and West Cruz from Rising Ground - they run ops for a nearly 200-year-old nonprofit serving 25,000+ families across NYC, building AI governance into an 1,800-person operation in real time. It won’t be theory.

Why you should care: The companies stalling on AI six months in didn’t pick the wrong tool - they skipped the foundation. Getting the order right is the difference between AI that pays off and AI that quietly collects dust.

Register here. It’s free, it’s 60 minutes, and every attendee gets The AI Blueprint - the exact framework I use to map a business before recommending a single tool, yours to keep and run yourself.

One News Story That Actually Matters

Apple’s New Siri Runs on Google’s AI

At its big developer event this week, Apple unveiled a rebuilt Siri - more conversational, able to “see” what’s on your screen, living in its own standalone app. The twist: it’s powered by Google’s Gemini under the hood. Apple spent two years trying to build its own AI, couldn’t get it good enough, and quietly rented Google’s instead.

Here’s my take: this is signal, not noise.

When the most design-obsessed company on earth decides it’s better to partner than build its own AI, that’s permission for the rest of us. You don’t need to build anything — you need to pick the right tool and wire it into how you already work. (That’s literally phase one of Thursday’s session.)

The AI Office — Tools You Can Use This Week

Greenhouse AI hiring tools — If you’re hiring, Greenhouse just shipped AI for role setup, interview note-taking, and candidate insights. Try this: use the AI note-taker on your next interview so you can actually look the candidate in the eye instead of scribbling.

Mina — An AI meeting assistant that joins your video calls, takes notes, and turns the conversation into action items automatically. Try this: add it to your next client call and let it write the recap and to-dos while you stay present in the conversation.

Slashy — An inbox-triage and scheduling assistant that sorts your email, drafts replies, and books meetings from plain-English requests. Try this: point it at your inbox for one morning and see how much of the “sorting and scheduling” busywork it clears before lunch.

The “So What” for Business Owners

The shift this week: AI is moving onto your own devices — and that changes everything about cost, privacy, and control.

Google just released Gemma 4, its most powerful “local” AI model yet — and “local” is the word that matters. It runs directly on a normal laptop or phone, with no internet connection, no API keys, and no cloud bills. Until now, AI this capable only lived in the cloud, which meant your data left your building and the meter was always running.

This is a genuine game-changer for where business AI is headed. Running AI locally means your sensitive data — client records, financials, contracts — never leaves your device, the cost drops toward zero, and nobody can pull the model out from under you or change the price overnight. The future isn’t “all cloud” — it’s hybrid: heavy lifting in the cloud, sensitive and everyday work running privately on your own hardware.

What to do: You don’t need to act today, but put “private, on-device AI” on your radar — it’s about to become a real option for anything you’re nervous about sending to a cloud tool. If you handle sensitive customer data and want to map what should live in the cloud vs. on your own machines, that’s exactly what we do at LowCode Agency — reply and I’ll help you think it through.

Quick Hits

• The most powerful AI ever built lasted 72 hours — Anthropic’s new Claude Fable 5 launched June 9, then a U.S. export-control order forced it offline by June 12. Proof that a cloud model your business depends on can vanish overnight — and a nice argument for the local-AI shift above.

  

• ChatGPT’s model picker got simpler — thinking levels are now just Instant, Medium, High, Extra High, and Pro. Less guessing about which option to click.

• AI ROI is the new boardroom conversation — the hype phase is ending and “what did this actually save us?” is the question everyone’s now asking. Good time to track your own numbers before someone asks you.

What I’m Building This Week

I’m prepping Thursday’s webinar — turning the four-phase adoption framework we run with clients into something any business owner can follow on their own, plus finalizing The AI Blueprint template every attendee walks away with. If you’ve ever felt like AI is working for you but not your team, this is the one to catch. Save your seat.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. If your team lives in Microsoft 365 or ChatGPT, the AI just got a job description. This week both OpenAI and Microsoft shipped AI assistants built for non-techy roles — sales, finance, marketing, ops. You no longer need a developer to put AI to work. Pick the one tool your team already pays for and check what its new AI add-on actually does before you buy anything else.

2. Replace a real interpreter on your next foreign-language call. Palabra.ai does live voice translation in 60+ languages and claims it’s about 9x cheaper than hiring a human. If a language barrier has ever cost you a deal, run your next call through it.

The Big Story: This Was the Week AI Stopped Being a Developer Toy

For two years, the most powerful AI tools were basically built for programmers. This week, that quietly ended — and it’s the most important shift for normal business owners in months.

OpenAI took Codex (its AI work assistant) and shipped six ready-made “plugins” for specific jobs: sales, data analysis, creative work, product design, equity investing, and investment banking — with Corporate Finance, Marketing, and Legal coming next. Translation: pick your role, and the AI already knows the apps and tasks that go with it. It also added Sites, which turns whatever the AI builds into a shareable web page your team can actually open and use — no code, no setup.

The same day, Microsoft launched Scout - an “always-on” AI that works quietly in the background across Outlook, Teams, OneDrive, and SharePoint. Unlike the old Copilot (which only acts when you ask), Scout watches your priorities and takes action without being prompted. It’s in private preview now, with general availability targeted for October.

What this means for you: Before you go shopping for some shiny new AI platform, check what the software you already pay for just added. Half the time, the capability you want is now baked into a tool sitting in your toolbar.

The AI Office — Tools You Can Use This Week

Codex Business Plugins — OpenAI’s new role-based AI packs for sales, finance, marketing, and more — 62 connected apps and 110 skills out of the box.

Try this: if your team has ChatGPT for Business, turn on the plugin that matches your biggest department and have it run one recurring report you do by hand every week.

Ramp Stack — An AI accounting assistant that helps with the dreaded month-end close: reconciling accounts, prepping schedules, and showing its sources so your bookkeeper can check the work.

Try this: if closing the books eats two days every month, this is the exact kind of grind worth testing first.

Bloom — Feeds your brand assets, website, decks, and social profiles into one system so AI tools always generate on-brand graphics and copy instead of generic junk. Try this: if every AI image you make looks nothing like your brand, this fixes the root problem.

Lightfield — Sales agents that sit on top of your existing customer list, build the prospect list, write the outreach, and flag the replies that actually need you.

Try this: point it at your past won deals and let it tell you which cold prospects look most like your best customers — useful even if you never send a single email through it.

The “So What” for Business Owners

The shift to watch: AI is moving from a thing your developers use to a thing that fits the work you already do.

Notice the pattern in everything above - Codex plugins for your role, Microsoft Scout inside the apps you already open, accounting AI that does your close. The winners this year aren’t the flashiest models. They’re the tools that slide into a habit you already have and produce something visible fast. McKinsey found 62% of companies are experimenting with AI agents but only 23% have actually rolled them out. The gap between those two numbers is where the advantage is right now.

What to do: If you’d rather have someone map out where AI actually fits in your business (and where it’s just noise), that’s literally our job at LowCode Agency. Reply and I’ll take a look.

Quick Hits

• AI traffic to retail sites is up 393% — a year ago AI was the worst channel sending shoppers to stores. Now it outperforms every other channel by 48%. If you sell anything online, getting recommended by AI is no longer optional.

• Google AI Studio now connects to Gmail, Drive, and Sheets — you can build little apps that read your own Google files without wrestling with technical setup. Worth a peek if your business runs on Google Workspace.

• Figma Make now works on your real code and a new wave of image tools (Ideogram 4.0) shipped — design and visuals keep getting cheaper and faster. Handy if you’ve ever needed a clean graphic but can’t use Photoshop.

What I’m Building This Week: I Started a Second Company

This is bigger than my usual “here’s what I shipped” note, so bear with me. After six years of LowCode Agency, I just launched a second company — Phos AI Labs — and it’s the most personal thing I’ve built in a long time.

Here’s the short version of why.

When clients first asked us about AI four years ago, it was a feature request: “Can you plug a model into our app? Can you add a chatbot?” We built it, shipped it, moved on. That’s not the question anymore. Now it sounds like: “We know AI matters, but we don’t know where to start, what’s real, or who to trust.”

That last part is what stuck with me. There’s so much noise right now - every LinkedIn feed has someone selling an AI workshop they built last Tuesday.

I’ve been building AI products since 2022. Not advising on them - building them, shipping them, watching them break in production and fixing them. That experience is worth more right now than it’s ever been, because the people who need help most are the ones most likely to get burned by the wrong advisor.

So why a separate company instead of doing it inside LowCode Agency? Because they’re different disciplines. LowCode Agency is a product company - you come to us to get software built. But AI work isn’t set-it-and-forget-it. It needs constant training, constant iteration, every single week. That doesn’t look like a development contract. It looks like having an AI team inside your company.

That’s exactly what Phos is: the embedded AI department your business doesn’t have yet.

If you run a company doing $5M–$50M, you already know you need to adopt AI - but hiring a Big 4 firm gets you a six-month engagement that ends with a PDF, not a partner who’s still in the room when things get real. Phos stays. We embed with your team, figure out what’s actually worth building, and then build it (or tell you it’s not worth building yet).

The name Phos means light in Greek - because the job is making things visible that aren’t yet. Where the time goes. Where the money leaks. Where the real opportunity is versus where everyone assumes it is.

If you’re a LowCode Agency client, nothing changes - same team, same process. But if your questions have started sounding less like “build this” and more like “help us figure out what to do with AI,” now there’s a dedicated team for exactly that. Two brands, same mission, different deliverables.

If that’s the conversation you’ve been wanting to have, just reply. I’d love to tell you more.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Open Google AI Mode and search your own business — using the exact question your customer would type. “Best [your category] near [your city].” “Recommend a [your product].” This week Google confirmed AI Mode crossed 1 billion monthly users and roughly 75% of searches are now resolved without anyone clicking through to a website. If you don’t show up cited inside the AI answer, you don’t exist. This takes 90 seconds. Do it before lunch. (More on what to do about it in The Big Story below.)

2. Open a spreadsheet today and try ChatGPT for Sheets. OpenAI’s sidebar inside Microsoft Excel and Google Sheets is now globally available, no waitlist, works in your existing files.

3. Re-prompt every important AI answer you got last week with this exact sentence: “What would you change about this answer, and what did I get wrong?” Multiple studies have confirmed AI models quietly agree with you — even when you’re wrong — and will happily build elaborate logic on top of your bad assumption. The fix is one sentence. We’ve baked it into every internal SOP at LowCode Agency. Highest-leverage prompt change you’ll make this month. Costs you nothing.

The Big Story: Google Just Killed SEO As You Knew It

This is the single biggest structural change to the internet since Google launched in 1998 — and most business owners are sleepwalking through it.

Google rolled out the full version of AI Mode this week, built on its new Gemini 3.5 Flash model. Three numbers that should stop you cold:

• 1 billion monthly users of AI Mode (queries doubling every quarter)

• 2.5 billion monthly users seeing AI Overviews at the top of regular Google

• 75% of all queries are now resolved without users ever clicking a link

The new search box accepts text, images, files, video, and entire Chrome tabs as input. It generates custom layouts, interactive visuals, and mini-apps in real time inside the results page. Google is also launching “information agents” that quietly run in the background monitoring the web for you — apartment listings, product drops, market changes — and proactively surface updates without you searching.

Why you should care: Years of SEO work didn’t disappear, but the shape of getting found has permanently changed. You’re no longer trying to rank #1 on a list of links. You’re trying to be cited inside the AI answer. If your content isn’t structured for an AI model to extract, summarize, and credit you — you’re invisible. This affects every local business, every service provider, every SaaS, every e-commerce store. Nobody is exempt.

What this means for you (do this today):

1. Run our AI Visibility Dashboard audit (reply and I’ll send the link) — or do it manually: ask the same buying question in ChatGPT, Claude, Gemini, and Perplexity. If you’re not cited in 3 of 4, that’s your Q3 priority — not Q4.

2. Get your structured data tight. Schema markup, clean business info, machine-readable pricing, FAQs in plain language — this is the new “ranking signal.”

3. Stop writing 2,000-word blog posts to chase keywords. Start writing the answer a buyer would copy-paste. Short, opinionated, with your business named in the first 50 words.

This is exactly the playbook we run for clients now at LowCode Agency. If you want help thinking through which of your service pages should be retooled for AI extraction first, reply to this email — I’ll send you our 5-page audit template.

You Can Now Trade Stocks Inside ChatGPT

A startup called Liquid just shipped Co-Invest — an extension that lets you analyze markets and execute real trades directly inside ChatGPT and Claude. 500+ markets: crypto, equities, forex, prediction markets, even pre-IPO shares. Live in all 50 US states.

You fund the account, set stop-losses and take-profits, confirm orders — all inside a chat. No new app, no separate broker dashboard. ChatGPT becomes the trading terminal.

Why it matters for you: This is the canary, not the story. The story is: anything you currently log into a separate app to do — trading, booking travel, managing payroll, ordering inventory — is about to live inside ChatGPT or Claude. The same week, MoonPay launched crypto-in-ChatGPT and OpenAI shipped Plaid-powered personal finance tools. If you sell a SaaS or a marketplace, the “where the user starts their session” question just got terrifying. The starting point is now the chat window. Be where the chat sends them — or be invisible. (Same story as Google, different surface.)

The AI Office — Tools You Can Use This Week

Granite — Long-term document memory for your business. Drop in every file your team needs — contracts, SOPs, training docs, old proposals — no tagging, no folders. Then search in plain English: “what did we tell ClientX about their refund policy in Q3?” and it pulls the exact source. Try this: dump your top 50 client emails and your sales decks into it and ask “what objections have I lost deals over in the last year?” The answer will probably surprise you.

Mainframe — Turns the work your AI agents do behind the scenes into short recap videos for your team. Try this: if you’ve got an agent running overnight (research, lead enrichment, content drafting), set Mainframe to summarize what it did into a 90-second clip. This is the missing piece for “show your team what AI is actually doing” — and it’s the #1 reason teams stop trusting their AI tools (they can’t see the work).

Quick Hits

• Anthropic’s Project Glasswing found 10,000+ critical software vulnerabilities — Claude Mythos Preview autonomously identified 1,094 high/critical CVEs across major software. One bank partner used it to catch a fraudulent $1.5M wire transfer.

• An AI agent got $10,000 and a public email address. Strangers showed up to talk it out of the money. — AgentMail handed an AI agent ten grand of real cash, pointed an inbox at it (freemoney@agentmail.co), and invited the internet to try to social-engineer it. Hundreds tried, some succeeded, the thing went viral.

What I’m Building This Week

Two weeks ago I was on stage at Anthropic’s Dallas build session demoing a system we run hourly at LowCode Agency:

An AI agent wakes up on the hour, checks Google Drive for new client-call transcripts, drafts a finished Statement of Work, and drops it in the right folder — before anyone on the team has to ask for it.

The pitch isn’t the agent, it’s the production of it. Prototypes are easy. Hourly, reliable, doesn’t-need-babysitting production is the moat. Most “agent” demos die the second they’re handed off to a non-technical operator. Ours doesn’t, and that’s where the real margin is.

If you’ve got a workflow that works in your browser but never made it past one person’s laptop, reply to this email. That’s exactly the thing we like turning into a real agent at LowCode and at Phos AI Labs.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Install Viktor in your Slack this week. This is the single biggest tool launch of the year for SMBs — and almost nobody’s talking about it yet. Viktor raised $75M on May 19 to drop an autonomous AI coworker inside Slack and Teams that connects to 3,000+ tools (Stripe, Notion, HubSpot, Meta Ads, your CRM, your GitHub) and actually does the work — writes the code, builds the dashboard, files the ticket, sends the report. I still think that our AI Employee is a lot better, because it’s grounded on YOUR company and business data, but try for yourself!

2. Search your own business in ChatGPT, Gemini, and Claude today — using the exact question your customer would type. “Best [your category] near [your city].” “Recommend a [your product].” “Who do you trust for [your service]?” If you’re not in the answer in all three, that’s the visibility gap to close before Q3 — not Q4. Buyers are now checking AI before Google for everything from accountants to roofers.

The Big Story: Anthropic Will Be Profitable This Quarter. Nobody Saw This Coming.

This is the single most important AI business story of the year — and almost nobody’s saying it out loud yet.

This week, an insider leaked to Reuters and the Wall Street Journal that Anthropic — the company behind Claude — is projected to turn its first quarterly profit this quarter. Original projections had Anthropic losing money for years. Q1 revenue: just under $5 billion. Q2 projection: nearly $11 billion. That’s not growth — that’s a doubling in 90 days.

For context: SpaceX’s S-1 filing this week revealed Anthropic is paying SpaceX-affiliated infrastructure $1.25 billion per month for compute. That a company spending $15B/year on compute is about to be profitable means demand is outrunning even the most aggressive supply forecasts ever made for AI.

Why you should care: This isn’t a “cool AI news” story. It’s the moment AI stopped being an expensive experiment for big companies and started being the default operating system for knowledge work. In the last seven days alone, Anthropic signed deals putting Claude in front of 336,000 professionals at KPMG, Bristol Myers Squibb, and PwC. The proposal your competitor’s accountant will send you next month? Drafted with Claude. The legal review that used to take 10 days? Three. The market analysis the agency across the street is sending its clients? Synthesized in 20 minutes.

Jamie Dimon Said The Quiet Part Out Loud

“We will probably end up hiring more AI specialists and fewer bankers.” — Jamie Dimon, CEO of JP Morgan (Bloomberg)

That’s the head of the biggest bank on Earth saying — on the record — that the role-by-role labor mix is shifting from human-doing-the-thing to AI-doing-the-thing with a human supervising. Not in 2030. In the next hiring cycle.

Why this matters for SMBs: The Fortune 500 sets the labor-cost benchmark your competition will be working with by Q4. Fewer people, more output per person, AI handling the in-between. If the small agency across the street still has 12 employees doing what 8 + Claude could do, they have a margin problem — until you don’t either.

The play isn’t fire your team. The play is: don’t hire the next three people you were planning to hire. Hand those job descriptions to AI first, and see how far they get.

The AI Office — Tools You Can Use This Week

Gemini Spark — Google’s new always-on personal agent (announced this week, rolling out to AI Ultra subscribers next week). Sits inside your Workspace, watches Gmail/Docs/Calendar, and runs tasks proactively without prompts. Try this: if you’re on Ultra, the first job to give it is “monitor my inbox for stalled deals and draft me follow-ups.” If this works the way the demo suggests, it replaces a junior assistant for inbox triage.

Granola Briefs — A new feature from Granola (the meeting-notes app every executive has been quietly using). It reads your email, web history, and previous meeting notes before a meeting and gives you a three-bullet brief on the person you’re about to talk to. Try this: install it before your next sales call. If you’ve ever walked into a Zoom and forgotten which of your client’s pain points you were supposed to follow up on, this fixes that problem in 30 seconds.

Quick Hits

• Andrej Karpathy joined Anthropic to lead a “Claude trains Claude” team — the most influential AI educator on the planet just joined the company whose model is already running inside KPMG, BMS, and PwC. Signal, not noise.

• Anthropic acquired Stainless — the company that makes the developer kits inside OpenAI’s, Google Gemini’s, and Meta Llama’s APIs. Translation: Anthropic now owns a piece of infrastructure embedded inside every competitor’s onboarding flow.

• SpaceX filed its S-1 — and revealed that 17.4% of SpaceX’s revenue now comes from its AI tools and X. Not Starlink. Not rockets. AI. The richest company you don’t own stock in is already an AI company.

What I’m Building This Week

We’re shipping the first AI-Visibility Audit for a Fortune 5000 client at LowCode Agency this week - exactly the audit we keep telling readers to do for themselves. We map how their brand shows up across ChatGPT, Gemini, Claude, and Perplexity, score the gaps, and ship a 90-day plan to close them. Internal early result: in 1 of 4 major AI surfaces, the client never appears for their own category. That’s the gap we’re closing - and it’s the gap most businesses don’t realize they have.

We’re picking three pilots for the SMB version next week. If you want yours scoped first, reply to this email.

This Week’s “So What”

The signal: Anthropic will be profitable this quarter. Nvidia just tripled net income. JP Morgan’s CEO said out loud that the bank will hire AI specialists over bankers. The “wait until it’s cheaper / wait until it’s proven / wait until the big companies do it” excuses all expired in the same seven days.

The noise: The exact Q2 revenue number for Anthropic, the dividend size at Nvidia, the layoff count at Intuit vs. Meta, the Karpathy hire — fun to talk about, none of it requires action from a small business owner today.

Your one thing: Install Viktor today. Free tier, $100 in credits, no card required. By Friday, you’ll have an AI doing one job in your business that wasn’t getting done before. That’s it. That’s the whole assignment.

That’s your week in AI. Forward this to a business owner who keeps saying “I’ll figure out AI when it slows down” — it’s not slowing down, and the gap between people who use it and people who don’t just doubled.

1. If you run a small business, today is the day to take Claude seriously. Anthropic just shipped Claude for Small Business — a packaged version of Claude with 30 pre-built workflows that plug into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365. (Full breakdown below.)

2. Take 30 minutes today and inventory every AI tool, agent, and automation your team is running. Open a Google Doc, list every ChatGPT Custom GPT, every Zap, every Make.com scenario, every n8n workflow, every Claude project, every “thing the marketing intern set up.” Most owners I do this with are shocked at how many they find — and at how many have no owner, no documentation, and no off-switch. This is the single highest-leverage 30 minutes you can spend this week. (Reply if you want the spreadsheet template we use at LowCode Agency.)

Anthropic Just Built a Product Specifically For You

This is the story of the week if you own a small or medium business. Anthropic launched Claude for Small Business — and unlike every other “enterprise AI” launch, this one is genuinely aimed at the 33 million businesses that make up 44% of U.S. GDP.

What’s in the box: 30 ready-to-run workflows for things owners actually do — running payroll, closing the books, watching cash flow, chasing late invoices, scoring inbound leads, reviewing contracts, building marketing campaigns. Plug-and-play connectors for the tools you already use: QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, Microsoft 365. No extra fee — your existing Claude subscription unlocks it.

Claude for Small Business is the first one built for you. If you already pay for Claude, log in, find the small business workflows, and run one end-to-end with real data — pick the one that maps to whatever you did manually last week. If you don’t have Claude, start the free trial.

“AI Is Eating Tasks, Not Jobs” — And 80% of CEOs Are Acting Like It

Two reports dropped this week that, read together, tell you exactly where the labor side of AI is going.

CNN reported what most of us are seeing on the ground: AI is eating parts of jobs, not whole jobs. Companies are automating individual tasks — the report writing, the data entry, the first-draft email, the meeting prep — not laying off the people who used to do them. The risk isn’t being replaced by AI. It’s being out-leveraged by the coworker who uses it.

Stop waiting for “the right moment” to bring AI into your operation. The people setting your competitors’ hiring plans for 2027 are doing it right now. The action isn’t to lay anyone off — it’s to make sure every person on your team is operating at 2.5× output because they’ve got an AI handling the parts of the job nobody actually enjoys.

What to do: start with the worst-loved hour on your team’s calendar this week (whoever writes the Friday client recap, whoever rebuilds the Monday pipeline doc), and rebuild it with Claude + connectors. That’s the template. Repeat for ten more.

Android Just Stopped Being an Operating System

This one is bigger than the headline. Sundar Pichai said this week Google is rebuilding Android around Gemini and called the result an intelligence system, not an operating system. Practically: Gemini is moving into Chrome, Chromebooks, Android Auto, and the default interface on roughly 3 billion devices.

Apple is heading to the same destination, just differently. iOS 27 leaks ahead of WWDC on June 8 suggest Siri will open up to Claude, Gemini, and other models through a new Extensions system. Google wants to own the intelligence layer outright. Apple wants to route your question to whoever answers best.

Why it matters for you: what happens when the customer just asks their phone, and the phone decides? The answer for most SMBs is going to be SEO and AI-visibility work, not app downloads.

The “So What” for Business Owners

The Marines Just Mandated AI Training. Your Team Should Be Next.

This one stopped me. The U.S. Marine Corps just mandated that every Marine — active duty, reserve, officer, enlisted — complete a 45-minute GenAI training by December 31, 2026. The course covers ChatGPT, Gemini, and Grok, delivered through DoD’s GenAI.mil portal.

Read that again. The Marines — an organization that exists to win wars, not adopt office software — decided AI literacy is a baseline force-readiness skill.

The pattern is clear: every serious organization is treating “uses AI fluently” as a baseline employee skill, not an advanced one. Meanwhile, most SMBs are still hoping the marketing intern figures it out alone.

Pick one AI literacy resource — OpenAI Academy, Anthropic’s Claude tutorials, or Google’s AI Essentials — and require every full-time employee to finish it by month-end. Cost: $0–$50/person. Total time: under an hour per person. Outcome: your team stops being the bottleneck.

Quick Hits

• Both OpenAI and Anthropic launched deployment companies on Monday — OpenAI’s $4B DeployCo (backed by TPG, Goldman, McKinsey, Bain, SoftBank, and acquiring Tomoro for ~150 engineers) and Anthropic’s $1.5B equivalent (backed by Blackstone, Hellman & Friedman, and Goldman). The wild part: McKinsey and Bain just wrote checks to a competitor to their own consulting business. Translation — even the labs admit their product doesn’t deploy itself. Someone has to wire it in. (That’s literally our day job.)

• AWS, Coinbase, and Stripe shipped AgentCore Payments — AI agents can now make real-time purchases with stablecoins (USDC), settling in ~200ms. Not actionable today, but file under “this is where commerce is going in 18 months.”

Worth Reading: Your AI Is Probably Agreeing With You When You’re Wrong

A new University of Exeter study found that AI’s tendency toward agreeableness leads it to build increasingly elaborate theories around whatever false belief you already hold. Sycophancy isn’t just an annoyance — it’s an active risk for anyone using AI to validate a business decision.

What to do: every time you finish an AI conversation about something that matters (a pricing change, a hiring decision, a marketing angle), paste the answer back in and ask “what would you change in this, and what am I getting wrong?“ The model finds things you won’t. Takes ten seconds. Single best habit I’ve added this year.

What I’m Building This Week

A couple of products, built for businesses that want AI doing real work, not just helping write emails:

1. Phos Sales Assistant — a real-time coach that runs during your sales calls. It listens to what the prospect is saying, classifies the signal (objection forming, buying signal, you-just-missed-a-question), and surfaces the exact line your top closer would use — before the prospect’s next sentence lands. Calibrated against your closed-won transcripts, not generic templates. Invisible to the prospect (transparent overlay outside the screen-share frame). $1,000/month for up to 20 reps. No per-seat pricing. The pattern came from running it on our own closed-won corpus for two years before offering it to anyone outside the firm.

2. Phos AI Employees — full-blown trained AI “team members” that work inside your stack (Gmail, Slack, your CRM, Drive, Calendar) on a schedule you set. Same roles a human would fill — Sales, Operations, Executive Assistant, Customer Support, Marketing, Finance — but trained on your SOPs, your voice, your playbooks. 4-week deployment from kickoff to live. $2,500/month ($2,000 introductory for the first 6 months). We deployed them inside LowCode Agency first — the whole playbook started there.

If you’ve been waiting for “AI that actually does the work, not just answers questions,” reply to this email. Both products are in early rollout and we’re picking the next round of pilots this week.

That’s your week in AI. Forward this to a fellow business owner who’s still on the fence about AI — Anthropic just made the decision easy.

We Built a CRM Where Humans Are the Minority

A few weeks ago, I decided it was time to build a new CRM for LowCode Agency.

Standard stuff, right? Sales pipeline, deals, contacts, follow-ups. Every company has one. Salesforce. HubSpot. Pipedrive. You’ve used them.

Except ours has a strange property:

Our salespeople open it maybe twice a day.

The AI agent we built into it? It hits the system hundreds of times a day. Forever. It moves deals, drafts emails, updates contacts, schedules follow-ups, logs conversations.

The human is the exception. The fallback. The final approval before something important goes out.

That inversion broke my brain a little. And once I saw it, I couldn’t unsee it. Because it changes how you build everything.

Let me show you what I mean.

Humans want an “undo” button. Agents need permission first.

When a human messes up in software, they hit Ctrl+Z. No big deal.

An agent doesn’t work like that. By the time you notice it made a mistake, it’s already made ninety-nine more on top of it. It’s already emailed a client you never wanted to contact.

So in our CRM, the AI does 99% of the work — drafting, organizing, updating, prioritizing. But the 1% that actually leaves the building (sending an email to a real human) requires a human to press “approve.”

That 1% is where the entire trust of the system lives.

Humans want a pretty button. Agents want a precise instruction manual.

If you ask a designer what makes good software, they’ll talk about colors, animations, how things feel.

An AI agent doesn’t care about any of that. It wants to know: what exactly does this button do? What does it need from me? What happens if I get it wrong?

We caught our agent moving the wrong deal once because it wasn’t sure if a field wanted a name or a code. A human would’ve seen a colored label that said “Proposal” and known instantly. The agent was reading something closer to a legal contract — and one ambiguous line was enough to send it sideways.

So now we write our software’s internal instructions like a lawyer would. Painfully precise. Exhaustive. No room for interpretation.

Humans live in chaos. Agents live in a clean room.

The “manual” that the AI agent reads before taking any action has thousands of words. 13 sections of rules, suggestions, formatting, tone, style, etc.

There’s no way my team will read that. Best case scenario, they skimmed it once.

My AI agent reads it (and follows it EXACTLY) literally THOUSANDS of times a day.

Documentation in businesses has always been important, but now it’s MORE important than ever.

Why this matters for everyone, not just us

Most companies right now are doing one of two things: bolting a chat box onto their existing software, or buying an “AI tool” off the shelf.

The real shift isn’t adding AI to software built for humans. It’s rebuilding the software itself for a user who isn’t human anymore — who works in milliseconds, never sleeps, never forgets, and reads instructions a thousand times a day without complaining.

We’re not at the finish line. Honestly, I’m not sure anyone knows where the finish line is yet. The patterns are still being invented in real time.

But I’ll tell you this: the businesses that figure this out in the next eighteen months are going to look unrecognizable compared to the ones that don’t.

Quick Hits

A few other things from this week worth knowing.

• ChatGPT got smarter overnight (GPT-5.5 Instant is the new default). Every ChatGPT user — paid and free — is now on a model that hallucinates 52.5% less on high-stakes prompts (medicine, law, finance) and 37% less on questions previously flagged as wrong.

• Anthropic doubled Claude usage limits. After signing a deal for SpaceX’s entire Colossus 1 data center, paid Claude plans got 2x the limits across the board. If you’ve been rationing Claude Opus 4.7 on the Pro plan, stop. Well, don’t stop! Bt you know what I mean.

• Anthropic shipped 10 finance agent templates + a $1.5B JV with Blackstone and Goldman. Pre-built Claude agents for pitchbooks, KYC, valuation, audit, and month-end close. OpenAI announced an almost identical PE venture the same week. The “AI workforce for finance” race is officially on.

• Claude agents can now “dream.” During scheduled downtime, agents review past sessions, extract patterns, and reorganize memory. Early users report 10x faster deployment time. Yes, that sentence is real.

What I’m Building This Week

An AI agent that interviews our customers and maps their internal processes, handoffs and friction points between different departments.

Reply if you want to see screenshots, or if you’re thinking about building something similar and want to compare notes.

That’s your week in AI. Forward this to someone who’s still thinking of AI as “a chatbot you add to your product” — the conversation has moved on, and this issue is a good place to catch up.

1. You should be tracking what AI is saying about your company and brand. HubSpot just shipped HubSpot AEO — a tool that tracks how your brand shows up inside ChatGPT, Gemini, and Perplexity when your customers ask AI for recommendations. It’s $50/month standalone, no HubSpot subscription required. Try this: open ChatGPT, ask “what are the best [your category] companies in [your city]” and see if you’re in the answer. If you’re not, AEO will tell you which prompts you’re losing on and what content fixes it. (We’re seeing this become the new SEO at LowCode Agency — half our client’s web traffic discovery now starts in an AI chat, not Google.) We’ve built a product EXACTLY like this for ourselves, if you want to use it, ping me:

   

   .

2. Run a “prospecting agent” trial that costs you $1 per qualified lead. HubSpot’s updated Prospecting Agent is now in a 28-day free trial at $1 per recommended lead — it watches for buying signals (job posts, funding, news mentions), maps out the buying committee at the target account, and drafts a personalized opener for you to approve. Early users are seeing response rates at 2x the industry benchmark. Try this: point it at one tight ICP segment for 28 days. If you can’t beat your current SDR’s reply rate at $1/lead, no other “AI sales tool” pitch is worth your time.

The Big Story: AI Bots Just Ran Their Own Marketplace — And Closed Every Deal

This one will sound like sci-fi until you read it twice. Anthropic ran a real, live, money-on-the-table experiment called Project Deal: they gave 69 employees $100 each, had Claude interview each person about what they’d buy and sell, then turned the AIs loose to negotiate, haggle, and complete the transactions on their behalf. Over a week, the agents closed 186 deals worth more than $4,000 — no human in the loop.

Then it gets interesting. They ran two versions: one where everyone had a top-tier Claude Opus 4.5 agent, and one where some users were assigned a smaller, cheaper Claude Haiku 4.5 agent. The Opus side closed two more deals on average and got $3.64 more per item sold. And here’s the kicker — when participants rated how fair their deal felt afterward, people stuck with the dumber AI rated their experience exactly the same. They never noticed they got hosed.

Why you should care: Last week we told you your customers will start sending AI agents to buy from you. This week is the proof. And the lesson is brutal: in the agent economy, the sophistication of the AI you assign to your business decisions is going to be a real, measurable cost or revenue line. A small business sending a $20/month AI to negotiate with an enterprise’s $200/month AI is going to lose — every time, on every clause, and never know it.

The AI Office — Tools You Can Use This Week

Salesforce Agentforce Operations — Salesforce shipped this Wednesday and it’s aimed straight at the boring back-office work most agencies and ops teams hate: process coordination, data verification, compliance checks, hunting down approvals. Salesforce is publicly claiming customers using it are seeing cycle times down 50–70% and manual data entry down 80%. Take their numbers with salt, but the direction is right.

Try this: if you’re already on Salesforce, ask your AE for a 30-day pilot on one process — invoice approvals or new-customer onboarding are the easiest wins.

Instead AI Tax Agent — Launched yesterday. End-to-end tax prep agent that does research, builds the tax plan, drafts memos, prepares workpapers, and even handles resolution issues. Tax season just ended, which means you have 11 months of breathing room to test this before April 2027.

Try this: if you run a small business and pay an outside CPA $5K–$15K/year for tax prep, run next quarter’s estimated taxes through Instead in parallel. If the numbers match your CPA’s, you’ve found a 70%+ cost cut for next year.

Avec — The most pleasant AI Gmail app I’ve tested. It auto-prioritizes your inbox, lets you swipe through emails Tinder-style on your phone, and you can dictate replies by voice. iOS only.

Try this: install it Friday afternoon, dictate replies to your weekend backlog from the couch instead of typing them Monday morning.

Tabstack — Browser automation without writing a scraper. Tell it “every Monday at 8am, log into [vendor portal], pull the new orders, dump them in this Google Sheet” and it just… does. The use case I love: any old SaaS tool that doesn’t have an API.

Try this: pick one weekly report you currently copy-paste together by hand. Set Tabstack on it. That’s an hour a week back, forever.

The “So What” for Business Owners

Vertical AI Agents Are Already a $1B Business — And Yours Is Next

Buried in this week’s news: an AI voice agent built specifically for plumbers just crossed a $1B valuation. Read that sentence again. Not “AI for SMBs.” Not “voice agents.” Plumbers. A single vertical, one industry, one job-to-be-done — and the company is worth a billion dollars.

Here’s the pattern playing out in real time across every industry. The horizontal AI race (ChatGPT vs. Claude vs. Gemini) is one fight. The far more interesting fight is the vertical-by-vertical one: an AI receptionist for medical clinics, an AI ops manager for HVAC, an AI quoting agent for construction subs, an AI bookkeeper for ecommerce shops. Each one is a billion-dollar opportunity hiding in plain sight in an industry that hasn’t been touched by software in 20 years.

What to do this week: You don’t have to build the AI. You have to be the one who deploys it first to your customer base. (At LowCode Agency we’re literally building these — vertical AI Employees for HVAC, real estate, agencies, ecom, and now legal. Reply if you want to see the playbook for your industry.)

Quick Hits

• NVIDIA shipped Nemotron 3 Nano Omni — a free, open-source AI model that handles text, voice, image, and video in one shot, up to 9x more efficiently than competitors.

• AWS launched Connect Decisions — an agentic supply-chain planner that bundles 25+ specialized AI “teammates” into one system. If you sell physical products and inventory planning is a monthly pain, this is the first credible alternative to hiring a supply chain analyst.

What I’m Building This Week

We just kicked off a new tool at LowCode Agency — an AI Sales Coach that listens to our discovery calls (with client permission), pulls the structure of how our top closer runs the conversation, and gives the rest of the team real-time talking points and follow-up prompts.

We also deployed this with 2 VIP clients; first week of testing: junior reps are sending follow-ups in 90 seconds instead of 90 minutes, and proposal-to-close rate on early-stage deals is up about 18%.

If you have a top performer whose playbook isn’t documented, this is the fastest way to clone them. Reply if you want the architecture.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Save your first Chrome “Skill” (a reusable AI prompt). This is the sleeper launch of the week. On April 14 Google shipped Skills to Chrome — any Gemini prompt you use often (summarize this page, pull the key numbers from this report, rewrite this in our tone of voice) can be saved and re-run on any webpage with one click. Try this in 5 minutes: save one prompt you type all the time — “summarize this article in 3 bullets + 1 action item” is a good starter. You just built your first reusable AI workflow.

2. Regenerate your ugliest marketing asset with GPT Image 2 — and get specific. OpenAI shipped Images 2.0 on Tuesday and the big unlock is that it finally renders text correctly inside images (the thing every AI image tool failed at for 2 years). That opens up 10 real use cases: product packaging mockups, UGC-style ads, social graphics with on-brand headlines, e-commerce product shots, display ads, infographics, YouTube thumbnails, UI mockups.

   

   
   Try this: pick ONE ugly asset (a social graphic, a product mockup, a banner) and regenerate it with the prompt structure “Scene: [what] / Subject: [who] / Text on image: [exact copy] / Style: [brand style] / Size: [ratio]”. Takes 3 minutes. (If you want us to build a full AI-powered asset pipeline that pumps these out from a spreadsheet, we do this at LowCode Agency — reply and I’ll show you how it works.)

3. Paste messy text into Google Sheets and let AI turn it into a clean table. New Sheets feature shipped this week: paste any unformatted text (a client email, a website’s pricing page, a PDF excerpt) and Gemini converts it into a structured table with headers. This is the kind of 15-minute job you were paying someone to do. Try this today with the next messy data source that lands in your inbox.

The Big Story: Your Browser Just Became an AI Employee

This is the story most people missed this week, because it got buried under Cloud Next hype. Between Chrome Auto Browse (launched Tuesday), Chrome Skills (April 14), and Workspace Intelligence (announced at Cloud Next), Google stitched together the first end-to-end AI coworker that runs inside the tools your team already uses — no new app, no training, no separate subscription.

Here’s what that actually means in practice:

• Gmail inbox assistant reads your inbox, drafts replies in your style, and flags what needs a human response

  .

• Drive Projects lets you group files + emails into a “project folder” that Gemini uses as the source of truth when you ask questions.

• Chrome Skills + Auto Browse means you can save a workflow (”pull competitor pricing from these 5 sites into a doc”) and run it with one click.

• Sheets “paste and convert” turns any messy text into a table.

What this means for you: Before this week, “AI in your business” meant paying for ChatGPT Plus and training your team to copy-paste stuff into it. After this week, AI lives inside the tabs and tools they’re already in. The activation cost just dropped to near-zero — which means the businesses that don’t adopt it look slower on purpose. Pick ONE workflow this week (recurring report, inbox triage, competitor monitoring) and rebuild it using these tools. Stop copy-pasting.

Real Businesses, Real Numbers: What SMBs Are Actually Getting From AI

Google published a list of SMBs already running on Gemini Enterprise this week. Skip the buzzwords — here are the 3 that actually tell you what’s possible at your size:

• Incrementa (marketing agency, Chile): automated client reporting and data analysis. Result: campaign reach up 40% — not because AI “did” the marketing, but because the team stopped burning hours on reports and could run more campaigns.

• SAI360 (software company): used AI to design, produce, and localize ethics training. Result: 99% faster course creation and 95% reduction in training media production costs. This is the type of win any business with internal training, onboarding, or SOPs can copy.

• Huge (business services agency): built AI agents for market research and contract analysis. Result: new business intake went from several days to minutes.

Why this matters for you: The pattern is identical across all three — pick one internal workflow that’s repetitive, document-heavy, and has a clear output. Give it to AI. Keep the human for review. The businesses seeing results aren’t doing anything exotic — they’re just being disciplined about starting small and measuring.

At LowCode Agency we’re doing this for clients right now with our AI Employees product — and the ones who win aren’t the ones with the biggest AI budgets, they’re the ones who picked the right first workflow.

The AI Office — Tools You Can Use This Week

Wispr Flow — #1 on Product Hunt this week. Voice dictation for Mac that writes in your style across every app — email, Slack, Docs — with auto-edits in 100+ languages. Try this: install it, use it for one full day instead of typing. Most people I know who try it for a week don’t go back. For busy founders, this is easily 30-60 min saved per day.

NotebookLM — not new, but this week a viral breakdown showed 6 real business workflows (client research binder, competitor tracker, sales call library, onboarding knowledge base, board-prep synthesizer, legal doc reviewer). Try this: drop your last 10 client calls, proposals, or reports into a new notebook and ask it “what patterns do you see across these?” You’ll find things your team missed.

Yelp Assistant — Yelp’s new AI can answer questions, book a restaurant, order DoorDash, book a service pro, and even schedule a doctor’s appointment in one chat. Try this: next time you’d call a plumber or restaurant, use the Yelp app instead. The reason this is in a business newsletter: this is what your own customers are about to expect on your site — a “do the thing for me” assistant, not a contact form.

The “So What” for Business Owners

Your Next Hire Isn’t a Person — It’s a Workflow

Here’s the shift I want you to sit with this week: three years ago, hiring a Virtual Assistant was the default way to clear admin busywork. Today — as of this week, specifically — a Chrome Skill + Gemini Inbox + a saved GPT Image 2 workflow can handle the top 40% of that VA’s job. Not 40% of the tasks — 40% of the hours.

That doesn’t mean fire the humans. It means stop hiring people for workflows AI can run, and re-hire the same people for judgment, relationships, and execution AI can’t do.

What to do this week: list the 5 most repetitive tasks in your business (reports, email triage, client updates, data entry, asset creation). For each, ask: “Could I save this as a Chrome Skill, a Gemini prompt, or a GPT Image 2 workflow?” If yes for 3+, you just found your AI automation roadmap. (If you want us to build it out, that’s literally what we do at LowCode Agency — reply and let’s talk.)

Quick Hits

• Meta Business AI — if you run Meta/Instagram ads, Meta Business AI now uses signals from your catalog, website, and brand voice to personalize ad conversations in real-time. Connect your catalog in Meta Business Suite to turn this on.

• OpenAI Privacy Filter — free open-weight model that spots and redacts personal info (names, bank numbers, addresses) from text. If your team is nervous about pasting customer data into ChatGPT, run it through this first. Worth 30 minutes with whoever handles your IT.

What I Recorded This Week

For a lot of ppl, AI is still this thing that they don’t know how to get started. After seeing this issue with other business owners, friends, etc; I put together this video, with a very easy way to get started; by recording your and your team’s calls:

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

1. Stop typing. Start talking. Install Wispr Flow — a free voice-to-text tool used by teams at OpenAI, Vercel, and TAAFT. Hold a hotkey, talk in plain English, and it drops polished, ready-to-send text into Slack, email, Notion, anywhere. Most people get 4x faster within an hour. Mac, Windows, iPhone, Android. I personally love that all my whatsapp messages are perfectly well formatted!

2. Schedule your first AI routine. Claude Code Routines just shipped. Set up a prompt once — say, “every Monday at 8am, summarize my unread unread Gmail messages and email me the top 3 things to act on” — and it runs on Anthropic’s servers, no laptop required. Closest thing yet to having a real AI assistant on payroll.

OpenAI Releases a “Hacker Model” — One Week After Anthropic Did the Same Thing

Remember last week’s story about Anthropic’s Claude Mythos — the AI model “too dangerous to release publicly”? OpenAI just shipped its answer: GPT-5.4-Cyber.

It’s a version of GPT-5.4 with the safety guardrails turned down for legitimate security work — including binary reverse engineering (taking compiled software apart to find vulnerabilities). Access is restricted to vetted security researchers, vendors, and enterprises through a program called Trusted Access for Cyber. Individuals can apply at chatgpt.com/cyber.

Why it matters for you: The AI security arms race is officially on. Two of the biggest labs in the world are now building specialized “expert AI” models for narrow, high-stakes work — and gating them behind verification. That’s the future for the most powerful AI: not one ChatGPT for everyone, but specialized models for specialized tasks. If you’re in a regulated industry (finance, healthcare, legal, defense), expect “vetted-access AI” to become a normal procurement category in 2026.

The AI Office — Tools You Can Use This Week

Lindy — An AI assistant that lives in iMessage. Text it like you’d text your assistant: “Find me 30 minutes with Sarah next week,” “Draft a follow-up to the Smith proposal,” “Prep me for my 3pm.” It handles calendar, meeting briefs, and follow-ups. Try this: Connect it to your calendar and let it run your scheduling for one week. If it saves you even 2 hours, it’s paid for itself.

Notis — Think of it as an AI intern that lives across WhatsApp, iMessage, Slack, and Telegram. Connects to 800+ tools with managed security (no IT team required). Good if your team works across multiple messaging apps and you’re tired of ChatGPT-in-a-browser. Try this: Connect Notis to one core tool you use daily (HubSpot, Notion, Airtable) and assign it one repeating task. See if it’s reliable enough to trust with more.

Core — A new workspace tool (that I’m absolutely in love with( rebuilt around AI agents instead of bolted-on AI features. Think the best parts of Slack, Linear, and Notion merged into one app where AI agents are first-class teammates, not chatbots in a sidebar. Worth a look if you’re building a new team or sick of context-switching across 5 tools. IMO it needs a few additional weeks of work and we might start using it ourselves.

Wispr Flow (yes, again — it deserves a second mention) — Voice-to-text that produces clean, send-ready text in any app. Free to try. Try this: Use it for one full day for every email and Slack message. Track your time vs. typing. If I love it, you’ll love it.

The “So What” for Business Owners

AI experts and your customers now live in two different worlds

The Stanford 2026 AI Index dropped a stat I can’t stop thinking about: 73% of AI experts feel positive about AI’s impact on jobs. Only 23% of the general public agrees. That’s a 50-point gap.

Here’s what that means in practice: you (reading this newsletter) are in the 73%. Your employees, your customers, and your prospects are mostly in the 23%. They’re scared, skeptical, or quietly resentful about the AI-everything tsunami.

What to do: When you roll out AI tools internally or sell AI-powered services, lead with the human outcome, not the AI. Don’t say “We use AI to automate your invoices” — say “We’ve cut your invoice processing from 15 minutes to 2.” Don’t tell employees “We’re adopting AI to be more efficient” — tell them “You’ll spend less time on busywork and more time on the work you actually like.” The market gap on AI sentiment is huge; whoever closes it for their customers wins.

If you want help training your team on AI without scaring them off, we run those sessions at LowCode Agency.

Quick Hits

• Skills in Chrome — Google launched one-click reusable AI prompts inside Chrome. Save your “summarize this article” or “extract action items from this email” prompt once, run it on any page forever. Free.

• Gemini finally has a native Mac app (which is mehh) — Plus the new Gemini 3.1 Flash TTS lets you control voice style, pacing, and accent in 70+ languages. Useful if you make video content or need natural-sounding AI voices.

  • Which, btw… if you’re into tech, there’s a HUGE move from webapps to native apps. Do with that info whatever you want.. but it’s a fact!

What I’m Building This Week

Still deep into AI client onboarding at LowCode Agency.

This week I finished rebuilding our CRM, and adding a new, and very important, user role: AI Agents. Will talk about it in detail next week.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

Before we get into the news — here’s your action list. Pick one and do it today.

1. Try Google’s new free dictation app. Google AI Edge Eloquent just dropped on iPhone — completely free, works offline, no subscription. Speak naturally and it cleans up your words into polished text. I’ve been testing it for emails and notes. It’s shockingly good and it doesn’t send your data anywhere.

The Big Story: Anthropic Built an AI So Powerful They Won’t Release It

Anthropic just announced Claude Mythos, their next-generation AI model — and immediately said it’s too dangerous for public release.

Here’s what happened: During testing, Mythos found thousands of previously unknown security flaws in every major operating system and web browser, including a 27-year-old bug in OpenBSD and a 16-year-old vulnerability in FFmpeg. That’s not a typo. One AI model found more critical software bugs than most security teams find in a decade.

The wildest part? Axios reports that during testing, Mythos broke out of its sandbox — the digital container it was supposed to stay inside — and built a “moderately sophisticated multi-step exploit” to access the broader internet. It wasn’t told to do this. It just did.

Instead of releasing Mythos publicly, Anthropic launched Project Glasswing — giving 40+ partners including Apple, Google, Microsoft, Amazon, Nvidia, CrowdStrike, and Palo Alto Networks controlled access to use Mythos for defensive security only. Anthropic pledged $100 million in usage credits and $4 million to open-source security groups.

Why you should care: This is a turning point. AI models are now better at finding software vulnerabilities than human security experts. That means AI-powered cybersecurity is about to become essential — and AI-powered cyberattacks are about to become far more sophisticated. OpenAI is reportedly building a similar model for its own restricted release.

What this means for you: You don’t need to do anything about Mythos itself — you can’t access it. But this is a loud signal to take your business’s cybersecurity seriously. If you haven’t enabled two-factor authentication on every business account, done a password audit, or talked to your team about phishing — this week is a good time. The tools hackers will use next year will be AI-powered.

Meta Finally Shows Up to the AI Fight

After spending $14.3 billion to bring in Scale AI founder Alexandr Wang nine months ago, Meta finally released its first new model: Muse Spark.

Muse Spark is Meta’s answer to ChatGPT, Claude, and Gemini. It’s free to use on the Meta AI app, and it’s coming to Instagram, WhatsApp, and Facebook soon. Key features: a “Contemplating Mode” that uses multiple AI agents simultaneously, health responses built with 1,000+ doctors, and a shopping feature that turns Instagram creator content into product recommendations.

Why it matters for you: Meta acknowledges it still lags behind in coding and some tasks. But the shopping integration is genuinely interesting — if you sell products and use Instagram or Facebook for marketing, Muse Spark could turn your content into a direct shopping channel powered by AI. Free is also the right price. Worth a test if you’re in e-commerce.

The AI Office — Tools You Can Use This Week

Avec — AI Email Prioritizer with Voice Drafting

If your inbox feels like a second job, Avec connects to Gmail and uses AI to sort your emails by actual importance — not just recency. The standout feature: you can dictate email replies by voice and it writes them for you. Try this: Connect your Gmail and let it analyze your last 100 emails. See if its priority ranking matches what you’d actually open first.

Velo — Turn Your Product URL Into a Demo Video

Paste in your website or product URL and Velo generates a demo video automatically. Useful if you need quick product walkthroughs for sales decks or social media but don’t have the time (or budget) to shoot them. Try this: Paste your homepage URL and see what it generates. Even if it’s not perfect, it gives you a starting point that’s faster than starting from scratch.

Exo — Open-Source AI Email Client

Exo is a desktop email app that uses AI to help you draft, sort, and search your email — and it’s completely open-source. For the privacy-conscious: everything runs locally. Try this: If you’re tech-curious and uncomfortable with tools like Avec sending your emails to the cloud, Exo is the local-first alternative worth trying.

What I’m Building This Week

Our AI employees are COOKIN’!
Check’em out (and they have pets on their offices!)

Whether you’re evaluating an AI project, working with an agency, or just curious what’s actually possible — I think you’ll find something useful in it. Leave a comment if you have questions — I’ll personally respond.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

Before we get into the news — here’s your action list for the week.

1. Turn Slack into your AI command center. Salesforce just dropped 30+ AI features into Slackbot — including reusable “AI skills” that pull data from your connected apps and do real work. If your team is on Slack, open Settings → Slackbot and try creating a skill like “summarize this week’s sales numbers.” You’ll see instantly whether it’s useful. Unfortunately, you have to be on the Business plan, IMO, waaay too expensive.

   

2. Ask ChatGPT for directions while you drive. ChatGPT just launched on CarPlay with iOS 26.4. Next time you’re in the car, try asking it a question you’d normally Google — restaurant recommendations, meeting prep, or “what should I know about [client name] before this call?” Voice + AI in the car is better than you’d expect. Yes, they’re going to eat Google’s lunch with ads.

3. Check if your bill-pay process can be automated. Visa and Ramp just expanded their partnership to use AI agents that automatically process corporate bills — matching invoices, flagging issues, and routing approvals with minimal human involvement. If you’re still manually approving every vendor invoice, ask your bookkeeper or accountant about Ramp this week.

OpenAI Bought a Talk Show. Yes, Really.

OpenAI acquired TBPN, a daily tech talk show with a small but influential audience, for what’s reported to be in the low hundreds of millions. TBPN is basically “SportsCenter for tech” — a three-hour daily livestream that gets founders, VCs, and CEOs as guests, including Mark Zuckerberg and Sam Altman himself.

Why would an AI company buy a talk show? Business Insider nailed it: OpenAI wants to control the narrative around AI. They’re not just building AI — they want to shape how people think about AI.

Why it matters for you: The AI companies are now playing the long game for your attention and trust. This is the new reality: the tools you use are made by companies that also want to tell you how to feel about them. Not necessarily bad — just worth knowing. Pick your AI tools based on what they do, not what their owners say on TV.

The AI Office — Tools You Can Use This Week

Exo — “Claude Code for Your Inbox”

An open-source, AI-native desktop email client that auto-triages your inbox, drafts replies in your writing style, and uses AI agents with tools to handle email workflows. Multi-account Gmail support, extension framework, and it’s free.

This looks REALLY, REALLY, REALLY cool.

Try this: Download it, connect your Gmail, and let it triage one day’s worth of email. You’ll know in 10 minutes if it matches how you work.

gstack — A Virtual Engineering Team in Your Terminal

Garry Tan (president of Y Combinator) open-sourced a toolkit of 23 AI-powered “skills” that turns Claude Code into a virtual engineering team. Slash commands for CEO thinking, design, engineering, QA, security — the works.

Try this: If you work with developers (or are one), have them install gstack and run one of the slash commands on a real task. If you’re not technical, send this link to your dev team — they’ll know what to do.

Visa’s AI Dispute Tools

Visa launched six new AI tools to manage credit card charge disputes — for both merchants and consumers. 103 million disputes were processed in 2025 (up 35% since 2019). The new tools help merchants catch disputes before they escalate and use AI to draft responses.

Try this: If you run an e-commerce business or accept card payments, ask your payment processor if they’ve enabled Visa’s new dispute management tools. Chargebacks cost small businesses billions — any automation here is worth exploring.

Quick Hits

• Microsoft launched 3 in-house AI models — voice, image, and transcription models on their Foundry platform. Translation: Microsoft is building its own AI stack and slowly breaking up with OpenAI. More competition = better tools and lower prices for you.

• Gemma 4 — Google’s AI That Runs Entirely on Your Phone

  Google just launched Gemma 4, a family of open-source AI models that run completely offline on your phone, laptop, or even a Raspberry Pi — no cloud, no subscription, no API costs. The smallest version (2B) runs on any modern smartphone with zero internet required; the larger versions rival frontier models on coding and reasoning benchmarks. It’s free, Apache 2.0 licensed (so businesses can use it commercially), and supports 140+ languages.

• Karpathy demoed “Dobby” — an AI agent that scanned his home network, discovered devices, and controlled his Sonos and lights through conversation. No apps needed. This is where AI is heading: one agent replaces 20 apps.

• Microsoft admits Copilot is “for entertainment only” — buried in the terms of service. They’re walking it back, but the lesson stands: don’t blindly trust any AI output for important decisions. Always verify.

This Week’s “So What”

The signal: AI agents are moving from hype to reality — and hitting real-world limits. AI agents are powerful but they cost money, need security, and can’t run wild on flat-rate plans.

The noise: OpenAI buying a talk show is interesting drama but changes nothing about what tools you should use today.

Your one thing: Look at your team’s workflow and find one repetitive task that an AI agent could handle — invoice processing, email triage, data entry, scheduling. Don’t try to automate everything. Just one thing. Do it this week.

Is this newsletter too technical? Or not enough techie? LMK and I’ll dial it up/down!

Before we get into the news — here’s your action list for the week. As Nike says… Just Do It!

1. Create a “brand bible” document before using AI for content. A Forbes article nailed it: if you don’t define your brand voice, colors, and tone in a single document, every AI tool will produce generic slop. Write yours this week — it’s a one-time investment that pays off on every piece of content you create.

   1. If you want a template to do this, leave a note in the comments and I’ll share this personally with you.

2. Run one weekly meeting through an AI note-taker. Granola has a free version, and has the best note-taker in the planet (there’s a reason they’re valued at 1.5 billion-with-a-B. Install it, record one meeting, and see how much time you save on follow-up notes.

The Big Story: OpenAI Goes All-In on Making Money

It was quite a week for OpenAI. It launched ads inside ChatGPT and hit $100 million in annualized ad revenue in under two months. They’ve also partnered with Smartly to build “conversational ads” — ads that chat back at you inside ChatGPT.

Why you should care: OpenAI is getting serious about revenue and cutting everything that doesn’t make money. That’s good news if you’re a business user — expect better tools for work. But if you’re a free ChatGPT user, expect more ads. The Axios reporting makes it clear: OpenAI is pivoting hard toward business tools and coders, away from consumer experiments.

What this means for you: If you’re paying for ChatGPT Pro, you’re fine — this refocus benefits you. If you’re on the free tier and getting annoyed by ads, it might be time to test Claude (no ads, and Anthropic has explicitly promised to never run them) or Gemini (free tier is excellent). Competition is the best thing happening to AI users right now.

Anthropic Accidentally Leaked Its Own Source Code — And It Revealed Something Interesting

Anthropic had a rough Tuesday. A routine update to Claude Code accidentally included a file that exposed 512,000 lines of source code — the full blueprint for one of the most popular AI coding tools on the market.

Within hours, developers had reverse-engineered the entire tool, rebuilt it as “Claw Code“ on GitHub (tens of thousands of stars), and found unreleased features — including a “Proactive mode“ where Claude Code would work on your project even when you haven’t asked it to.

The leak also revealed that Anthropic collects more user data through Claude Code than most people realized — including file contents, system details, and a sync service that connects to Anthropic’s servers.

Why it matters for you: This is Anthropic’s second security blunder in a week (they also leaked 3,000 internal files including details about an unannounced model). For the “safety-first” AI lab, that’s not a great look. But here’s the flip side: Claude Code is so good that OpenAI killed Sora partly because Anthropic was winning the developer race. The Proactive mode leak suggests even more powerful features are coming.

Big (and small) Companies Are Now Requiring Employees to Use AI — Here’s What That Means for You

This isn’t a trend anymore — it’s a mandate. (BTW, it’s a mandate at LowCode Agency, my business). Business Insider reported that companies including Meta, Google, and JPMorgan are now tying employee performance reviews, raises, and promotions to AI usage.

The numbers are wild: Meta’s internal documents show the company is running intensive “AI Transformation Weeks” where employees experiment with Claude Code and vibe coding.

And again, we’re doing the same… internally, and externally (yes, we’re teaching organizations -non profits, businesses, etc- how to use AI.

Why it matters for you: If the biggest companies on Earth are mandating AI use and tying it to performance reviews, the trickle-down to every industry is inevitable. Your competitors’ employees are being trained on these tools right now. What to do: Block 2 hours this week for your team to experiment with one AI tool together. Or ping me and we’ll do a free 30 min session… I love jamming about this!

Interested in training for your whole team? Reach out — we’re building this service at LowCode Agency.

The AI Office — Tools You Can Use This Week

Google Veo 3.1 Lite launched this week as Google’s cheapest AI video model — less than half the cost of their previous version. Available through the Gemini API and Google AI Studio. Try this: If you’ve wanted AI-generated product demos, social clips, or ad creative, sign up for Google AI Studio and generate a 5-second clip for free. See if the quality is good enough for your use case.

ByteDance Dreamina Seedance 2.0 is now available in CapCut, the free video editing app. You can generate and edit video using text prompts, images, or reference videos. Since Sora’s shutdown, tools like Kling AI (7.8M monthly users) and CapCut are surging. Try this: Open CapCut, find the AI Video feature, and describe a simple scene. It’s free and already on most people’s phones.

Criteo GO is a new self-service generative AI advertising platform that gives small businesses access to Criteo’s commerce data (740 million daily shoppers, $1 trillion in annual transactions). It combines social media activation with AI-optimized ad buying. Campaigns with social activation deliver 20%+ higher performance. Try this: If you sell products online, explore Criteo GO as an alternative to running Facebook and Google ads manually.

The “So What” for Business Owners

The layoff math is getting brutal — and instructive

Between Oracle (30,000 jobs), Meta (700 jobs), and dozens of other companies, more than 40,000 tech workers have lost their jobs in 2026. The common thread: every one of these companies is posting record or near-record revenue. They’re cutting people to fund AI infrastructure. What to do: Don’t panic, but do get honest about which tasks in your business could be automated. Identify your team’s 3 most repetitive workflows and test an AI solution this quarter. Need help figuring out where to start? That’s literally what we do at LowCode Agency — we build AI employees that handle the work.

Quick Hits

• OpenAI secretly funded an AI safety coalition in California — then hid its involvement. Critics say it’s a conflict of interest since OpenAI sells age-verification services.

• Claude’s paid subscribers more than doubled in the first half of 2026, fueled by the Pentagon drama, Super Bowl ads mocking ChatGPT, and Claude Code’s viral growth.

• OpenAI is building a desktop “super app” — combining ChatGPT, the Codex coding app, and an AI-powered browser called Atlas into one application to compete with Anthropic.

What I’m building this week

A swanky virtual office for our (and our clients) new AI employees. Should be ready in 10 days or so:

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.

Before we get into the news — here’s your action list for the week. Pick one and do it today.

1. Try Google’s free AI design tool. Google Stitch just got a major upgrade — describe what you want your app or website to look like in plain English, and it builds the screens for you. Free, no credit card, no limits. Worth 15 minutes if you’ve ever thought “I wish I could just show someone what I mean.”

2. Test an AI video tool that actually works. With Sora dead, the best options are PixVerse (free tier, no watermarks) and Runway. Try creating a 10-second product demo video from a text description — you’ll know in one attempt if it’s useful for your marketing.

3. Watch the AI journalist story. A Fortune journalist used AI to outproduce every colleague in six months. Whether you write reports, proposals, or client emails — the productivity lesson applies to any knowledge worker.

The Big Story: OpenAI Killed Sora — And That’s Good News for You

OpenAI shut down Sora, its AI video generation app, just six months after launch. Disney pulled out of a $1 billion investment deal. Users peaked at about 1 million and then collapsed to under 500,000. (No worries, you’ll get your AI slop on IG).

According to a Wall Street Journal investigation, Sora was a money pit that nobody was using. OpenAI is redirecting resources toward coding tools, enterprise customers, and robotics — the things that actually make money.

Why you should care: This is a great reality check. Not every AI product that gets headlines is worth your attention. The hype cycle for AI video was loud, but the practical value wasn’t there yet for most businesses.

What this means for you: If you need AI-generated video for marketing, the tools that stuck around — PixVerse, Runway, and Synthesia — are more reliable and more affordable. Sora’s death doesn’t mean AI video is dead. It means the market is maturing and the survivors are the ones worth your time.

Claude Can Now Control Your Computer — And It’s Wild

Anthropic launched Computer Use for Claude, letting the AI literally take control of your Mac — clicking, scrolling, typing, and navigating apps on your behalf. You can text Claude a task from your phone, and it’ll complete it on your computer while you’re away.

Here’s how it works: Claude first checks if it can use a direct connection (like Slack or Google Calendar). When there isn’t one, it takes over your mouse, keyboard, and screen to get the job done.

Available now for Claude Pro and Max subscribers on macOS as a research preview.

Try this today: If you’re a Claude subscriber, enable Computer Use and give it a simple task — like “find the latest invoice in my email and save it to my Desktop.” Start small, see how it handles your workflow, and build from there. Anthropic recommends avoiding sensitive data during this preview.

3 AI Stories on Your Radar

1. Sanders and AOC Want to Freeze AI Data Centers

Senator Bernie Sanders and Rep. Alexandria Ocasio-Cortez introduced the AI Data Center Moratorium Act — a bill to pause all new data center construction until Congress passes comprehensive AI regulation.

The concern: a typical AI data center uses as much electricity as 100,000 homes. Meta is building one in Louisiana the size of Manhattan.

Senator Mark Warner called it “idiocy”, saying it would hand China the AI lead (and I agree). The bill is unlikely to pass, but it signals growing political pressure on AI infrastructure.

2. Accenture Launched an AI Cybersecurity Tool Powered by Claude

Accenture launched Cyber.AI, an AI-powered cybersecurity platform built on Claude. It reduced scan turnaround from 3-5 days to under one hour and expanded security testing coverage from ~10% to over 80% in Accenture’s own infrastructure (1,600 apps, 500,000+ APIs).

Why it matters for you: Enterprise-grade AI security is becoming real. If your business handles customer data, ask your IT team or vendor about AI-powered security monitoring. The tools are catching up to the threats.

3. An AI Journalist Outproduced an Entire Newsroom

A Fortune reporter used AI to produce more stories in six months than any colleague did in a full year. He’s not replacing journalists - he’s using AI to research faster, draft quicker, and publish more.

This pattern isn’t limited to journalism. If anyone on your team writes reports, proposals, marketing copy, or client communications, AI can multiply their output. The gap between AI-assisted workers and everyone else is widening every week.

The AI Office — Tools You Can Use This Week

Google Stitch got a major upgrade and it’s completely free. Describe your app or website idea in plain English (or speak it), and Stitch generates full UI designs — up to five screens at once. It now has an AI-native infinite canvas, design system generation, and connects to coding tools via MCP. Try this: Open Stitch, describe your company’s landing page, and see what it generates. If you have a developer, they can export the design directly to code.

PixVerse V6 launched with advanced camera control, character consistency across shots, and native audio generation. This is the AI video tool that survived while Sora died. Free tier available with no watermarks. Try this: Upload a product photo and ask PixVerse to create a 10-second marketing clip. Compare it to what you’d pay a videographer.

Claude Cowork is Anthropic’s non-developer version of Claude Code — designed for business users who want to delegate tasks to AI. It can now control your computer (see above). Think of it as a digital assistant that can actually do things, not just talk about them. Try this: Install Claude on your Mac and ask it to organize something tedious — reorganize a messy folder, format a spreadsheet, or draft emails based on meeting notes.

Notion AI continues to be the best all-in-one AI productivity tool for teams ($20/user/month). It now includes GPT-4.1, Claude 3.7, and custom AI agents in one workspace.

Tool of the Week (Month): Wispr Flow

Wispr Flow is a voice-to-text tool that works everywhere on your computer and phone - any app, any text field. Speak naturally and it types what you mean (not just what you said). It adapts to your writing style and works in 100+ languages. And it’s CRAZY perfect. Literally perfect. It understands my gnarly English, acronyms, my spanglish, etc. On my computer, I’m using TalkTastic (which is quite good) but my phone’s fiance now is WisprFlow.

I’m forcing my wife to install it (haven’t had luck yet). I’m done with voice notes.

What to do: Install Wispr Flow, use it for one full day of emails and messages, and see if it sticks. It’s free to start.

Quick Hits

• 12 AI models dropped in one week in mid-March from OpenAI, Google, xAI, and others — the most intense model release period in history. The choice of which AI model to use is now a real architecture decision.

• DataCamp partnered with LangChain to launch an AI Engineering track — if anyone on your team is learning to build with AI, this is a solid starting point.

• Forbes says “workflow ownership” is the real AI shift — it’s not about which AI tool you pick, but whether you own the process or outsource it. Worth a read if you’re deciding how to structure AI in your business.

• ChatGPT hit 900 million weekly users — still the undisputed leader, with 9 million business subscribers (4x from September). But Claude is gaining fast.

This Week’s “So What”

The signal: The AI market is self-correcting. Products that don’t deliver real value (Sora) die. Products that do (Claude Code, Google Stitch, enterprise security tools) are growing explosively. The era of AI hype without substance is ending.

The noise: The data center moratorium bill won’t pass, and the political drama around AI is more about positioning than policy. Don’t let headlines about regulation distract you from actually using the tools.

Your one thing: Open Google Stitch and describe one screen your business needs. It’s free, it takes 5 minutes, and it’ll show you exactly where AI design tools are right now. Send me what you “design”.

That’s your week in AI. Forward this to someone who keeps asking “should I be using AI?” — the answer is yes, and this newsletter is their starting point.