I saw the announcement about Claude watermarks and my first reaction was... really nothing. Scrolled right past it. Had to go back and make myself sit with it for a second, because that felt like the kind of thing I should have a stronger feeling about as an AI creator, right?

The more I’ve sat with it, the more I think that reaction is correct. I believe this whole conversation is pointing at the wrong thing.

Claude has watermarks now. Pangram is flagging content everywhere. Most social media platforms are scanning. And underneath all of it is this ongoing friction about whether using AI to write is some kind of authenticity problem.

But what about all the other tools we use daily?

The tools nobody checks for anymore

When calculators showed up in classrooms there was a real panic. Students aren’t learning real math! They’re outsourcing the thinking! The integrity of the subject is at risk! And then we just moved on, because nobody checks. No one in their adult life gets handed a math problem and asked to solve it longhand. What matters is whether you got the right answer and what you did with that answer.

GPS went through the same thing. You’re going to lose your sense of direction! People won’t be able to navigate without their phones! Spatial thinking is being eroded! And now, nobody asks if you used Google Maps to get there. It’s just how people get places.

And then more recently, there is Grammarly. Grammarly (or even any spellcheck) has been editing your sentences for years, adjusting your tone, rewriting your phrasing. And nobody has said a word about whether your emails are “authentically yours.”

Writing is heading the same way. Maybe not this year. But the direction is pretty clear.

The detection tools work fairly well. The moral panic around them is going to fade. Look at how much the conversation has already shifted in the past month.

What your reader is asking

Before I started creating content, I was a long-time reader/scroller (as most humans are today). When I think about the newsletters I open regularly, the posts I screenshot/save, the stuff that’s changed how I think about a topic, not once do I stop and think “I wonder if AI wrote this?”

What I’m asking instead is: did this give me anything?

Did it shift how I see something? Does this person know my world well enough? Is the idea sharp, or is this just filler with good formatting?

Your readers are not running Pangram on your posts to decide whether to read or not. They’re deciding in the first two paragraphs whether you have something worth their time.

The detection scores are not what matter here. The creators doing well at the moment have a clear point of view, they know their audience, and they say something worth reading. That part isn’t changing.

How to make AI actually useful for your content

Where I see people go wrong is bringing AI in at the wrong part of the process.

It can’t give you your perspective on your industry. It doesn’t know what you’ve noticed after years of doing the work, the thing that seems obvious to you but that nobody else is saying clearly yet. You have to bring that. That’s still wholly yours.

What it’s good at is getting ideas out of your head and into a draft shape faster than staring at a blank page.

Voice notes have been a game-changer for me here. Talk through your idea out loud like you’re explaining it to someone, drop the transcript into Claude, ask it to clean it up into a rough draft. Then rewrite it back into your voice. The thinking stayed yours the whole time. The AI just helped you get it on the page.

Other things that work: giving it your messy bullet points and asking it to find your clearest through-line. Asking “what would someone push back on here?” when you want a gut-check on your argument. Using it for the transitions and the sections you’re less sure about, not the sections where you have the most to say.

The version of this that doesn’t work is using it to generate ideas you don’t already have. That’s where the content feels hollow.

What doesn’t transfer

AI can help you execute. It can’t help you have a perspective worth executing on. It can’t tell you what your specific audience needs to hear right now, or give you the credibility that makes them want to hear it from you specifically.

The reason people subscribe to you is that they’ve come to trust your read on something. They’ve learned that when you show up in their inbox, you usually say something worth their time. That relationship took a while to build and it’s not something you can delegate. Make sure you are respecting that relationship!

and 5 years later I’m making content for the companies I studied in class?!?

welcome to my new content series:

Business as Unusual 🎉

yes my business does executive video, creator content, brand strategy, marketing consulting, videography for the coolest companies in the world.

and yes, this is unusual and couldn’t have existed 5 years ago.

built entirely around my most unique strengths.

so this series teaches you to do the SAME.

building a business that’s unusually YOURS.

all of the learning, behind the scenes, industry tea.

shared!!

so stay tuned!

I was at a lovely dinner party a week ago with some good friends. At some point in the night, I watched someone ask their ChatGPT which contractor to hire for a kitchen remodel.

No comparing three quotes or scrolling through reviews or asking anyone else at the table if they’d heard of the company. The answer came back in about four seconds, with a name and a short reason why, and she read it out loud, nodded, and texted the number before it was time for dessert.

Nobody at the table thought this was strange. It was just how she found a contractor now, the same easy way you’d ask a friend for a recommendation, except the friend was an AI model pulling from who knows where, and the person on the other end of that text had no idea how they’d won a new customer.

I couldn’t stop thinking about that exchange. Somewhere out there, a model had already formed an opinion about which contractor was worth calling, and that opinion traveled from a phone to a decision to a text message faster than most people can finish their dinner.

The customer has changed

For twenty years, marketing operated on one assumption: a human being would eventually land on your website, read your content, and make a call.

Every tactic we built came from that assumption. Headlines written to catch a scrolling thumb. Subject lines written to earn a click. Landing pages built to walk one specific person through one specific decision.

That assumption is getting shakier by the month.

A growing share of the people who used to land on your website now ask a question first, get an answer from a model, and never visit at all. The model already read your content, summarized it, decided if it was worth mentioning, and handed a verdict to your future customer before that person ever typed your name into a search bar.

The audience hasn’t disappeared. The path to them just added a gatekeeper, and that gatekeeper reads differently than a person does.

The speed of this trend

Two years ago, almost nobody treated an AI chatbot as their default way to search. Now it’s close to half of the searches, and the number keeps climbing every quarter.

This didn’t creep in slowly the way most platform shifts do either. Google took years to become the default. This one took a couple of product launches.

And the tools stack on top of each other. Someone asks ChatGPT a question, gets a summary, then checks Perplexity to confirm it, then asks their phone’s built-in assistant a follow-up while they’re still deciding. Three separate models forming three separate opinions about you, all before a human eyeball hits your homepage.

Every model reads you differently

ChatGPT searches the web through Bing’s index. OpenAI’s own engineering team has confirmed this directly. If a page isn’t indexed by Bing, ChatGPT generally can’t find it, no matter how well it ranks on Google.

Gemini is built by Google and runs on Google’s own search index. It’s integrated directly into Google’s ecosystem, which means it behaves a lot like traditional SEO with a conversational layer on top. If your pages already rank well in Google, you have a real shot at being cited here.

Perplexity is the most transparent of the group by design. Every answer comes with visible citations you can click through to check, which is core to how the product positions itself as a research tool rather than a chatbot.

Three different platforms, three different foundations. Visibility on one doesn’t hand you visibility on the others.

Why this matters even outside of tech

This applies to anyone whose customers gather information before they buy, hire, or sign up, which covers nearly every industry that exists.

If someone is choosing which agency to hire, which product to buy, or which service to trust, there’s a real chance the first opinion they hear didn’t come from a person at all. It came from a model that already read about you, formed a summary, and delivered it with total confidence, whether or not that summary was accurate.

And if that summary is wrong, outdated, or missing entirely because your content was never built in a way a model could actually use, you lose that customer without ever finding out they were looking.

What actually changes for marketers?

The skills built over the last two decades, writing that earns attention, design that builds trust, positioning that makes a human choose you, still matter. That work doesn’t disappear.

What gets added on top of it is a new kind of literacy: understanding how these models pull information, what they treat as a credible source, what gets cited, and what gets skipped entirely. It means treating your content less like a pitch aimed at one person and more like a reference document a machine might quote from when someone asks it a question you can’t hear.

It’s an odd thing to sit with. Writing for a reader who doesn’t feel persuasion, doesn’t scroll, doesn’t respond to a clever headline, and just decides in a fraction of a second whether you’re worth bringing up at all.

But that’s who’s reading first now. The brands who figure out how to show up for that reader are going to be the ones still getting found by everyone else

.

Our team reviews hundreds of LinkedIn connection requests every week. Hundreds!! And within about two seconds of landing on someone’s profile, we’ve already formed an impression that is very hard to undo.

First impressions happen fast, they’re based on visual information, and they stick. The good news is that the things that create a weak first impression are almost always fixable, and most of them take less than an afternoon to address.

This is a post about the specific, practical things that signal to another professional that you take your work seriously. A lot of them cost nothing. A few are worth spending money on. All of them are worth knowing.

Virtual first:

Lighting.

If there’s a window behind you on video calls, you look like a silhouette. The light source needs to be in front of your face. A simple ring light fixes this for under $30 and makes a noticeable difference on camera, in headshots taken at home, and in any video content you record. If you’re on camera regularly in any capacity, this is the single highest-return purchase you can make.

Glasses glare.

If you wear glasses and spend time on video calls, glare is probably something you’ve learned to live with. You don’t have to. Tilting your glasses very slightly downward on your nose angles the lenses away from the light source and eliminates most of it. Angling your light so that it’s slightly off-center, instead of pointing center direct at your face also helps. It takes about ten seconds to figure out your specific angle. Anti-reflective coating on your lenses handles it permanently.

Your video background.

A clean wall, a bookshelf with some thought put into it, or a simple neutral background in your video call tool all work well. The question to ask: does this background add to or distract from what I’m trying to communicate?

Your profile photo.

On LinkedIn specifically, your headshot is the first thing people see. It needs to be recent, well-lit, and actually look like you. Smiling helps. Eye contact with the camera helps. A clean or simple background helps.

If you have a decent phone camera, a window with natural light, and someone to take the photo, you can get a perfectly usable headshot for free this week. A professional shoot is worth doing if you’re investing in your personal brand, but it’s not a requirement to get this right.

In-person notes:

Wrinkled clothes.

Steaming your clothes before a meeting, a shoot, a speaking slot, or any moment where you’re going to be seen in person or on camera makes a HUGE difference very quickly. A handheld steamer costs around $20 to $30 and only takes about two minutes to use.

A small number of things you wear a lot.

This is about the math of cost-per-wear. A well-fitting blazer or jacket you reach for before every important meeting or client call will pay for itself quickly if you buy it once and buy it well. Fit matters more than brand. Something well-fitted reads as put-together in a way that an expensive but ill-fitting piece does not.

Professional photos.

A good set of professional photos gives you assets you’ll use for years: a LinkedIn headshot, a speaker bio photo, a website photo, a press photo if you ever need one. If you’re building any kind of personal brand or running a business where clients are evaluating you as a person, having photos that were taken intentionally makes a real difference in how you show up.

You don’t need a full day shoot. A one-hour session with a photographer who knows how to shoot headshots and brand content is enough to walk away with everything you need. OR alternately, a $10 tripod and good phone camera can be just as effective.

The through-line

The through-line across all of this is pretty simple: the details that are easy to overlook are also the ones that are easy to fix. Lighting, background, a recent photo, a steamed shirt before a big meeting. None of it requires a significant investment of time or money. Most of it just requires knowing it matters.

Our team sees hundreds of profiles a week. The people who make a strong first impression are usually just the ones who’ve paid attention to the small stuff.

Maybe you’ve experienced this version of networking before: You go to an event, collect some business cards, connect on LinkedIn the next morning, and then never speak again. Or worse, you connect and immediately get a pitch in your inbox. The whole thing feels SO transactional and makes you want to avoid it entirely.

That version of networking doesn’t work. It never did. It just used to be easier to hide that it wasn’t working because the feedback loop was slower.

Now the feedback loop is very fast, and the bar for what people will tolerate in their inboxes, their DMs, and their feeds has gotten much higher. People have less patience for feeling like a name on a list. And honestly, after years of video calls and then a bumpy entry back into in-person professional life after COVID, a lot of us forgot what it actually felt like to build a real professional relationship from scratch.

This is our attempt to think through what that actually looks like.

The slowness is the point

Every instinct in professional life right now pushes toward speed. Ship faster, post more, respond immediately, scale everything. And there are places where that speed is genuinely useful.

Building trust with another human being is not one of those places.

Trust accumulates through repeated small interactions over time. It’s just how people work. You cannot manufacture familiarity. You cannot batch-process a relationship. One really long conversation does not replace ten shorter ones spread across six months.

I think about this a lot when I hear people describe networking as something they need to do more of. The framing of “doing more networking” already points in the wrong direction. It treats relationship-building like a volume game, and the people who approach it that way tend to have large networks that don’t do much for them.

The people with networks that really work usually have a smaller number of real relationships built over a longer period of time, through more touchpoints than they probably consciously planned for.

Multiple touchpoints

A single cold message, no matter how well-written, is a long shot. Not because people are unfriendly, (although many definitely don’t appreciate the sheer volume of cold pitches nowadays in the DMs), but because one message doesn’t give someone enough information to know whether they want a relationship with you.

What changes that is showing up repeatedly in ways that are genuinely useful or interesting to the other person, before you ever need anything from them.

This might look like commenting thoughtfully on someone’s posts over a period of weeks. Sharing something they made with a real note about why it resonated. Mentioning them or their work in something you’re writing. Showing up at the same events. Responding to their newsletter. None of these things feel like networking in the traditional sense, but they are. They’re building a record of who you are and what you pay attention to, and that record is what makes a cold outreach feel warm.

By the time you actually reach out to someone directly, the goal is for it to feel like a natural next step to both of you rather than a surprise.

Giving before asking

There’s a principle that gets repeated so often it’s almost lost its meaning: give before you ask. Most people interpret this as a transaction with better optics. Do one nice thing, then ask for the thing you wanted.

That’s not what it means, and people can feel the difference.

Giving before asking, taken seriously, means orienting your networking activity around what you can contribute rather than what you can extract. It means asking yourself, before you reach out to someone, what you actually bring to that relationship. It means making introductions when you see a genuine fit, sharing opportunities you won’t take yourself, sending articles with a real thought attached, and being the kind of person in your professional circle who notices things and follows up.

It compounds in a way that transactional networking never does, because people remember how you made them feel and they talk to other people.

And how someone feels about you matters A LOT when networking.

What COVID changed

A lot of professional relationships that used to be maintained by proximity, by running into someone at a conference or grabbing lunch nearby, got interrupted during the pandemic, and never fully recovered.

The relationships that survived that interruption were the ones with enough depth to exist over text or video. The ones that were held together mostly by physical proximity mostly faded.

This tells us something about which kinds of relationships are worth building: ones that have enough substance to them that they could survive a long gap, a format change, a season where neither of you had much bandwidth.

That standard is actually a good filter for where to put your energy. Are you building relationships with enough real texture to last through disruption? Or are you building a network of acquaintances that requires constant in-person maintenance to stay warm?

The AI piece

AI can write a perfectly good cold email. It can research someone before a call, draft a follow-up, generate a list of talking points. It can do a lot of the administrative work of professional communication very well.

What it cannot do is actually care about someone. It cannot have a memory of the conversation you had two years ago that made you think of this person when you saw a job posting. It cannot feel the weight of a favor returned or the warmth of someone who showed up when things were hard.

The parts of networking that AI can do are the parts that didn’t matter that much anyway. The parts that really do matter are still stubbornly human and stubbornly slow. And in a professional environment where more and more surface-level outreach is getting automated, the fact that you are a real person who actually pays attention is more of a differentiator than it used to be.

A few things that actually work

1. Ask better questions and remember the answers. People remember people who remember things about them. Keep notes if you have to.

2. Follow up without a reason. Most follow-ups are tied to an ask. A message that says “I saw this and thought of you” with no agenda attached is rare enough that it tends to land.

3. Make introductions before you’re asked. If you know two people who should know each other, connect them. You don’t need a reason or a personal benefit. The goodwill this generates is real and it circulates.

4. Show up consistently in small ways. A thoughtful comment, a reshare with a real sentence attached, a reply to a newsletter you actually read. These things add up over time in ways that are hard to track but easy to feel.

5. Be patient with the timeline. A relationship that turns into something meaningful, a collaborator, a client, a mentor, a friend, usually takes longer to build than you’d think.

Compounding

A real network works like this: the people in it think of you when something comes up, and they tell other people.

You cannot manufacture that with volume or a good cold email sequence. It comes from a long track record of showing up well, following through, and being the person who gives without keeping score. Do that with a handful of people over enough time and it stops feeling like networking at all. Dare we call it just, relationship building!

In an era where AI is handling more and more surface-level professional communication, being the person who actually pays attention is rarer than it used to be. That rarity compounds!

Before a recruiter decides to reach out, before a brand partner looks you up, before a sales rep writes you a personalized message, an AI agent has already read your profile, summarized it, and scored it. The human comes second now.

Sales teams are running automated workflows that pull your LinkedIn data before anyone picks up a phone. LinkedIn’s own Hiring Assistant is scanning profiles to match candidates to open roles. SDR agents pull your headline, your About section, your recent posts, and hand a summary to a human who then decides if you’re worth a personalized message. Research agents are building full dossiers on people before a single human ever looks at your name.

Most people are still writing their LinkedIn profiles for that one human who might land on the page and spend 30 seconds skimming. That person is still real and still matters! There’s just a lot more happening before they ever arrive.

How agents get to your profile:

A few different methods are in play, and they often work together.

Public scraping. Tools like Clay, Apollo, and dozens of others pull publicly visible profile information into contact databases that sales and recruiting teams feed into their AI workflows. Your name, headline, location, About section, job history, follower count, and articles are all fair game. Anything you’ve made public is accessible, which for most people is almost everything.

LinkedIn’s own AI. LinkedIn has been training models on profile data and posts for years. In late 2025, they extended that to users in the EU, UK, Canada, and other regions. If you’re US-based, your data has been in the mix even longer. The platform’s internal AI, including Hiring Assistant, is building its picture of you from everything you’ve put there.

Third-party enrichment databases. Tools like Apollo, Cognism, and ZoomInfo maintain massive databases of professional data aggregated from LinkedIn and other sources. When an AI agent researches you before an outreach, it’s often pulling from one of these databases rather than visiting your actual profile. Your data is living somewhere you can’t see or update in real time.

MCP integrations. Developers are now building AI agents that connect to LinkedIn via the Model Context Protocol, letting AI assistants query LinkedIn data directly inside a workflow. It’s early, but it’s growing.

All of this means the text on your profile is being read, parsed, and summarized by machines on a regular basis. Whether they surface you as a strong candidate, a relevant prospect, or pass you over entirely depends a lot on what you’ve actually written there.

What agents are extracting:

A person visiting your profile is picking up on a full range of signals: whether your company name rings a bell, whether your photo looks approachable, whether the overall feel matches what they expected.

Agents work differently. They extract structured information: your current title, your industry, your stated expertise, your location, your follower count. They match that against a set of criteria and score you. The vibe doesn’t factor in.

A few things that carry the most weight here:

Your headline. It’s one of the first fields pulled in a scrape. A headline like “Founder | Speaker | Consultant” is hard for an agent to categorize. Something like “COO at [Company] | Content strategy and brand partnerships for B2B companies” gives it exactly what it needs to decide if you’re relevant.

Your About section. This is where agents build a richer picture of you. Use plain, specific language to describe what you do, who you work with, and what you’ve built. Write it like you’re explaining yourself to a smart colleague at a dinner party, not like you’re writing the bio for your own Wikipedia page.

Your skills. LinkedIn’s internal AI uses skills data as part of how it matches people to opportunities. The more specific and accurate your skills are to your actual work, the better the platform’s models understand where to place you.

Your keywords. Think about what a recruiter or potential partner would type into a search bar to find someone doing what you do. Those words need to appear somewhere on your profile, written naturally, in the context of actual sentences.

Do LinkedIn articles and newsletters actually matter for this?

More than most people assume!

All LinkedIn posts can get indexed by Google, especially if they generate strong engagement. But the shelf life is short. Most posts disappear from any real search visibility within 48 to 72 hours. Articles work differently. They have their own permanent URLs, they rank more reliably in search results, and they stay on your profile indefinitely. They also get included in the profile data that scraping tools pull. So when an agent is building a picture of who you are, an article gives it something to actually work with beyond your headline and job history.

The practical difference: a post you wrote six months ago is essentially gone. An article you wrote six months ago is still findable, still scrapeable, and still showing up when someone researches you.

LinkedIn newsletters add one more layer: a visible subscriber count on your profile. That number gets scraped and functions as something an agent can actually quantify when assessing your credibility. A growing subscriber base signals that real people have opted in to hear from you regularly, which carries more weight than follower count alone.

What to do about all this?

Start with your headline. Does it say what you do and who you do it for? If someone read only that one line, would they know whether you were relevant to them?

Rewrite your About section in plain, specific language. Drop the language about being passionate and purpose-driven. Say what you do, who your clients or audience are, what kinds of problems you solve, and what you’ve built. First person is fine. Short paragraphs read better than walls of text.

Add your actual skills. If you work in content strategy, add “content strategy.” If you do brand partnerships, add “brand partnerships.” Specific and accurate beats impressive-sounding every time.

Put something real in your featured section. Agents pull this. Link to your newsletter, your best content, a portfolio piece, a case study. Something that shows the actual work.

Keep posting. Activity signals matter to LinkedIn’s algorithm, and recent post content sometimes gets included in what scraping tools surface. Consistent posting keeps your profile current in a way that matters to both audiences.

Publish a few articles. They don’t need to be frequent. A small archive of long-form content gives agents and humans more to work with when they’re trying to understand what you know.

To end with:

LinkedIn profile advice is notoriously slow to catch up to how the platform actually works. Most of what’s out there was written for a human reader with a short attention span. That reader still exists. There’s just a machine summarizing your profile for them before they ever get there.

The fundamentals of a strong profile haven’t changed much: be specific, be clear, show your actual work. What has changed is the cost of being vague. An agent that can’t figure out what you do in a few seconds of parsing your profile will move on, and the human on the other end may never know you existed.

Last month was Google I/O, the annual conference where the company pretty much tells you what computing is going to look like in two to three years. Most years it’s fine. Incremental updates, some new Pixel features, a lot of AI buzzwords.

This year was WOW.

Two things happened that are worth discussing, especially if you work in marketing, content, or anything adjacent to how people find and consume information online.

The Search Bar

For the first time in more than 25 years, the search bar is getting an overhaul.

The new version builds custom layouts dynamically, assembling things like interactive widgets, data tables, simulation tools, and mini-apps in real time, tailored to whatever you asked. Type in something complex and instead of ten blue links, you might get a functional dashboard, or a tracker, or an actual tool built on the spot for your specific question.

For ongoing tasks like planning a move or managing a project, Search can now build persistent custom experiences, like dashboards you can come back to over time. And those are shareable.

Think about what that means! You’re not getting a page of results anymore. You’re getting something closer to a custom piece of software generated from a single sentence.

The old distinction between software UI and AI output is starting to collapse.

For marketers and creators, the entire concept of “ranking in search” was built on a model where Google returns links and humans click them. Position one CTR has already collapsed from 27% to 11% as AI summaries have taken over the top of the page. Now the page itself is being replaced by a generated experience.

The Laptop

Separately (and somewhat discreetly given everything else announced) Google also killed the Chromebook.

The Googlebook was announced as the Chromebook’s successor, running a new operating system codenamed Aluminium OS, which merges Android and ChromeOS into a single platform. The first devices from HP, Lenovo, Acer, and ASUS are said to be coming this fall.

Google’s Alexander Kuscher summed up the positioning bluntly: “Intelligence is the new spec.” The Googlebook is built around Gemini at the OS level, meaning the AI is woven into the operating system itself. There’s an AI-powered cursor called Magic Pointer that reads context on whatever’s on your screen and can act on it.

Previous Chromebooks ran Android apps through a clunky translation layer that caused crashes and poor performance. Aluminium OS eliminates that entirely because the OS itself is Android. Apps run natively. This is Google’s answer to the MacBook, built from scratch around the assumption that the AI is always there.

So what does this all imply?

Taken together, these two announcements point at something bigger than any individual product.

The interface between humans and computers has basically looked the same since the 1980s. You open a program. You click things. You type. You navigate menus. The search bar was a significant evolution of that model, but it was still fundamentally the same idea: you enter a command, you get output.

What Google is describing now is a different relationship. This is AI that takes initiative, executes multi-step tasks, and works on your behalf without needing you to hold its hand the whole way.

For creators / marketers, the question now is: if the interface between people and information is changing this fundamentally, what does that mean for how you create content, how you get found, and what “showing up” even means in that environment?

There isn’t a clean answer yet. But the people who are thinking about that question now are going to have a real head start on the ones who figure it out later.

Something happened in the first two weeks of May 2026 that felt significant enough to write about here.

Cloudflare, Coinbase, and Upwork all announced major layoffs within days of each other. Thousands of people losing jobs. And in every case, the explanation was some version of the same thing: AI is changing how we work, and we’re restructuring accordingly.

Which is true. But it’s not the whole truth.

Let’s start with the clearest case

Cloudflare is the hardest one to dismiss as spin. The company says internal AI usage surged more than 600% in just three months, with employees across engineering, finance, HR, and marketing now running thousands of AI agent sessions every day. Then they cut 1,100 people, about 20% of their workforce.

But even more interesting: they did it while posting record revenue of $639.8 million for the quarter, a 34% increase year over year. These are not emergency cuts. The company is not struggling.

CEO Matthew Prince was pretty direct about what’s actually happening. He said the company is keeping “builders” and “sellers” while cutting what he called “measurers,” the support and operational roles that exist to manage other people’s work.

This is the clearest articulation any major tech CEO has given about what AI is actually displacing, and it’s not engineers. It’s the layer of people who coordinate, report, verify, and support.

Now for the murkier one

Coinbase cut 700 jobs in May 2026, citing AI-driven efficiency gains. CEO Brian Armstrong framed it as restructuring toward a leaner, AI-native organization.

But Coinbase also reported a $667 million net loss and a 21.6% revenue decline in Q4 2025. So is this an AI story or a revenue story with an AI label on it? Probably both. Their own earnings deck shows engineers are now generating 78% more pull requests year over year. But a company bleeding that much money was also going to cut headcount regardless. AI just gave them a better story to tell.

This is what’s been called “AI-washing,” and it’s happening a lot right now. There’s a real degree of suspicion among business leaders that some of these layoffs are economic cuts dressed up in AI language. Both things can be true at once.

And then there’s Upwork

Upwork CEO Hayden Brown cut roughly 25% of her workforce and wrote in the memo, “Two-pizza teams are dead. AI means smaller, differently resourced teams in product and engineering can make a bigger impact than ever.”

The irony here is almost too on the nose. Upwork is a platform whose entire business model is built on the premise that companies need human freelancers. And they are laying off their own humans while telling clients that AI means you need fewer people.

Their stock dropped 19% on the announcement, partly because revenue grew just 1.4% year over year and forward guidance came in well below analyst expectations. So again, is this an AI transformation or a struggling business making necessary cuts? The real answer (we guess) is probably somewhere in the middle.

So what’s actually happening?

The real signal here isn’t any one single company, but the entire pattern.

The tech sector saw over 102,000 job cuts in the first four months of 2026 alone, with AI cited as the primary driver for two consecutive months. S&P 500 headcount fell in 2025 for the first time since 2016.

But the roles disappearing are specific. The coordinator. The person who manages the report that another person uses to make a decision. The layer of operational middle. What’s growing is just as specific. AI-related freelance work on Upwork grew 60% year over year. Prompt engineering demand is up 240%. Basic writing, data entry, and translation are all declining.

So why does all this matter? (you might ask) Jobs being cut are still jobs being cut, no matter the reason.

The difference matters because it tells you where to put your energy. If the cuts are genuinely structural, the answer is to build skills that sit above what AI can automate. Strategic thinking, judgment, creative direction, client relationships. The stuff that requires a person who actually understands the context.

If the cuts are economic cover, most of those roles come back when the business stabilizes.

Either way, the safest place to be right now is not in the middle layer. It’s at the edges, where you’re either building something or selling something, and doing it in a way that’s hard to replace.

LinkedIn used to be the platform you opened when you were job hunting. Maybe you’d even post something thoughtful once a month, congratulate a few people on work anniversaries, and then close the tab. That version of LinkedIn is gone now.

Full-year revenue hit $17.1 billion in 2024, and LinkedIn recorded $5 billion in a single quarter by Q4 2025. Ad revenue alone is projected to hit $8.2 billion in 2025, with forecasts climbing well past $11 billion by 2027. Here’s where these numbers are coming from, and what it tells us for the future.

A New Market of Buyers

According to LinkedIn’s own 2025 B2B Buyer Report, millennials now make up 73% of all B2B buyers and 44% of final purchasing decision-makers. Gen Z is right behind them, moving into roles with real budget authority. And this generation does not evaluate vendors the way their predecessors did.

Forrester’s 2024 Buyers’ Journey Survey found that 81% of buyers already have a preferred vendor when they first make contact, and 85% have defined their requirements before raising their hand. The sales call is often a formality at that point. The actual decision formed weeks or months earlier, in the content they were consuming. More than 50% of younger buyers rely on social media and their value networks to inform purchasing decisions. They trust people they follow. A company’s banner ad does not factor in.

This is the new environment LinkedIn is building for.

So What is LinkedIn Actually Building?

Thought Leader Ads are the clearest signal of where the platform is headed. The idea is simple: a brand sponsors an individual creator or employee’s post and runs it as paid media. These ads boost engagement by 252% CTR compared to standard formats, and the math on cost per click is hard to argue with.

The reason it works is simple. People engage with people. The algorithm is just reflecting what is already true about human behavior.

The Partnerships tab is what makes Thought Leader Ads scalable. It lives inside Campaign Manager and surfaces any posts that mention your brand, across first, second, and third-degree connections, so you can filter by content type, request promotion rights, and turn organic mentions into sponsored content without hunting for them manually. If someone at a company is already saying great things about your product, you can now amplify that in a few clicks. That used to require a whole workflow to pull off.

Paid 1-on-1 calls are now something thought leaders can offer directly from their profile. Through a Calendly integration on Premium accounts, creators can book paid consultations straight from their LinkedIn profile, turning their audience into actual revenue without sending anyone off-platform. It is a small feature in the grand scheme of things, but it signals something larger: LinkedIn is trying to make staying on the platform financially worth it for the people who drive its engagement.

Alpha Creators Program is still under the radar, but growing rapidly. LinkedIn has started testing a new program with a small number of reputable creators (Lindsay included!), where brands can partner with creators through the LinkedIn platform directly. This erases the communication issue of using either external platforms or sending DMs that get buried.

Other Little Things that show the direction LinkedIn is heading: profile pop-ups when you hover over someone’s name, deeper impressions metrics with percentage breakdowns, a big focus on events, and surely much more to come.

But Why Does This All Matter?

There is a concept in B2B called the dark funnel. It describes everything that shapes a buyer’s opinion before they ever show up in a CRM. 92% of B2B buyers start with at least one vendor already in mind before making contact. The question is how that vendor got there.

More often than not, it’s a post someone saved months ago, a video they watched between meetings, a creator they follow who mentioned a tool in passing. 74% of B2B decision-makers say thought leadership is a more trustworthy way to assess a company than its product marketing materials, and 95% say strong thought leadership makes them more open to sales outreach.

Good content pre-sells the relationship.

LinkedIn generates the highest return on ad spend of any major ad network at 113%, compared to 78% for Google Search and 29% for Meta. The budgets are moving here because the performance data supports it.

The brands doing the best right now are investing in the personal brands of the people inside their companies. An engineer who writes about a deployment that almost broke everything. A sales leader who shares the objections she hears on every single call. Content that sounds like a real person because it is a real person.

The rise of B2B creators is one of the three primary drivers LinkedIn is citing for its own growth, and it’s a whole strategy they are building around.

If you are building a B2B brand or a personal brand in this space, LinkedIn deserves more of your attention than you are probably giving it right now.

A few months ago, I watched an AI agent open my browser, navigate to a website, fill out a form, and complete a task without me touching a single key. I sat there and just... watched it work.

That moment felt so significant. It felt like only months prior that we had just gotten used to chat bots. And something much bigger is already being built and ready to integrate.

From Screens to Streets

Right now, AI agents live mostly on our computers. They book meetings, write code, manage inboxes, browse the web, and move between tools the way a person would. Gartner projects that 40% of enterprise applications will have AI agents built in by the end of 2026, compared to less than 5% just last year. We went from “this is a cool demo” to “this is inside our software” in basically 12 months.

However, less talked about is that the same technology that learned to navigate your screen is now being trained to navigate your world.

Physical AI, which is what researchers call AI embedded in real, moving, physical systems, is no longer a lab project. It is in warehouses. It is on factory floors. It is being funded at a scale that makes what happened with software agents look like a warm-up act. Figure AI raised $675 million this year for physical robots entering commercial deployment. Google’s Gemini Robotics platform, which gives robots the ability to turn visual information into physical actions, is already in the hands of partners including Boston Dynamics. Nvidia’s GR00T foundation model, built specifically for humanoid robots, went from announcement to rapid iteration within a single year.

We are no longer asking whether physical AI will work. We are asking how fast it will scale.

The Marathon Moment

If you want a single image that captures how fast this is moving, look at what happened at the Beijing E-Town Half Marathon over the past two years.

In April 2025, 20 teams entered a humanoid robot race alongside human runners. Six of those robots actually finished. The winner crossed the line in 2 hours, 40 minutes. The robots fell down, lost parts, and needed human handlers nearby the whole time. It looked more like a science fair than a race, but it was historic. First of its kind.

One year later, in April 2026, over 100 teams entered. A robot named Lightning, built by Honor, a Chinese smartphone company that had been making humanoid robots for exactly one year, ran the 21-kilometer course in 50 minutes and 26 seconds. That is faster than the human world record. Lightning operated autonomously, using navigation technology borrowed from self-driving cars, and ran at an average pace that would leave most elite athletes behind.

One hundred minutes faster than the previous year’s winner. In twelve months.

To be fair about what that does and does not mean: Lightning ran a pre-mapped course, had a support crew following it, and crashed into a barricade after crossing the finish line. MIT robotics researcher Rodney Brooks, who co-founded iRobot, pointed out that the robot had no ability to interact safely with real people in real environments. The achievement was real. The general competence it implies is not there yet.

But the trajectory is the story. The gap between what robots could do in April 2025 and what they could do in April 2026 is not incremental. It is a different category of capability entirely.

What Is Actually Being Built Right Now

Here is what is real and deployable today versus what is on the near horizon.

What exists now: Humanoid robots are working in structured industrial settings. Amazon’s fulfillment centers are using Agility Robotics’ Digit for warehouse tasks. BMW and GXO have humanoids in pilot programs on factory floors. These robots work in controlled, predictable environments where the variables are limited and the consequences of an error are manageable. Manufacturing costs for these robots dropped 40% between 2023 and 2024, which is accelerating how fast companies are willing to try them.

What is coming by 2030: The projections point to robots handling semi-structured tasks like tote picking, palletizing, and logistics support at commercial scale. Bain & Company’s research suggests that within the next few years, robots could match human capabilities in intelligence, perception, and physical handling in these kinds of settings, though battery life (currently around two hours per charge) remains a real limiting factor. The global humanoid robot market is projected to hit somewhere between $6 billion and $38 billion by 2030, depending on which research group you ask.

Home robots are a later story. Most serious researchers put meaningful household deployment in the 2030s rather than by the end of this decade. The unstructured nature of home environments, different furniture arrangements, kids, pets, spilled things on the floor, is genuinely harder to solve than a warehouse, where the layout stays the same every day.

The wild card: Bank of America published a research note in March 2026 projecting that by 2060, more people will own a humanoid robot than a car. They are predicting this because of labor math: aging populations, persistent worker shortages, and the economics of automation make humanoid labor increasingly attractive even before robots fully match human ability.

What This Actually Means for Us

I want to be honest about something. When I watch an AI agent complete a task on my screen, there is a version of me that finds it genuinely exciting and a version of me that finds it a little unsettling. I think both of those reactions are reasonable, and I think you are probably allowed to hold them at the same time.

The useful question is not whether this is happening. It clearly is. The useful question is what to pay attention to as it does.

A few things I keep coming back to:

The pace of this is not linear. The robot marathon is a clean example of that. A jump from 2:40 to 0:50 in a single year is a signal that when the foundational pieces fall into place, capability can accelerate in ways that feel sudden, even when they were technically building for years. The same pattern showed up with language models.

The early deployments are in controlled environments for a reason. Warehouses before homes. Factories before streets. There is an emphasis being placed on trust, safety protocols, regulatory frameworks, and what happens when something goes wrong. The physical world has consequences that a software bug does not. This is why the timeline for broad consumer-facing deployment is longer than the hype might suggest, and also why that timeline matters.

The human skills that will hold their value are the ones that require genuine presence and judgment. Reading a room. Making a call with incomplete information. Caring for someone in a way that requires actual empathy. Building trust with another person. These are not things that get automated in the next five years, and probably not in the next ten.

Thinking about that agent working on my screen for the first time, it felt significant because something had shifted without a big announcement. Physical AI is going to feel the same way. One day it will just be part of how the world operates, and we will try to remember what it was like before.

That day is further away than the headlines suggest, and closer than most people are prepared for.

Cannes Lions is June 22-26 and I leave in a few weeks!! If you’ve never been, it’s part industry conference, part networking marathon, part chaos, part “I can’t wait to come back next year.” 15,000 delegates from 90 countries descend on the French Riviera for a week. Mania.

This year feels even bigger than usual. AI is officially on the beach. There’s a dedicated AI and Tech Sandbox at Miramar running all week alongside the main festival, and Cannes added a brand new AI Craft subcategory recognizing work where AI was genuinely essential to the outcome (not just a tool that touched it). The whole conversation has shifted from “should we use AI” to “what did you actually make with it.”

Why This Year Hits Different

Here’s the thing I keep coming back to: the more AI handles the routine parts of work, the more the in-person stuff becomes the actual job.

Global experiential marketing spend hit $128 billion in 2024, finally surpassing its pre-pandemic peak. 79% of 18 to 35-year-olds plan to attend more events this year, and almost a quarter of that same group say they feel lonely. People are showing up in person because the scroll isn’t cutting it anymore, and because the things worth doing increasingly don’t happen on a screen.

Which is exactly why a week like Cannes is so high-leverage right now. A beachside intro that turns into a collaboration six months later. The dinner where someone tells you what they’re actually building. A conversation that shifts how you’re thinking for the next year. You can’t automate any of that, and you can’t replicate it from your laptop.

So showing up matters more than ever. But showing up isn’t the whole game. Going to Cannes without a plan is how you end up exhausted, $5K lighter, and unable to remember what any of it was for. The festival rewards people who walk in with a POV and a real sense of who they want to meet.

That’s the problem I’ve been trying to solve, and it’s what led to the project I’m about to share.

The Unofficial Cannes Guide

I’ve been heads down building a Cannes prep resource with today’s sponsor, Perplexity Computer, and super excited of how it came together.

You can poke around at https://pplx.ai/lindsay-rosenthal-newsletter-1

The build itself was wild. I just described what I wanted via a few prompts with Perplexity Computer, and a full interactive dashboard came back. Things that would’ve taken me and a developer weeks!!

What’s inside:

• An interactive map with every brand activation on the Croisette

• Filters by category (AI, B2B, agencies, and more) so you can plan around what you actually care about

• Probable + confirmed brand list for the week

• Live pulse of what’s being talked about across the event

• And my personal favorite: a prep to-do list to actually walk in ready

If you’re going, bookmark it now and come back to it the week before you leave. If you’re not going, it’s still a pretty good window into what the industry will be obsessing over for the back half of the year.

Better Prep

For something the size of Cannes, 45 minutes of planning on the plane will not cut it. I have been very guilty of this. The week becomes reactive instead of intentional and you leave with great moments but no real throughline.

This year I’m using the guide to get oriented before I take off. A few other shifts in how I’m prepping:

• Identify the three conversations I want to walk away having had. Specific people, or at least specific types of people. Write them down. Do real research on them before I get there so I have something to actually say.

• Know which sessions matter and which are just good room vibes. The AI and Tech Sandbox at Miramar has hands-on sessions all week and I’m prioritizing those over panels where five people more or less agree with each other for 45 minutes.

• Walk in with a POV. Cannes rewards people who actually have something to say. If you’re still forming your opinions on AI, creator commerce, or whatever your lane is, get that work done before you land.

Better Moments

This is where it actually happens. A week like Cannes is a relationship trip first and a content trip second. A few things I’ve learned the hard way:

Ask better questions than everyone else in the room. Most people at these things are in pitch mode by default. If you can ask one genuinely interesting question at the end of a session, or over a drink, you become memorable in a way that no business card can fake.

Create a reason for people to find you. Post from the ground in real time, host a small breakfast, share what’s actually worth attending each day, whatever fits your style. Having something to offer flips the dynamic. You stop chasing introductions and they start coming to you.

And use the tools you already trust in the moment. I’ll have Perplexity open all week for pulling context on a company someone just mentioned, double-checking a stat mid-conversation, or capturing a thought before the next thing swallows it. In an environment that fast, having something to lean on matters.

Better Follow Up

This is what makes or breaks the actual ROI of the trip. You come home with a full connection request list and the best intentions, and then real life rolls back in within 48 hours.

The window is short. The people you met are jet-lagged and back in their inboxes faster than you’d think. I’m building my follow-up system before I leave so it isn’t something I have to figure out while running on four hours of sleep.

How I’m using the dashboard and Perplexity to make it less painful: pulling fast context on anyone I want to reach back out to, so the message references something specific instead of the generic “great to meet you at Cannes.” Organizing what I actually learned across the week by theme, which then becomes the foundation for content when I’m home.

The follow-up is also where you figure out what you actually believe now. A full week of ideas, pitches, and conversations went into your head. Which ones changed your thinking? That synthesis is worth protecting real time for. It’s the part most people skip and it’s genuinely where the value compounds.

See you on the Croisette.
And seriously, go play with the guide before you pack your bag or make your own pre-event prep in Perplexity Computer: https://pplx.ai/lindsay-rosenthal-newsletter-1 .

The job market is brutal right now. We all know it.

Applications are going into black holes. Interviews are getting canceled. Roles that were posted six months ago just disappeared. And on top of all of that, the skills the industry actually wants have shifted so fast that a lot of people are not even sure what to be building toward anymore.

We get asked about this constantly. From people in our community, from followers, from friends trying to figure out their next move. What are you actually looking for? What does it take to get into a company like Cred? What is the industry asking for that a resume does not show?

So here is our honest answer, for anyone who is job searching, hiring, or just trying to figure out where to put their energy right now.

Attribute #1: AI Skills (Not Just the Basics)

We actively look for people who know more about AI than we do. That is not a humble brag. It is just the logic. If we could already do it ourselves, we would not be hiring for it.

We are in this space every single day and we still do not know everything. Not even close. The pace of change is genuinely humbling when you let yourself admit it, and anyone who tells you they have it all figured out is not paying close enough attention. We are not the bar, we want to raise the bar, so we hire above ourselves.

So what does that actually look like in a hire? It is less about knowing the tools and more about knowing how to connect them. People who can look at a workflow and figure out how to make it run on its own. People who understand agents, not just prompts. People who have the technical literacy to set something up and hand it off without needing someone to check on it every hour.

This is not coding. Most of what we are describing does not require writing a single line of code. It is more like systems thinking applied to software. The ability to see a process, understand its moving parts, and build something that removes the human from the repetitive middle of it.

We want people who bring ideas into the room that we have not seen yet. New tools, new structures, new ways of thinking about how this work gets done. If you can teach us something, that is when we get excited. And we think more companies, whether they say it out loud or not, are starting to hire the same way.

Attribute #2: Flexibility (and what that actually means)

This one is worth being direct about, because a lot of companies dance around it.

Content and marketing do not run on a 9 to 5. Brands operate across time zones. Campaigns go live and need adjustments fast. A client has urgent feedback on a Sunday and it actually is time-sensitive. That is just the reality of this industry, and the people who do well in it tend to be people who are not thrown by that.

What flexibility does not mean is being available around the clock with no boundaries. I (Katie) am a mom, and being there for my family is not a side note, it is a real part of how I runs her schedule and life. Lindsay is the same. We both work hard and we both protect the things that matter outside of work.

The version of flexibility the industry is actually asking for is more like: can you manage your own time well enough to show up when it counts? Can you communicate when something is coming up that might affect your availability? Can you be a professional adult about the unpredictable nature of client work without either disappearing or burning out?

We have found that the people who thrive in environments like ours are not the ones who never log off. They are the ones who know themselves well enough to flex without falling apart, and who do not need someone managing their schedule for them.

Attribute #3 Self-Drive

This is the hardest one to screen for in an interview, but it might be the most important.

There is a version of a good employee who does exactly what is asked of them, does it well, and stops there. That person has value. But in a fast-moving industry like this one, the people who stand out are the ones who finish the task and then come back with three ideas about what to do next.

We are always looking around the corner. What is the industry going to need 12 to 18 months from now that it does not know it needs yet? What can we be building toward today so that we are already positioned when everyone else catches up? These are questions we ask ourselves constantly, and we genuinely love being around people who ask them too.

We have a weekly ritual we call the Impossible Ask. One of us brings something to the table that seems genuinely out of reach. Something we want but are not sure we can pull off. Then we work together to figure out if there is a way to make it happen. So far, 45% of those “impossible” things have actually worked, and that number is the reason we keep doing it.

We think a lot of companies, even the ones that would not phrase it this way, are looking for this same quality right now. The willingness to think bigger than the job description. To ask what is coming rather than just what is current. To pitch the idea that sounds a little out there and then follow through on it.

It’s all about having enough genuine curiosity and enough personal investment in the work to keep pushing when the obvious answer would be to stop.

Why We Are Talking About This

We are not actively hiring at the moment. This post is less about finding our next team member and more about a conversation we keep having with people who are navigating this industry right now.

We get asked a lot: what actually matters? What should I be building? What are companies like yours looking for that a resume does not show?

This is our honest answer. Stay curious about AI in a way that goes past surface level. Be someone people can count on when the work gets unpredictable. And bring your own thinking to the table, not just your ability to execute someone else’s.

The industry is shifting fast enough that the people who will do best are not necessarily the ones with the longest resumes. They are the ones who are genuinely paying attention and who are not waiting to be handed a playbook.

One of them spent the last 18 months behind-the-scenes replacing entry-level roles with AI. Customer support, content, data processing, back-office tasks, gone. Replaced almost entirely.

The other one is still debating whether to let employees use ChatGPT.

This is not a hypothetical. This is the reality of 2026, and the gap between these two companies is widening every quarter.

The Numbers Are Not Subtle

The data here is hard to look away from.

According to a 2026 survey by Writer and Workplace Intelligence of 2,400 global executives and employees, 60% of companies are already planning layoffs for workers who won’t adopt AI. Goldman Sachs projects generative AI could replace the equivalent of 25 million full-time jobs in 2026 alone. A Harvard Business School study (March 202), analyzing nearly all U.S. job postings from 2019 to early 2025, found that openings for routine, automation-prone roles fell 13% after ChatGPT launched, while demand for technical and analytical jobs grew 20%.

And it’s not just about the jobs disappearing (although this raises much concern). It’s who is feeling it first. Research from the Federal Reserve Bank of Dallas found that the decline in employment across AI-exposed sectors is falling disproportionately on workers under 25, and not because of layoffs, but because those entry-level jobs simply are not being posted anymore. The people who were supposed to be learning the ropes are finding out there are no more ropes to learn on.

By the end of 2026, Gartner projects that 20% of organizations will use AI to flatten their management structure, cutting more than half of current middle management positions in the process. This is what “AI taking over full roles” actually looks like. Not a robot in a suit and tie. A budget line that disappears.

So Why Are So Many Companies Still Sitting It Out?

Here is where things get genuinely complicated, because the resistance is not coming from the people you would expect.

That same Writer/Workplace Intelligence survey found that 79% of organizations face real challenges adopting AI, a double-digit jump from 2025. This despite 59% of those same companies spending over $1 million annually on AI tools. They are paying for it. They are just not using it well. And according to PwC’s 2026 Global CEO Survey, 56% of CEOs report getting nothing from their AI adoption efforts. Nothing. After millions spent.

Deloitte’s 2026 State of AI in the Enterprise report, drawing on 3,235 global leaders, found that while access to AI tools has expanded 50% year over year, fewer than 60% of employees with access actually use them regularly. Only 25% of organizations have converted 40% or more of their pilots into production systems.

What is actually stopping them? A few real things.

Legacy infrastructure. Nearly 60% of AI leaders say their biggest technical challenge is getting modern AI to connect with systems built long before their employees graduated. You cannot attach a 2026 tool to a 2009 backbone and expect it to work reliably.

Data problems. AI performs relative to the quality of the data it learns from. Many enterprises hold years of siloed, inconsistent records that nobody wants to pay to clean up. So they defer it, and the AI initiatives stall waiting for data that never gets organized.

Compliance and legal exposure. In finance, healthcare, and law, an AI error is not just embarrassing, but legally actionable. The Writer survey found that 67% of executives believe their company has already suffered a data breach due to employees using unsanctioned AI tools. That kind of headline makes legal teams very slow to approve anything.

Culture. This one gets underestimated the most at the moment. The same survey found that 29% of employees admit to actively working against their company’s AI strategy. People are scared, and scared people find workarounds. They ignore the approved tools. They use personal accounts. They do things the old way and hope no one notices.

The result is a kind of organizational paralysis: companies paying to be on the cutting edge while their employees route around it.

The Dichotomy Is Real

On one side: organizations that treated AI as a genuine operational shift. They redesigned workflows rather than piling tools on top of existing ones. They measured adoption instead of just deployment. The productivity gap is now measurable and significant. According to the Writer 2026 survey, AI super-users inside organizations save nearly nine hours per week, 4.5 times more than employees who use AI minimally. Those same super-users were three times more likely to have received both a promotion and a pay raise in the past year.

On the other side: organizations stuck in what researchers call pilot purgatory. They buy licenses. They run a successful demo. They call it a strategy. Then the pilot does not scale, the ROI does not appear, and someone shelves the project until the board brings it up again. According to the Writer survey, 75% of executives admit their company’s AI strategy is more for show than actual internal guidance. That number is not surprising.

The gap compounds. McKinsey data from 2025 found that industries with high AI exposure saw revenue per employee grow by 27% between 2018 and 2024. Industries with low AI exposure grew by 9%.

Who Comes Out Ahead?

Not automatically the fastest mover. Speed without structure is how you end up with 95% pilot failure rates.

The organizations that will look back on this period as a win share a few things in common. They treated adoption as a people and culture problem at least as much as a technology problem. They brought employees in early, gave them real training rather than a single onboarding session, and built clear governance around which tools were approved and what data could go where. They connected AI spending to specific, measurable outcomes rather than general productivity hope.

They also went deep on specific use cases instead of spreading thin across every department at once. Engineering and product teams moved fast because the tooling in those areas is mature and results are easy to track. Legal and finance moved slower because the tools are less proven and the cost of an error is higher. The companies doing well understood that difference and did not try to force a uniform rollout across functions with completely different risk profiles.

The organizations that will struggle are the ones treating this as a checkbox. Buying a tool to say they have AI. Announcing a strategy in an all-hands and then leaving the actual work unchanged. Expecting productivity gains without rethinking the processes that govern how work gets done.

What This Means for the Rest of Us

If you are not running a company, you are still living inside this divide.

The entry-level pipeline is being cut. The Federal Reserve Dallas research makes this specific: young workers in AI-exposed fields are not being laid off, they are just not being hired in the first place. The traditional career ladder is losing its lower rungs before people get a chance to climb them. That is a real, slow-moving problem with consequences that will take years to fully show up in the data.

But a few things hold up regardless of how fast the technology moves.

Skills that sit at the intersection of judgment and human communication are holding their value. Figuring out what a client actually needs, reading a situation when the data is ambiguous, making a call when there is no clear right answer. AI is not good at these things yet, and for many roles, these are the skills that actually matter at a senior level.

Understanding how AI works, even at a non-technical level, is quickly becoming a basic expectation rather than a standout quality. Not because every job requires building models, but because every job will involve working alongside tools that use them. The people who understand those tools well enough to question their outputs, catch errors, and direct their use effectively will be worth more than those who either blindly trust them or refuse to touch them.

And physical and human-facing work continues to hold a floor. Healthcare, skilled trades, teaching, care work. These are holding up both because AI cannot replicate physical presence or real human judgment, and because demand keeps rising anyway.

The Honest Bottom Line

Companies going deep on AI are making a real bet, and many will win it. Companies holding back are also making a bet, and some have legitimate reasons for the caution. The problem is that caution has a cost that compounds over time, and most of the leaders pumping the brakes do not have a clear plan for when they will stop.

For the rest of us watching this play out, the job is not to pick a side or predict a winner. It is to stay clear-eyed about what is changing, build skills that are not going to expire in 18 months, and refuse to confuse buying a tool with having a strategy.

The gap is real and it is growing. The question is less about which side you are on and more about whether you are actually doing anything about it.

Open almost any article, scroll almost any feed, and the narrative is the same.
Jobs disappearing. Security risks. Hallucinations. Bias.
The slow creep of automation swallowing up everything humans used to be valued for.

And look, those conversations are important. They are worth having. But somewhere in all of that noise, we have almost completely lost sight of something that we think deserves a lot more airtime.

AI is not just doing what humans do, faster. It is raising the ceiling of what is possible at all. And in some of the most important domains in human history, it is cracking open doors that no amount of human effort, funding, or brilliance ever could have reached on its own.

That is what we got into on this week’s episode. And we left genuinely fired up.

The Cancer Breakthrough

Researchers at the Technical University of Denmark used an AI program to train cancer patients’ own immune cells to specifically identify and attack their cancer cells. The process took a matter of weeks.

In a traditional lab setting, that approach would not have taken ten years. It likely would not have been on the horizon at all.

And that is the framing shift that we keep coming back to. This is not a story about doing the same thing faster. It is a story about a door that was not open before, and AI opening it. That is a fundamentally different kind of progress.

Part of why cancer has been so historically hard to crack is that it is not one problem with one solution. Every tumor is distinct. Every patient’s biology is different. The variables are almost incomprehensibly complex. But that complexity is exactly where AI thrives. It is built for this. And we are only just beginning to see what that means.

The Domino Effect of Solved Problems

Cancer is not the only example. Not even close.

AlphaFold, developed by Google DeepMind, solved protein folding. This was a problem that some of the sharpest biological minds on the planet had been working on for over 50 years. It was not that they were not smart enough or not trying hard enough. The problem was simply at a scale and complexity that human cognition could not fully process.

AI solved it. And in doing so, it unlocked a cascade of downstream possibilities, because protein folding sits at the foundation of so much of what we understand about disease, drug development, and human biology.

The same pattern is emerging in climate science. Climate systems involve so many interdependent variables that comprehensive modelling has always been constrained by what was computationally possible. AI is beginning to break that constraint, running simulations at a resolution and timescale that simply were not available before.

What these breakthroughs have in common is that the bottleneck was never really human intelligence or effort or funding. The bottleneck was computational and cognitive scale. And AI removes that bottleneck in a way that nothing else ever has.

The Problems We Have Been Waiting Our Whole Lives to Solve

Once you start thinking this way, the list gets long quickly.

Antibiotic resistance. Alzheimer’s. Rare diseases that have never had enough cases to generate meaningful research. Mental health treatments that still rely on methods developed decades ago. The interconnected crises of food security, water scarcity, and ecological collapse.

These are not small problems. They are the problems that have defined human suffering for generations. And for the first time, there is a genuine argument that we are approaching the tools needed to meaningfully tackle them within our lifetimes.

That is not nothing. That is everything.

The Honest Counterpoint

We would not be doing this topic justice without acknowledging the other side.

These models are not perfect. And when AI operates in high stakes domains like medicine, imperfection has consequences. A hallucination in a content tool produces a bad paragraph. A hallucination in a diagnostic or treatment system could cost someone their life.

That is not a reason to stop. But it is a reason to build carefully, to maintain human judgment in the loop, and to make sure the guardrails keep pace with the capabilities.

The future we are most excited about is not one where AI replaces the doctor or the researcher. It is one where AI handles the computational scale and pattern recognition it is genuinely better at, and humans bring in the judgment, the context, the ethical weight of the high stakes calls that genuinely need a person in the room.

That partnership, built responsibly, is where the magic is.

So here is the question we will leave you with:

For the first time in history, the limits of human progress might not be intelligence, funding, or effort. If that constraint is gone, what are we going to do with that?

Will it take our jobs? How many? How fast?

It is a fair question. But we think it is the wrong one.

The more interesting question, is this: what does talent even mean in a world where AI can do most of what we used to call skilled work, faster and more consistently than any human?

That is what we dug into. And the answers were more unsettling, and more freeing, than we expected.

The Jobs Are Already Changing

Let’s start with the data, because it is hard to ignore.

Entry level job postings are down 29% year over year. And this is not a coincidence. Entry level roles have historically existed to handle the smaller, more defined tasks that keep larger projects moving. Research, drafting, basic coding, data entry, coordination. These are exactly the tasks AI is best at. Fast, accurate, tireless.

For anyone graduating right now, or trying to figure out whether college is even worth it, the traditional career ladder that generations before them climbed is not just wobbling. It has fundamentally changed shape.

Comprehension Debt: Borrowing Against Your Future Self

Anthropic ran a study with 52 software engineers. Half used AI to complete a task, half did not. When both groups were given a follow up quiz on the material, the AI users scored 17% lower.

They got the task done. But they did not actually learn anything from doing it.

We defined this as: comprehension debt. The idea that every time you use AI to do something you do not fully understand, you are borrowing against your future self. You look capable today. But six months from now, a year from now, you are further behind than you would have been if you had just done the hard thing.

And as anyone who has ever carried debt knows, it compounds.

We are seeing this in content creation too. People can appear to be experts on almost any topic. Post confidently, sound authoritative. But push a little on the actual depth of understanding and sometimes it is just not there. The output exists. The knowledge does not.

Your Brain on AI

It gets more interesting. And a little more alarming.

MIT Media Lab researchers used ECG scans to study what happens in the brain when people use AI for cognitively demanding tasks. The result? The parts of the brain that typically fire during complex thinking quite literally switch off when AI is doing the work instead.

We have probably all felt a version of this. A task lands in your inbox that would have required real focus a couple of years ago, and now your first instinct is to hand it off. The brain has been trained to offload. And if we never exercise those cognitive muscles, the question of what happens to them over time is one worth taking seriously.

What Employers Are Actually Looking For

Unfortunately, there is a real disconnect happening right now in the job market.

Gen Z, understandably, has grown up prioritizing technical skills. AI fluency, tools, systems. That makes complete sense given the world they have entered.

But almost every workforce report coming out right now is saying something different. The skills that organizations are struggling to find are not technical.
They are human. Judgment. Creativity. Taste.
The ability to read a room, make a call, communicate clearly, lead through ambiguity.

And Gartner is predicting that by the end of 2026, a significant number of organizations will require AI-free skills assessments during hiring, specifically to understand whether candidates can actually think clearly and independently when the tools are not available.

Because the tools do go down. Wifi has outages. Bugs happen. And if your entire capability is contingent on AI being available and working, that is a real liability.

The Reframe That Changed It For Us

AI does not replace deep human expertise. It multiplies it.

If you bring genuine knowledge, real judgment, and developed interpersonal skills to an AI tool, that tool makes you significantly more capable than you were before. The depth you bring in is what determines the quality of what comes out.

Which means the answer for anyone navigating this moment, whether you are Gen Z just starting out or further along in your career, is not to frantically chase every new tool. It is to go deep on your human skills. Know one or two tools really well. And let AI amplify what you have already genuinely built.

So here is the question we are sitting with in the end of all this:

AI can multiply your skills, but only if you actually have them. Are you still building them?

Usually the conversation is about the endless scrolling and shortening attention spans on social media.

But a different version of brain rot might be emerging at work.

AI.

AI tools are quickly becoming the first step in many workflows.

Need an idea? Ask AI.

Need a summary? Ask AI.

Need a draft? Ask AI.

And while these tools make us faster, they also raise a deeper question.

What happens to our thinking when we outsource too much of it?

The Rise of Cognitive Offloading

There’s a concept in cognitive science called cognitive offloading.

It describes what happens when we rely on external tools to handle mental tasks our brains used to perform.

One well-known study from Columbia University explored something called the Google Effect.

Researchers found that people were significantly less likely to remember information when they believed they could easily access it later online.

Instead of retaining knowledge, the brain simply remembered where to find it.

We’ve already seen similar patterns with tools like GPS. When navigation became automated, people’s spatial memory declined.

AI could accelerate this trend.

When the first step in our thinking process becomes asking AI instead of thinking ourselves, we may begin skipping the mental work that strengthens our cognitive abilities.

The Problem With AI Micro-Loops

Another challenge is attention fragmentation.

Many AI workflows look like this:

Prompt.
Scan the result.
Edit.
Prompt again.
Scan again.
Send.

And repeat.

These rapid micro-loops keep our brains operating in shallow processing mode.

But the most valuable thinking often requires the opposite: deep focus.

Research from University of California, Irvine has shown that after a distraction, it can take over 20 minutes to fully return to deep focus.

Constant interruptions make sustained thinking much harder.

And deep thinking is where the most valuable ideas often come from.

Use It or Lose It

Cognitive skills follow a simple rule.

Use it or lose it.

The more we practice certain types of thinking, the stronger those neural pathways become. But if we stop using them, those connections weaken over time.

It’s similar to physical training. If you stop using a muscle, it atrophies.

The same principle applies to mental skills like writing, analysis, and creative problem-solving.

If AI handles those tasks too often, we may slowly lose some of our ability to do them ourselves.

How to Use AI Without Losing Your Edge

The goal isn’t to stop using AI. These tools are incredibly powerful and essential in today’s work ecosystem.

But they need to be used intentionally.

A few simple habits can help protect your thinking:

1. Think before you prompt.
   Add your own perspective before asking AI for input.

2. Use AI as a sparring partner.
   Ask it to critique or challenge your ideas instead of generating them from scratch.

3. Protect deep work time.
   Schedule blocks where you think and create without AI assistance.

4. Interrogate the output.
   Don’t assume AI is correct. Push back, refine, and improve the result.

Used this way, AI becomes a thinking partner rather than a thinking replacement.

And that difference may determine whether these tools strengthen our thinking or slowly erode it.

I need to talk about ClickUp.

Not the product (though yes, it’s the one tab that never closes on my machine).
I want to talk about something most people overlook:
how they built their audience.

A media engine inside the company

Content is unfortunately a side task for many companies.
A few posts here and there, but quickly deprioritized over other tasks.

ClickUp, however, went the other direction.

They built entire teams and systems around their content.
Multiple accounts, formats, angles, structures.

It’s a core foundation of the company itself.
A crazy internal engine.

Two layers that feed each other

What drives this whole engine is two-fold.

1. Company content

They run multiple profiles with different styles and ideas.
Some are more entertaining.
Some are more educational.

A lot of it is built to feel native to the platform it’s found on. (Funny for TikTok, informational for LinkedIn, etc.)

And you will see this content everywhere.

2. Employee content

This is where it gets more interesting.

People inside the company are posting consistently about the brand too.

Founders like Chris Cunningham and others are sharing how they think, what they are testing, and what they are learning. It feels totally unscripted and like you are getting access to how they actually operate.

Content planning is scheduled

They make time for content planning every week.

They sit down, come up with ideas, and batch content. Then they test, review, adjust.

That consistency is a big reason it works (and how such a busy team can get 10 videos out per week).

They follow a simple structure

Not every post is the same (and ClickUp team knows this well).

They break content into 4 key categories:

• Things that have already worked

• Ideas that worked somewhere else

• Trend-based content

• New experiments

Then they spread that across the week.

They are constantly learning + iterating every single week, and learning from what they find.

They invest in people who can actually create

This is not just a marketing hire and a content calendar, and then call it good.

They bring in people who really know how to make content that people actually want to watch.

That includes even unemployed actors and comedy writers. Hires dedicated to this strategy!

Volume helps, but quality still decides

They post a lot. Up to 10 times per week. That gives them more chances to learn what works.

But they have also tested pushing beyond that. And when they did, quality dropped, so they pulled it back.

There is a clear balance between output and quality, and they have tested and tried to find there’s.

They understand how platforms behave

One recent example that stood out: Chris Cunningham has shared about a post that performed super poorly on one platform. Instead of deleting it, they tried it somewhere else. And surprisingly, it ended up performing really well there.

Same content. Different context. If you understand how each platform operates, you can tailor your content more precisely to what those audiences want.

What to take from this

You do not need the ClickUp team size or budget to make this work. But you do need structure!

Start with something simple:

• Pick a posting cadence you can maintain

• Block time each week for ideas and creation

• Test different formats on purpose

• Pay attention to what performs and reuse it

Then adjust based on your own team and bandwidth.

It’s all about prioritizing, testing, and actually paying attention to the results.

Content is how companies show up now.

It is how people find you, how they build trust, and how they decide to pay attention.

ClickUp made that a priority early. And built around it.

If you are building anything right now, take a page out of their playbook!

And the internet kind of just... moved on.

That bothered us. Because when one of the most visited websites in the world draws a hard line on AI content, that is not a “scroll past it” moment. It’s is a signal!!

And on this week’s episode, we sat down to really dig into what it means.

What Actually Happened

In March 2026, the English Wikipedia community passed a formal policy stating that AI-generated content cannot be used to create or rewrite articles on the platform. The only exceptions are basic copyediting of your own writing and assisted translation, and even then, only if the editor is fluent enough in the language to catch errors themselves.

This wasn’t a vague guideline or a soft recommendation. It was a near-unanimous vote from one of the internet’s most trusted communities, and the implications stretch way beyond Wikipedia’s own pages.

The Two Big Reasons Behind the Ban

The reasoning Wikipedia put forward makes a lot of sense when you sit with it.

The first is the obvious one: AI hallucinates. It generates confident, well-structured, completely fabricated information. And Wikipedia is, at its core, a source of truth. Those two things cannot coexist. If people are going to your platform to find accurate information, you simply cannot afford to have AI filling in gaps with invented facts.

But the second reason is the one that really got us talking. Wikipedia is a human-built platform. Its entire identity is rooted in the idea that real people, thousands of volunteers, have spent hundreds and hundreds of hours researching, writing, editing, and debating the content on those pages. That process is the point. It is what gives Wikipedia its credibility.

When AI can generate and edit content instantly, it does not just create an accuracy problem. It creates a fairness problem. It devalues the human labor that built the thing in the first place. And it fundamentally threatens the nature of what Wikipedia actually is.

The Detail That Stopped Us in Our Tracks

Here is something Wikipedia’s own policy statement admitted that we found genuinely fascinating: they cannot reliably detect AI writing by style alone. They have to look at the edit history of an article to figure out whether AI was involved.

Let that sink in. Even Wikipedia, with all its community oversight and editorial infrastructure, cannot always tell whether a piece of writing is human or machine. That is a testament to how far AI writing has come. And it raises some uncomfortable questions about every other platform that has not taken a stance yet.

Model Collapse: The Bigger Picture

Here is where it gets really interesting, and honestly a little unsettling.

Wikipedia is one of the most heavily scraped sources of data on the entire internet. AI companies use it constantly to train their models. So if AI-generated content starts making its way onto Wikipedia, and then that content gets scraped to train the next generation of AI models, you end up in a loop. AI training on AI, training on AI, with no human source of truth anchoring any of it.

Researchers call this “model collapse,” and it is a serious concern in the AI world. The idea is that the quality of AI outputs degrades over time when models are trained predominantly on AI-generated data rather than authentic human knowledge.

Wikipedia, whether intentionally or not, is stepping in to protect that loop. Which raises an interesting question: they do not own any LLMs. They do not build AI models. So why do they care? Our take is that this is probably also a long-term play. Wikipedia’s content is extraordinarily valuable. If the quality of that content degrades, so does their leverage. And there is likely a future in which they can monetize that content significantly. Protecting its integrity now makes a lot of sense in that context.

What AI Can Do That Humans Simply Cannot

We would not be doing this topic justice if we did not talk about what gets lost in an AI-free world, because the other end of the spectrum is genuinely remarkable.

On the episode, Lindsay shared a story that really moved all of us. A friend of hers had recently lost a family member, someone with an incredible life and history. They used Grok to pull together everything that existed on this person across the web, including things pulled from Instagram videos, obscure mentions, corners of the internet that no journalist would ever think to check. Within a short time, they had a portrait of this person’s life that would have taken weeks of dedicated research to compile manually.

That information does not live on Wikipedia and never would. But an AI tool, functioning like a million researchers working simultaneously, was able to surface it and weave it into something meaningful.

That is not a small thing. That is a genuinely new human capability.

So we have two completely different philosophies sitting side by side. Human first, human only, with all the trust and accountability that comes with it. And then human completely removed from the process, but with a depth and speed of synthesis we have never had access to before.

Fair to Hold AI to a Higher Standard?

This is the question we kept coming back to, and we do not think there is a clean answer.

Humans are wrong all the time. Wikipedia has errors. Journalists make mistakes. The news gets things wrong. We accept that, because with human error comes human accountability. You can trace it. You can point to a person, a decision, a moment where something went wrong and understand how it happened.

With AI, that accountability is murky at best. And the nature of the errors is different too. A human editor might miss something, skip over a gap, leave something incomplete. AI will fill that gap, confidently, with something that may have no basis in reality whatsoever. The error is not a hole. It is a fabrication. And that is a fundamentally different problem.

But it does not fully answer the question. Because the standard we hold AI to is effectively “be more accurate than humans.” And that is a bar we do not apply to ourselves.

So here is what we are sitting with this week, and we would love to know where you land on it:

If humans have always been the authors of truth on the internet, and humans have always been fallible, what does it actually mean for something to be “true” online? And who gets to decide?

Let us know your thoughts in the comments, and catch the full conversation on this week’s episode wherever you listen to podcasts.

What’s the big deal with chips?

I was recently at OFC Conference and kept hearing the same word over and over again:

Chips.

Chips. Chips. Chips.

(made me crave some salsa too!)

Everyone is building them.
Everyone is talking about them.
Everyone seems slightly stressed about them.

So I spent a few days trying to understand why.

First, what people actually mean by “chips”

When people say “chips” right now, they are usually talking about AI chips.

Not the ones in your credit card, or even the basic processors in your laptop.

They mean things like:

• GPUs (what a lot of AI runs on today)

• Custom AI chips built for specific models

• Smaller processors designed to live inside your phone or laptop

Companies like NVIDIA, AMD, and Apple are all competing here.

Because these chips are what actually run AI. No chips, no AI.

Why everyone suddenly cares so much

AI models keep getting better. But they also get heavier.

They need more compute to train, and more compute every time you use them.

Something like ChatGPT is not just “magically there” It’s running on hardware somewhere every single time you ask it something.

Multiply that by millions of users, and you get the problem: There is more demand for compute than supply.

Chips are the supply.

What kept coming up at OFC

Needing more chips isn’t the main topic at play here. It’s where those chips are specifically going.

For the last few years, most AI has lived in the cloud and on servers somewhere.

But today, more AI is starting to run directly on your device.

Your phone, laptop, and soon much more.

Cloud AI (what we are used to)

• Runs in data centers

• Very powerful

• Requires sending data back and forth

• Costs add up quickly at scale

On-device AI (what is picking up speed)

• Runs locally on your device

• Faster response times

• Works without internet in some cases

• Keeps more data private

This is why companies like Apple are investing so heavily in their own chips. They want your device (their device) to handle more of the work itself.

Why new chips make this possible

Older chips were not designed for AI. They can handle it, but not efficiently.

The newer ones are built specifically for:

• Running neural networks

• Handling lots of calculations at once

• Doing it without using too much power

That last part matters the most here. Your phone cannot run like a data center. So if AI is going to live on your device, the chips need to be good enough and efficient enough to make that work.

What this actually changes

This changes how products feel and how they get built entirely.

A few things that stood out:

Faster experiences

Less waiting on servers. Things feel instant.

More offline capability

Some features will work without internet.

More privacy

Your data does not always need to leave your device.

Different cost structures

If companies rely less on cloud compute, their costs change. That affects pricing, margins, and how products scale.

The cloud is still part of the picture

One thing that was clear from conversations at OFC:

This is not a full shift away from the cloud.

It is more of a split.

• Bigger, heavier tasks stay in the cloud

• Faster, everyday tasks move to your device

So instead of everything happening in one place, it gets distributed.

The part I was missing before

Before this week, I mostly thought about AI in terms of models.

Which one gives better outputs.
Which one is faster or more accurate.
Which one has new features to test.

But the hardware behind all this is a deeper level that will affect allllll the new features and models that consumers get to see.

Chips decide:

• How fast things run

• How expensive they are

• Where AI can actually exist

That is why this keeps coming up in every conversation.

The simplest way to think about it

AI is the software. Chips are what make it usable.

And right now, a lot of the progress people are excited about is coming from improvements on hardware side.

Chips will be the future of tech innovation.

(now… where is that salsa?)