Your team has five fewer people than it had in January. Your title says the same thing it said in January.

That gap is the whole story this week.

Everyone in this niche, including me, has spent the last six months watching individual contributors for signs of displacement. Data analysts. Designers. Junior engineers. The obvious layer, because it’s the layer that shows up in a layoff announcement with a headcount number attached.

Nobody’s been watching the layer above it.

The Manager Nobody Warned

A friend of mine runs engineering at a mid-size SaaS company. Series C, a few hundred employees, nothing dramatic. Eighteen months ago she had four direct reports, each running their own team of six to eight. Today she has two direct reports. Same title. Same seat at the same table in the org chart. The other two teams got folded in “temporarily” after two managers left within a few months of each other, and nobody backfilled either role.

She’s now indirectly responsible for close to three times the headcount she managed in January of last year. Nobody sat her down and told her the job had changed. It just quietly did.

She didn’t get laid off. She got what looks, on paper, like a promotion. More scope, more people, same comp band for another two review cycles at least. It took her four months to realize that what actually happened was the company stopped paying for the parts of a manager’s job that AI tools now handle. Status rollups. First-pass performance calibration. Resourcing math across a spreadsheet. What’s left, the part where she actually has to sit across from someone and deliver bad news, or decide who gets the one good project three people are fighting for, doesn’t scale the same way. But apparently the company decided it scales enough.

What Actually Happened

For most of the last decade, the working ratio in tech orgs sat somewhere around one manager for every six to eight direct reports. That number wasn’t arbitrary. It reflected how much coordination overhead a single human could carry: status updates, one-on-ones, performance documentation, resourcing, escalations. All of that took real hours.

A meaningful chunk of that overhead is now something an AI tool does faster than a person, and without needing a one-on-one to explain itself. Status rollups write themselves from the tools everyone already uses. Performance calibration drafts itself from ticket data and calendar patterns. None of it needs a manager sitting in the middle collecting it by hand anymore.

So the actual job of managing, the part that was always genuinely human, owning a hard call, delivering unwelcome news, deciding who gets the opportunity, still needs a person. It just doesn’t need one person per seven anymore. Maybe one per fifteen. Maybe twenty, once the paperwork underneath is gone. Same job. Much bigger denominator. Companies noticed the arithmetic before workers did.

None of this needs an announcement, which is what makes it different from the layoffs everyone’s already tracking. There’s no single day where the company says “we are reducing the management layer by 30%.” It happens through attrition and non-replacement. A manager leaves, the team gets folded into someone else’s, and the org chart looks tidy again within a quarter. Nobody has to explain a headcount reduction because, technically, there wasn’t one. Just a series of small mergers that add up to the same outcome with none of the visibility.

So if you manage people, the question worth actually asking tonight isn’t whether your job feels safe. Does your role right now require the same headcount underneath it that it did eighteen months ago? For a lot of you the honest answer is already no. That’s the signal, already fired. You’re just still catching up to what it means, the same way my friend spent four months catching up to it.

The Move

Two things I’d actually do this week, if I were sitting where she is.

Stop tracking your value by headcount managed. That number’s inflating for reasons that have nothing to do with how well you’re doing the job, so it’s about to stop meaning anything as a status signal. Start a running list instead. The calls only you made. Who got the stretch assignment and why. What you escalated versus quietly absorbed. Which fire you let burn on purpose versus which one you jumped on immediately. Write it down as it happens, not from memory later, because memory later is exactly how this evidence disappears when you need it most.

Then have the conversation before the org chart has it for you instead. Ask whoever you report to, directly, what they think this role actually consists of now that half the old overhead runs itself. Information gathering, not a confrontation. The answer tells you whether you’re quietly being re-scoped into something bigger, or quietly being prepped to fold into someone else’s team the way the last two managers were.

She wishes she’d asked eight months earlier than she did. Says it plainly every time we talk about it.

Nobody eliminates the manager outright. They just stop giving them anyone left to manage.

I said something blunter about this same pattern on X earlier this week, if you want the sharper, less patient version of this argument. Go check out @mangoaggro on X.

Read more

There’s a company I’ve been watching that handled its AI-driven reorg differently than almost anyone else this year, and it still cost people their jobs.

I want to be honest about that up front, because the version of this story that just says “here’s the good company, go work there” is exactly the kind of tidy narrative that doesn’t actually help anyone. Nobody reading this can go work there. That’s not the point. The point is what they did differently, because the differences are copyable even when the company isn’t.

Most of what I track in this newsletter is the secretive version. Vague town halls. “We’re still figuring out the org chart” for months at a stretch. A first round of cuts followed by a second one nobody saw coming because nobody was told a timeline in the first place. That pattern isn’t cruelty exactly. It’s mostly cowardice wearing a strategy document.

This company did something different. When they announced the AI-driven restructuring, they gave an actual number. Not “some roles may be impacted.” A number, and a date range for when decisions would be finalized. People had six weeks of genuine uncertainty instead of six months of managed uncertainty stretched out to avoid an uncomfortable quarter.

Six weeks is still brutal. I’m not pretending otherwise. But six weeks with an end date is a completely different experience than six months without one. You can make decisions inside six weeks. You can start looking, start saving differently, start having the conversation with your partner earlier instead of pretending everything’s fine at home the way I’ve heard from more people than I can count. The secretive version steals that time from you without you ever knowing it was being stolen.

The other thing this company did, which I’ve genuinely never heard described before by anyone going through this: they let people who were being cut choose their own last day within a window, rather than walking them out immediately. Two weeks, four weeks, whatever they needed to hand off cleanly and start their search with some dignity intact. It cost the company a little in productivity during the transition. It clearly cost them less than the alternative, which is a workforce that spends the next year assuming every “AI efficiency” announcement is a trap.

A friend of a friend went through the version I described first, the six-month secretive kind, earlier this year. I mentioned it a few weeks back. I’ve been thinking about the contrast since I found this second story, because it’s made me second-guess something I say a lot in this newsletter, which is that the corporate behavior around these cuts is uniformly bad. It mostly is. It doesn’t have to be, and that distinction matters more than I think I’ve been giving it credit for.

Here’s the contrarian part. Everyone assumes the honest version costs the company more. Slower deployment stories, more employee pushback in the short window, harder conversations with the board about why the number was said out loud instead of buried in a footnote. I don’t think that’s actually true past the first quarter. A company that lies about its restructuring timeline spends the next year managing the mistrust it created. A company that tells the truth about a hard number spends that same year rebuilding, with a workforce that mostly still believes what leadership tells them. Trust is expensive to buy and it compounds either direction.

If you’re currently inside the secretive version, and most of you are, here’s what to actually take from this. You can’t make your company honest. You can act as if they’re going to behave like the secretive version regardless of what they’re currently telling you, because that’s the higher-probability outcome. Build the plan for the six-month version even while hoping for the six-week one. Ask for a specific timeline anyway, not because you’ll get an honest answer, but because how they respond to the question is itself information. Watch whether they give you a real date or a version of “we’re still figuring it out.” You already know which one that answer usually means.

Doing it honestly didn’t make it not a layoff. It just made it one you could actually plan for.

Subscribe for the specific move below — Career insurance caption:

Read more

The calendar invite for the “AI efficiency town hall” and the calendar invite for the layoff meeting are never the same week. They’re never even the same quarter, usually. But they’re closer together than anyone wants to admit while they’re sitting in the first one.

Six months. That’s the number I keep landing on when I go back through the cases I’ve actually tracked, not the ones I’ve heard about secondhand. Company announces an AI-driven restructuring. Doesn’t fully deploy what they promised. Six months later, give or take a few weeks either direction, the second round of cuts lands on the same team that survived the first one.

Not every company. Not every time. But often enough now that it stopped looking like coincidence around the fourth or fifth case.

I want to be precise about what this actually is, because precision is the whole point of tracking predictions publicly instead of just vibing about doom. This isn’t a claim that AI caused the second round. It’s a claim about timing. The gap between the promise and the reckoning has a shape to it. Once you’ve seen the shape a few times, you stop being surprised by it.

Here’s the mechanism, as far as I can tell. A company announces AI-driven efficiency gains alongside a first round of cuts. The promise buys them roughly two quarters of runway with the board and with Wall Street, if they’re public. Somewhere in that window, someone actually tries to deploy the tooling at the scale that was promised. It works partially. It works enough to look plausible in a slide deck. It doesn’t work enough to cover the full gap between the headcount they cut and the headcount they actually need.

By month five or six, that gap becomes visible internally before it becomes visible externally. Someone in finance notices the numbers aren’t holding. The second round gets planned quietly while the company is still publicly discussing “the success of our AI transformation initiative” in whatever forum they discuss things like that in.

I tracked this pattern first back in Issue 9, in May, as a prediction rather than an observation. Second wave of cuts between late 2026 and early 2028, at companies that announced AI restructuring without completing deployment. That prediction is still live. Nothing about the last two months has made me want to walk it back. If anything the six-month rhythm has gotten more specific the more cases I’ve actually looked at closely.

A friend of a friend went through this exact sequence earlier this year. Survived the first round at a mid-size SaaS company in March. Was told directly, in the town hall, that the team was “right-sized” for the AI tooling being rolled out. Believed it, mostly, because there wasn’t much choice not to. By September, six months later almost to the week, half of what remained of that team was gone in a second round nobody had been warned about. The tooling from March was still not doing what it was promised to do in March. It just took six months for that gap to become someone else’s problem to solve with headcount instead of engineering.

I don’t think that’s an unusual story anymore. I think it’s becoming the normal one, and normal is the part that should worry people more than dramatic ever does.

Here’s the contrarian part, the bit that goes against what most people assume when they hear “AI layoffs.” The AI usually isn’t lying about what it can do. The org chart is lying about how fast it can absorb that capability. Everyone wants to blame the model. The model mostly does what it says on the label. What breaks is the six-month bet the company made that deployment would move faster than it actually moves.

If you’re still employed somewhere that had its first round of AI-branded cuts in the last few months, that six-month window is your window too. Not to panic in. To pay attention in. Watch how much of what was promised is actually shipping versus how much is still slide-deck. Watch whether the team above you is getting quieter about timelines. Watch whether contractor postings start showing up for scope that technically already got automated away.

I audited my own week against this same instinct recently, actually. Went looking for the places where I’ve been assuming a tool handles something that I’m actually still doing manually and just not tracking honestly. Found two. Neither one was flattering to notice.

The gap between what got promised and what got shipped doesn’t announce itself. It just sits there accumulating until someone in finance decides six months is long enough to wait.

The six months isn’t a rule written down anywhere. It’s just how long it takes for a decision to become a rumor, then a memo, then your last day.

Subscribe for the full role-specific breakdown below — Career insurance caption

Read more

Ford rehired the engineers it laid off for AI last year, and called it a staffing update.

No press release. No line anywhere admitting the automated quality system missed defects a person would have caught. Just new job listings, same titles, quieter this time, filed under something that sounds routine.

CBA did roughly the same thing with 40 customer service roles. The AI voice system couldn’t handle actual call volume once real customers started using it instead of the demo script. Nobody at CBA wrote “we were wrong” anywhere a shareholder could see it. The correction happened the way these corrections always happen. Quietly, in a spreadsheet, months after the original announcement got the headline.

I’ve watched a version of this happen to someone I know directly. A former colleague’s team got the AI-cut memo in March. Two of those exact roles reopened by June, different job title, same responsibilities, same desk practically. Nobody circled back to explain the gap. The gap just sat there, unnamed, the way it always does.

IBM did a version of this too, in the opposite direction, which is almost more revealing. Their own HR leadership said out loud this month that the entry-level pipeline dries up if a company stops hiring into it, and announced tripling entry-level hiring as a direct response. That’s a company admitting, in public, that the earlier decision to gut junior roles created a problem two or three years down the line that somebody now has to fix in a hurry.

Three companies, three different functions, one pattern underneath all of them. The decision gets made fast, gets an AI story attached to it because that story is available and plausible, and then gets quietly walked back or patched over once the actual consequences show up. The correction never gets the same audience as the original announcement. It shows up in a hiring update, a staffing note, an HR memo nobody outside the building ever reads.

Here’s what actually moved this year, and it isn’t the thing everyone’s pointing at.

The Number That Actually Moved

AI was cited as the reason for layoffs in roughly 40 percent of all job cuts announced in May, according to Challenger Gray’s tracking. That’s the third month running where AI has been the single most-cited reason for a workforce reduction. In January, the same category sat at seven percent.

Read that jump out loud. Seven percent to forty percent in four months. Most coverage treated that as proof the technology finally caught up to the hype. I don’t think that’s the actual story, and the reason I don’t think that is a different number that dropped in the same window.

A RAND analysis published this year found frontier-adjacent models coming out of China running at somewhere between a sixth and a quarter of the cost of comparable US systems. Model access got dramatically cheaper across the board this year, not just from one lab, across the entire competitive field.

Here’s the connection nobody in the mainstream coverage bothered to draw. When the underlying tool costs almost nothing to run, citing “AI” as the reason for a cut costs a company almost nothing to claim. Whether or not the automation actually replaced the function cleanly stops being the relevant question. The citation is nearly free. The deployment is a separate, much slower, much messier problem that most companies haven’t actually finished solving.

Ford and CBA are the receipts for exactly that gap. The technology got cited on the way out the door. The technology hadn’t actually finished the job when the door closed.

The Gap Between The Citation And The Reality

The conventional read on the Challenger numbers is straightforward. AI displacement accelerated hard this year, the data proves it, brace for more of the same.

The uncomfortable read, the one I think is closer to what’s actually happening on the ground, is different. A rising AI-citation percentage doesn’t prove AI is displacing workers faster. It proves AI has become the cheapest, lowest-friction line a company can write in a memo, whether or not the actual deployment behind that line is finished, half-finished, or barely started.

That distinction matters enormously if you’re the person losing the job over it. It means the thing supposedly replacing you might not even work yet. It means the excuse arrived faster than the automation did, and you’re the one absorbing the gap between the announcement and the reality.

This isn’t a comforting reframe. I’m not telling you your job is safe because the AI probably can’t actually do it. I’m telling you the reason given for the cut and the actual technical truth underneath it have come apart from each other, and almost nobody currently discussing this wave is distinguishing between the two.

Companies aren’t lying, exactly. They’re doing something closer to using a genuinely real trend as cover for decisions that were mostly about headcount targets and cost efficiency all along. The AI story is true enough, broadly, industry-wide, to make a specific citation plausible in any specific case, whether or not it’s the actual reason in that specific case.

Think about how fast that citation became available to use. Two years ago, “we’re restructuring around AI” wasn’t a sentence most leadership teams could say with a straight face internally, let alone put in an external memo. Now it’s the default explanation, reached for automatically, the way “synergies” used to get reached for during an earlier decade of cuts. The explanation industrialized faster than the technology did. That’s not a conspiracy. It’s just what happens when a plausible story becomes free to tell and nobody’s checking whether it’s actually load-bearing in any individual case.

The people most at risk right now aren’t necessarily in the roles AI is best at automating. They’re in the roles where citing AI is cheapest and least likely to get challenged internally, whether or not the underlying capability is actually there yet. Those are two different maps of exposure, and most people are currently studying the wrong one.

What This Means If You’re Reading This At Work

If your company just cited AI as a factor in a cut, a hiring freeze, or a restructuring announcement, don’t take that citation at face value. It’s not that they’re definitely lying. It’s that the citation tells you almost nothing about whether the actual technical capability behind it exists yet, in your specific function, at your specific company.

Ask a different question instead, quietly, to yourself first. Did the actual output of the eliminated function change, or did the finance decision change? Those are two different problems wearing the same explanation, and they call for two completely different moves on your part.

If the output genuinely changed, if the work is genuinely getting done differently now, that’s a real signal about where the function is headed and you should plan around it honestly.

If the finance decision changed and the AI citation is doing the explaining, that’s a different kind of exposure entirely. It means the actual deployment is still catching up to the announcement, which means there’s a window, not a permanent state, and windows can be used.

A quick version of the audit, for whichever situation you’re actually in. Has your own function’s output visibly changed in the last two quarters, or has the conversation about your function’s budget changed. Did the tool get better, or did the target get smaller. Would the “AI did this” explanation survive someone actually checking the deployment logs, or is it doing the work a much blunter explanation used to do.

None of this means panic less. It means panic more precisely, at the actual thing happening, instead of the version of it that fits neatly into a headline. The generic version of this fear, the one that treats every layoff as proof the robots finally arrived, is easier to sit with in some ways. It’s also less useful, because it tells you to prepare for the wrong threat. The specific version, the one where the citation and the deployment are two different clocks running at two different speeds, tells you exactly where to look and roughly how much time is actually on it.

The citation is cheap. The deployment still isn’t finished. You can tell the difference if you know where to look, and most of the people currently getting cut don’t get told which situation they’re actually in.

If you’ve been sitting with a citation you were never actually shown the evidence for, that’s the same gap a lot of you have been sending me examples of on X lately. Worth a look over there too.

Read more

DeepSeek dropped a number on Friday that should have been the only AI story this week. Its newest flagship model is priced roughly 97% below OpenAI’s GPT-5.5. Cached input costs got cut to about a tenth of what they were months ago. Input pricing on some configurations is now down near a fraction of a cent per million tokens.

Every outlet that covered it called it a win.

A win for developers. Cheaper compute. Healthy competition doing what competition does. Bloomberg ran it as a price war story. The trade press ran it as a China-versus-Silicon-Valley story. A RAND report came out around the same time putting Chinese frontier models generally at a sixth to a quarter of the cost of comparable US systems, and that got folded into the same framing. Progress. Access. Good news.

Nobody wrote the other headline.

The cost of replacing a person with a model just fell by close to the same percentage as the discount on the model itself. Those are not two separate stories happening to land in the same week. They are the same story, told from the side that doesn’t get the press release.

The excuse that just ran out

For the past two years, every company sitting on an AI deployment roadmap has had a built-in delay valve. The technology isn’t fully there yet. The economics don’t quite work yet. We’re piloting. We’re being careful. That language did real work. It bought time. It let leadership sound responsible while quietly building toward something nobody wanted to say out loud yet.

The economics just stopped being the delay valve.

Alibaba’s Qwen passed a billion cumulative downloads on Hugging Face. It took over half of all global open-source model downloads in a single month this year, more than the next eight competitors combined, including Meta, DeepSeek, and OpenAI. That’s not a niche developer community experimenting. That’s the default starting point for the next wave of deployment, already chosen, already downloaded, already running in places nobody’s writing headlines about yet.

A flag on a model doesn’t change what it does to a budget conversation. If a US-based team has been holding off on full deployment because the per-token cost made full automation hard to justify against keeping a person on payroll, that justification just got a lot harder to hold onto. Not because anyone made a moral choice about it. Because the spreadsheet changed.

Worth being specific about what this actually looks like in a real budget meeting, because it rarely gets said the dramatic way it gets written about here. Nobody stands up in a Q3 planning session and announces the robots are taking over now. What actually happens is smaller and harder to point at. A line item that used to read “external agency support, content production” gets a note next to it asking whether the volume could run through an internal tool instead. A headcount request that would have sailed through eighteen months ago gets a follow-up question it didn’t used to get. A team lead gets asked, almost as an aside, whether the new hire really needs to start in Q3 or whether Q1 makes more sense once the current tooling pilot wraps up. None of that reads as a layoff. All of it is the layoff, just spread out across enough small decisions that no single one of them looks like the moment everything changed. By the time anyone names it out loud, the team has already been smaller for two quarters and everyone just sort of got used to it.

Here’s the part that connects to last week. The US government spent real effort trying to slow access to the most capable models built on American soil. Anthropic’s Mythos and Fable went offline worldwide within hours of a single finding, no warning, no exceptions for paying customers, citing national security. Read on its own, that looked like a story about control. About a government deciding it could move fast when something threatened it.

Read next to this week’s pricing news, it reads differently. You can gate one lab’s frontier model. You cannot gate a price collapse happening across an entire industry on a different continent, with multiple labs releasing in rapid succession, each one undercutting the last. The alternative was already cheap enough that the gatekeeping barely slowed anything down. The control that mattered, the one that actually protects people, was never about which model gets restricted. It was always about whether there’s any floor under what happens to a person once the model, any model, gets cheap enough.

There isn’t one. Not yet. Not federally, anyway. Colorado’s AI Act took effect this week and covers algorithmic discrimination in hiring. It says nothing about disclosing when AI played a role in a layoff. No federal law requires that disclosure anywhere in the country. The Worker Technological Displacement Act, which would have forced 90 days notice and named the systems involved, still hasn’t passed. A companion bill asking for human oversight on AI employment decisions is stalled in the same place bills like that usually go to wait.

The framing everyone’s getting backwards

Most of the commentary on the Chinese pricing story treats it as a competitive threat to American AI companies specifically. Will OpenAI cut prices to match. Will Anthropic. Will this dent the valuations everyone’s been arguing about for two years. That’s the framing built for people who hold equity in AI labs, not the framing that matters for people who hold a job that touches the output of one.

The competitive question and the worker question used to track together, more or less. Slower frontier progress meant slower deployment meant more runway before any given role got touched. That link is broken now, because the thing driving the runway was never really the technology’s ceiling. It was the price floor. Once price stops being a constraint anywhere in the world, it stops being a constraint for the companies making decisions about your role too, regardless of which lab’s name ends up on the API they’re calling. A contrarian read of this week’s news isn’t “America is losing the AI race.” It’s that the race framing itself was always a distraction from the only question that actually lands on a person’s desk, which is simply whether the task in front of them got cheap enough to hand to a model yet. This week, for a lot more tasks than last week, the answer changed.

What I’m actually predicting, specifically

The first public prediction this newsletter made, back in May, was a second wave of cuts between late 2026 and early 2028, at companies that announced AI restructuring without finishing the deployment. That one is still tracking. Nothing this week changes it.

Here’s a second one, sitting alongside it, not replacing it. By the first quarter of 2027, expect at least one Fortune 500 company to cite AI infrastructure cost efficiency directly, in writing, as the explicit reason for a layoff of 500 people or more. Not a euphemism. Close to the language Oracle already used in its own 10-K filing this month, when it told the SEC in plain terms that AI adoption and deployment had resulted in workforce reductions. That kind of language didn’t exist in a public filing eighteen months ago. It exists now because the math finally makes the boring true reason the legally safer one to admit.

If it doesn’t happen by then, I’ll come back here and say so, the same way I’d say it if I were right. That’s the deal with making a call in public. The wrong calls get the same airtime as the right ones, because the airtime was never the point. The point was being someone who’s willing to be checked against reality instead of someone who only talks in language vague enough to never be wrong about anything.

Read more

It’s been eight days since Anthropic announced Project Glasswing. Still no confirmed partner list.

At some point a silence this long stops being suspenseful and starts being the actual condition you’re operating inside. Nobody’s going to tap you on the shoulder and tell you whether your company got early access to the most capable AI model that exists right now. You have to go find out yourself, or accept that you might never get a clean answer.

So this issue isn’t more analysis of what Glasswing might mean. It’s the checklist. What to actually go check this week, regardless of whether confirmation ever comes.

Why waiting doesn’t work

Some companies that got Glasswing access will never announce it. There’s no upside in saying so publicly, and real downside in inviting questions about org structure, headcount plans, or what exactly “trusted partner” means for the people doing the work. Silence protects the company more than disclosure would.

Which means the binary you’re looking for, confirmed or not confirmed, might just never resolve. Planning your next six months around an announcement that may not come is planning around nothing.

The information you actually need is closer than a press release. It’s sitting in your own company’s behavior, if you know where to look.

Three things to check this week

Job postings. Pull up your company’s careers page. Search for language that wasn’t there two months ago. “AI-augmented,” “human-in-the-loop oversight,” “working alongside advanced models,” anything that frames the role as a partnership with a capability rather than ownership of a task. Hiring managers write these descriptions based on what’s actually changing day to day, often before anyone in legal or comms has approved saying it publicly. The job posting tells the truth before the announcement does.

Internal tooling. Has something rolled out in the last month that nobody fully explained? A new internal assistant. A new approval workflow that quietly changed how a process works. A “pilot program” that got mentioned once in a Slack channel and then went quiet. Pilots that go quiet are frequently pilots that are working well enough that leadership stopped needing buy-in and started just rolling it out.

Leadership language. The softest signal, but a reliable one. Listen for what’s missing as much as what’s said. If your leadership used to talk about “efficiency” and now talks about “capability” or “positioning” or “future readiness,” that drift toward vagueness usually means something specific is being held back from a name. Vague language at the top is rarely an accident. It’s a placeholder for something not ready to be said plainly yet.

None of these three things confirm Glasswing access specifically. You’re not going to find a smoking gun that says “yes, we have it.” But all three together, especially if they’re new in the last 60 days, are a strong enough signal that something has changed in how your company is thinking about AI capability. Whether that’s Glasswing specifically or just the broader access race around it matters less than the fact that it’s happening at all.

What it means depending on what you find

If you check all three and find nothing, that’s also useful information. It probably means your company is behind the curve on access, which has its own exposure profile, just a different one. The gap between your company and a better-resourced competitor is now the thing to watch, not your own internal signals.

If you find one or two signals, pay attention but don’t panic yet. A single job posting with new language might just be one hiring manager experimenting with phrasing. Internal tooling rollouts happen for plenty of reasons unrelated to frontier model access.

If you find all three at once, especially in the last two months, that’s a different situation. That’s a company actively repositioning around AI capability, whether or not they’ve said so out loud. The honest move at that point isn’t dread. It’s documentation.

There’s a version of this that applies even if you find absolutely nothing. Plenty of companies are genuinely behind on this, not because they’re badly run, but because access races like Glasswing don’t distribute evenly. Smaller companies, companies outside the US, companies without existing enterprise relationships with frontier labs, all of them are realistically watching this from further back. Being behind isn’t a personal failure of your employer. It’s a structural position. But it’s still worth knowing which position you’re in, because the moves that make sense differ depending on where you sit.

If your company is behind, the relevant question isn’t “are we getting Glasswing access.” It’s “how is leadership thinking about closing the gap once general availability arrives, and what’s the timeline they’re imagining.” Some companies will move fast the moment the model goes public. Others will sit on legacy tooling for another year out of inertia. Both are useful things to know about where you work.

The part that doesn’t show up in any checklist

This part is harder to audit than job postings or internal tools. Even companies with zero AI access changes happening internally are still operating in a market where their competitors might have them. That competitive pressure shapes decisions even before any tool gets deployed. A company that suspects a rival has frontier access sometimes makes defensive headcount decisions just to look efficient, completely independent of whether AI actually changed anything internally yet.

This is the part of the story that’s genuinely hard to see from inside a single company, because it’s happening between companies, in board rooms and investor conversations you’re not in. You can’t audit your way to visibility on this one. You can only stay aware that it’s a factor, and treat any sudden “efficiency” language from leadership as potentially about competitive anxiety rather than actual internal AI deployment.

The move that doesn’t depend on any of this being confirmed

Whatever you find, or don’t find, the actual move is the same. Write down what you specifically own that would be hard to replicate if you disappeared for three months. Not your task list. The judgment calls that live in your head because of context nobody else has. The relationships that exist because of years, not job titles. The thing that breaks quietly if you’re not there, even if nobody notices immediately.

This isn’t busywork. It’s the difference between knowing where you actually stand and finding out the hard way, later, from someone else’s decision.

One more thing worth saying plainly

None of this is about becoming paranoid at work. Checking your company’s job postings once and noting a phrase shift isn’t a reason to spend the next month in a state of low-grade dread. The point of a checklist is to run it, get an honest read, and then go back to doing your actual job with slightly better information than you had before.

The companies that benefit most from this story are the ones who get to operate while everyone else stays frozen, watching the news cycle for a confirmation that may never arrive. Don’t give them that. Run the checklist once this week. Write down what you’d want documented either way. Then get back to the work that’s actually in front of you.

I went and checked my own former employer’s careers page while writing this. Half expecting to find nothing, mostly doing it for the example. Found three postings with “AI-augmented” in the title that weren’t there two months ago. Sat with that longer than I expected to. Knowing the framework and actually finding it somewhere real are different feelings entirely.

UK readers: the job posting language check works the same way regardless of where you are. The phrasing shift tends to start in US postings first and lag by a few weeks elsewhere, so if you’re not seeing it yet, that’s more about timing than about being safe.

The companies running this play don’t need to confirm anything. They just need you to keep waiting for confirmation while they make the decisions that actually matter.

Go check your own company’s postings this week. That’s the whole assignment.

Posted the same checklist on X this morning. If you want the shorter version to send to someone who keeps saying they don’t know if they should be worried, that’s the one. Link in bio.

Read more

Your engineering manager didn’t mention it during the standup on Tuesday morning, but your three-year architectural roadmap just became an active corporate liability.

Everybody spent this week obsessing over the “Mythos” model release. They ran benchmarks, posted synthetic latency charts on social feeds, and argued over whether context windows have truly plateaued.

It’s comforting noise. It keeps your hands on the keyboard, looking at the technical variables you feel qualified to control. But while you were looking down at the terminal, your executive leadership team was looking up at a board deck that frames your entire team as a legacy cost center waiting for an automated transition pipeline.

The buzzword floating through executive suites right now isn’t capability. It’s “AI sovereignty.”

When you strip away the institutional marketing speak, sovereignty means your employer is terrified of being permanently locked into a single infrastructure vendor’s price increases. To solve this, enterprise technology organizations are rapidly moving toward uniform orchestration layers. They are standardizing environments so that models can be hot-swapped out without human developers rewriting a single line of custom middleware.

The problem for you is that when a company builds an engineering pipeline designed to make advanced infrastructure completely interchangeable, they accidentally make the human beings who write the integrations completely interchangeable too.

I sat in a coffee shop in downtown three days ago with a lead systems engineer who has spent eighteen months building a proprietary data processing pipeline for a major logistics firm.

He was exhausted. His eyes were bloodshot, and he kept looking at his phone as if a production alert was about to drop. He told me his team had successfully cut latency by 22 percent using custom embedding wrappers.

He felt safe. He felt indispensable. Then he showed me an internal memo from his CTO office dated last Thursday. The company is adopting a standardized vendor layer that natively processes those embeddings at the infrastructure level. His eighteen months of optimized architecture is being turned off at the end of the quarter. His entire department is being consolidated from twelve people down to an operations team of three.

He didn’t build an enduring career pillar; he built a temporary scaffolding that the system no longer requires.

This is the hidden mechanics of the second wave of white-collar displacement.

The first wave was blunt cuts. Firing junior engineers and copywriters to preserve margin.

The second wave is structural. It happens when your proprietary work is absorbed by an upstream update.

Consider the baseline data across the sector. When enterprise software firms rolled out native automated documentation and unit testing suites over the last twelve months, initial consensus suggested developers would simply reallocate their time to higher-level product design. The reality on the balance sheet reflects a completely different outcome. Across mid-market tech firms tracking operational efficiency metrics, internal deployment velocity increased by 41 percent, while net technical headcount across those same teams dropped by an average of 18 percent through quiet attrition and targeted restructuring.

The work didn’t disappear. The requirement for individual engineers to manually bridge the gaps between components did.

If your professional leverage is built entirely on navigating the arbitrary constraints of a platform you do not personally own, you aren’t an architect.

You are a temporary tenant.

You are spending your nights studying documentation for a system that is actively incentivized to make your understanding obsolete. Every time an enterprise vendor drops a model update that abstracts away another layer of the stack, your specialized knowledge doesn’t just lose value. It transforms into an unnecessary friction point.

The companies that survive this shift aren’t the ones racing to adopt every new model version within minutes of a press release. They are the businesses treating models as a commodity, like electricity or municipal water. If you are positioning yourself as the person who knows how to configure the plumbing for a specific brand of pipe, you are tracking the wrong asset. The leverage is moving entirely to the edges of the ecosystem: data absolute ownership at the bottom, and absolute user distribution at the top. Everything in the middle is an operational cost that is being compressed toward zero.

You have to stop optimizing for the integration. You have to start optimizing for the raw primitive. If your day job consists of pulling data out of one corporate repository, restructuring it slightly, and pushing it into an external model endpoint, you are training the very automated routine that will replace your seat before the end of the year.

The Paid Audit: The Structural Exposure Framework

This section is for paid subscribers who need to audit their true workplace leverage before the next infrastructure update shifts their team’s headcount.

Read more

The Samsung workers who got a share of this had to threaten a strike to get it in writing. I’ve been tracking the pattern on X @MangoAggro if you want the running commentary.

There’s a version of this story that makes the company look incompetent.

They hired a senior engineer to build internal tooling. He spent two years on it. The tooling worked. Within eight months, his role was redundant and his team was cut from six to two.

The incompetent version says: they didn’t plan for this. They didn’t think through the consequences. Classic corporate short-termism.

That version is wrong.

They planned for exactly this. The engineer built what they asked him to build. The productivity gain landed where it was supposed to land. His role was always going to compress once the system ran without him. Nobody lied. Nobody made a mistake. The system worked as designed.

That’s what makes it worth examining.

The productivity goes somewhere

In April, Anthropic published a survey of 81,000 Claude users. The average user reported being “substantially more productive.” Ninety percent said the gains stayed with them, not the company.

Read that alongside what GM reported the same quarter. They beat earnings estimates by 40%. They also cut another 500 to 600 IT workers in Texas and Michigan. In total, GM has eliminated over 10,000 white-collar salaried roles since 2022.

Both numbers are true. The Anthropic survey and the GM layoffs are not contradictions. They are the same equation viewed from two different positions in the org chart.

The worker experiences the productivity gain. The company captures the headcount reduction. Those are two separate transactions and they don’t happen at the same time or to the same people.

The senior engineer who is 40% more productive this year is not getting a 40% pay increase. He’s getting a performance review that says he’s performing well. The company is getting a business case to not replace the two people who left his team last quarter.

This is the redistribution problem in its most practical form. Not an abstract policy debate. A spreadsheet decision made by someone in finance who has never met the engineer in question, working from headcount targets approved in a planning meeting three months ago. The engineer isn’t in that meeting. His productivity numbers are.

Who is running the AI taxation conversation

Governments in the EU, UK, and parts of the US are now actively discussing AI taxation. The proposals vary. Some target compute. Some target companies above a headcount reduction threshold. Some propose redirecting a share of productivity gains into retraining funds.

The proposals share a structural problem. They’re written for the displacement that happened in 2023. This newsletter called the second wave in Issue 9: cuts between late 2026 and early 2028 at companies that announced AI restructuring without completing deployment. By the time any of this legislation passes, that wave is already over.

The consultation process itself is worth examining. When governments open AI taxation consultations, they invite submissions from affected parties. The parties with the most resources to submit detailed, technically credible responses are the same companies that will be taxed. The parties most affected by the displacement are represented by unions where they exist and by nobody where they don’t. The outcome of a consultation shaped primarily by the entity being consulted about is predictable.

The other structural problem is the measurement gap. A lot of the displacement doesn’t show up cleanly. Roles that stop being backfilled. Teams that quietly compress over 18 months. Junior positions that disappear without a layoff announcement. You can’t tax what you can’t see and the most significant displacement in knowledge work right now is designed to be invisible.

This is not an accident. When a company announces a layoff, it generates headlines, regulatory scrutiny, and sometimes legal exposure. When a company simply stops hiring for roles made redundant by AI, none of that happens. The headcount compresses silently. The people affected aren’t laid off. They just don’t get hired. Their jobs never existed in a way that generates a statistic.

The companies lobbying hardest against AI taxation this year cited AI efficiency in their last round of layoff announcements. That is not ironic. That is a coherent position: we captured the gains, we are not interested in redistributing them, and we have the resources to ensure the policy conversation moves slowly enough for the fait accompli to be complete.

The one deal that worked

In May 2026, Samsung and its union reached an agreement after months of standoff. Seventy-four percent of workers voted yes. The deal gave workers a share of AI chip profits, a pool covering 78,000 employees, structured around operating profit thresholds over ten years.

It worked because the workers had direct leverage over what the company was selling. Samsung makes the chips that power AI. The workers in the chip division could threaten to stop making them. That leverage is specific, credible, and hard to replace with tooling.

Most tech workers don’t have that. A senior software engineer’s leverage is real but diffuse. The company can absorb losing one person. It cannot absorb losing the workers who make the physical infrastructure that the whole AI economy runs on.

The Samsung deal is the floor, not the model. It required a near-strike to get in writing. It required workers with direct leverage over an irreplaceable part of the supply chain. Most of the people reading this don’t have either of those things.

There is one other case worth examining. In April 2026, the Intermediate People’s Court of Hangzhou ruled that a company could not cut a worker’s salary simply because AI had absorbed his role. The court applied existing Chinese labor law: AI adoption doesn’t qualify as the kind of extraordinary event that justifies unilateral changes to employment terms. No new legislation. Standard labor protections applied to an AI case.

That’s a different mechanism than taxation. It doesn’t redistribute the gains. It says the cost of the automation decision stays with the employer, not the worker. Small protection in practice. But a different way of thinking about where the obligation sits.

What the engineer actually did wrong

Back to the engineer who trained his replacement.

He made one mistake and it wasn’t technical. He built a system that ran without him and then stayed in the role that built it rather than moving to the role that owned what the system produced.

The system generated output. Someone had to be accountable for whether that output was good. Someone had to make judgment calls when the system produced something wrong or incomplete. Someone had to sit in the meetings where the output was used and be the person who caught the problems before they became expensive.

That’s an outcome job. He was doing a process job and he automated the process. He should have automated himself into the outcome role before someone else noticed he’d done it.

The process versus outcome distinction is the frame this newsletter returns to because it’s the one that actually holds. A process job executes known steps to produce a deliverable. AI does this well and gets better at it every quarter. An outcome job owns a result you’re held accountable for when it breaks. AI assists. The human owns it. Those two things are not the same and the companies currently cutting roles understand the distinction even if they don’t use the same words for it.

I’ve watched this happen to people I know. Not the dramatic version. The slow version where the role doesn’t get eliminated. It just stops growing, stops getting backfilled when someone leaves, stops coming up in planning meetings. You’re still there. The role just has less gravity than it used to. That version is harder to fight than a layoff notice.

The honest read on policy

AI taxation will arrive. Some version of it will pass in some jurisdictions in the next five years. It will be underfunded, partially captured by the companies it’s meant to tax, and four to five years behind the displacement curve it was designed to address.

That’s not cynicism. That’s the track record of every previous attempt to legislate the distributional consequences of a technology wave in real time. The EU’s GDPR took six years from proposal to enforcement. By the time it had teeth, the data practices it was designed to prevent were already the infrastructure of the entire industry.

AI taxation will follow the same arc. The gains will have been distributed, the roles will have been eliminated, the new equilibrium will be established. Then the legislation will arrive to govern a situation that has already resolved itself in favor of whoever moved fastest.

The person reading this newsletter has a career window that doesn’t overlap with that timeline. Policy is not the map. The map is something else.

Paid subscribers get the specific move below.

The founding tier is still open. First 50 lock in the rate forever. Link below.

Read more

Everyone keeps talking about layoffs.

I think we’re missing an earlier signal.

Burnout.

Not the normal kind.

Not “I need a vacation” burnout.

Not “too many meetings” burnout.

A different version.

The kind that comes from feeling like your profession is moving underneath your feet while you’re still standing on it.

I’ve been noticing something strange.

When people talk about AI displacement, they usually talk about jobs disappearing.

Headcount reductions.

Hiring freezes.

Restructuring.

Those are easy to see.

The harder thing to see is what happens before any of that.

The engineer spending nights learning new tools because they don’t know whether they’ll still be competitive in two years.

The analyst quietly rebuilding workflows every quarter because the expectations keep changing.

The product manager trying to figure out whether they’re becoming more valuable or less valuable every time a new model gets released.

Nobody counts those hours.

Nobody tracks the psychological cost.

Yet.

That cost is real.

And I think it’s becoming one of the biggest labor stories nobody is talking about.

The second job nobody asked for

For most of modern professional life, staying current was relatively straightforward.

You learned your field.

You built experience.

You improved gradually.

The environment changed, but it changed slowly enough that experience remained valuable.

A software engineer who spent ten years becoming excellent at their craft could reasonably expect that expertise to remain useful.

A marketing professional could build deep knowledge of their domain.

An analyst could spend years mastering tools and methods.

The rules changed.

Not overnight.

But fast enough that people noticed.

Now a growing number of knowledge workers have two jobs.

The first job is the one they’re paid for.

The second job is staying employable.

Reading model announcements.

Testing new tools.

Watching what peers are doing.

Trying to determine whether a new capability is hype, meaningful, or existential.

Monitoring the labor market.

Monitoring the technology.

Monitoring themselves.

Most of this work happens after hours.

Most of it is unpaid.

And because everybody assumes everyone else is handling it fine, very few people talk about it openly.

Which creates a strange illusion.

Everyone looks calm.

Many people aren’t.

The adaptation treadmill

One of the most common pieces of advice right now is simple.

Adapt.

Learn the tools.

Stay current.

Keep up.

The advice sounds reasonable.

The problem is that adaptation itself has become labor.

Imagine telling a software engineer in 2015 that they would eventually spend part of every week evaluating new AI capabilities, deciding which mattered, rebuilding workflows around them, and wondering whether those same tools would eventually reduce demand for the work they spent a decade learning.

Most would have laughed.

Now it’s normal.

The expectation isn’t merely that workers perform their jobs.

The expectation is that they continuously reinvent how those jobs are performed.

Some people thrive in that environment.

Others don’t.

Neither response is irrational.

Human beings are not designed to operate indefinitely in a state of professional uncertainty.

At some point uncertainty stops feeling exciting.

It starts feeling expensive.

Something about engineers keeps bothering me

Software engineers are a useful example because they’re often presented as the winners of the AI transition.

They’re technical.

They understand the tools.

Many already use them daily.

From the outside that looks like a strong position.

The closer you get, the more complicated it becomes.

Many engineers are being asked to learn the tools that could eventually reduce demand for certain parts of their work.

They’re expected to do it quickly.

They’re expected to do it mostly on their own time.

They’re expected to become more productive.

Then they watch companies celebrate productivity gains and quietly wonder where those gains eventually go.

Toward higher compensation?

Toward shorter working hours?

Toward larger teams?

Or toward fewer people producing the same amount of output?

That’s not a comfortable question.

A lot of engineers won’t ask it publicly.

Many are asking it privately.

The emotional burden isn’t coming from ignorance.

It’s coming from understanding exactly what is happening.

The people who leave first

Most conversations about displacement assume workers stay in place until they’re replaced.

I don’t think that’s what happens.

I think a growing number leave before displacement arrives.

Not because they lost.

Because they’re tired.

Tired of rebuilding.

Tired of uncertainty.

Tired of constantly recalculating where value lives.

Tired of hearing every productivity breakthrough framed as good news while quietly wondering who benefits from the savings.

Over the next two years I expect this to become more visible.

Not through dramatic announcements.

Through subtle decisions.

People declining promotions.

People switching industries.

People accepting lower-paying roles with more stability.

People walking away from high-status careers because they no longer want to live on the adaptation treadmill.

That won’t show up in AI press releases.

It will show up in career decisions.

And by the time it becomes obvious, it will already be happening at scale.

I’ve spent months writing about who gets displaced next.

Lately I’ve found myself thinking about the people who are still employed.

The people still doing the work.

Still adapting.

Still learning.

Still trying to stay ahead.

A lot of public discussion treats employed workers as evidence everything is fine.

That feels incomplete.

You can be employed and exhausted.

You can be employed and anxious.

You can be employed and quietly wondering how long you can keep running at the current pace.

Employment status doesn’t tell the whole story.

Sometimes it barely tells any of it.

What I’d actually do

This is the part where many newsletters say to learn AI.

You already know that.

The people reading this are not unaware.

Most are already experimenting.

Most are already adapting.

Most are already paying attention.

The bigger question is where you’re positioning yourself.

Run a simple audit.

Forget job titles.

Forget tasks.

Forget the software you use.

Ask one question.

What breaks if you disappear?

Not what stops temporarily.

What genuinely breaks?

What outcome are you responsible for?

What decision relies on your judgment?

What risk sits on your shoulders?

The closer you are to accountability, the stronger your position tends to be.

The closer you are to repeatable process work, the weaker it tends to become.

That doesn’t guarantee safety.

Nothing does.

But it changes the game.

People obsess over protecting tasks.

The more useful move is protecting responsibility.

Tasks move around.

Responsibility tends to stick.

This week, identify one area where you can own an outcome rather than execute a process.

One project.

One customer problem.

One decision.

One result.

Move one step closer to accountability.

Then do it again next month.

The workers who navigate this period successfully won’t necessarily be the smartest people in the room.

They won’t necessarily be the fastest learners.

They won’t necessarily predict every technological shift correctly.

They’ll be the people who understand where human responsibility still matters and position themselves accordingly.

The people still standing in five years won’t be the ones who never felt the pressure.

They’ll be the ones who recognized it early enough to build differently.

The people still standing in five years won’t be the ones who never felt the pressure.

They’ll be the ones who recognized it early enough to build differently.

There’s a number missing from every AI productivity report.

Not the efficiency gains. Not the headcount reduction. Not the tools adoption rate.

The hours.

The hours software engineers are spending learning tools they didn’t ask for, in a job market that now treats those tools as table stakes, on a timeline nobody voted on, with no additional pay and no guarantee the tools they’re learning this quarter will still be the relevant ones next quarter.

That’s the tax. Nobody is adding it up.

What the productivity story leaves out

The official narrative goes like this: AI tools make engineers faster, companies get more output per head, everyone wins.

Here’s the part that gets quietly dropped from that sentence.

A study by METR, one of the few that actually tested experienced software engineers on real-world tasks rather than controlled demos, found that those engineers were 19% less productive when using AI tools. Not more productive. Less. The researchers noted that engineers spent significant time managing the AI's output, correcting errors, and re-doing work the tool had done wrong the first time. Time that wouldn't have been spent if they'd written the code themselves.

The demo always works. The production environment is where the actual experience of being a software engineer in 2026 lives.

And the actual experience looks like this: learning a new tool fast enough to list it on a resume before it stops being the relevant tool. Figuring out which part of the AI's output is wrong before shipping it. Performing fluency in an interview for a workflow you've been using for six weeks. Doing all of this while the job postings keep adding new tools to the requirements list, the tools keep updating their pricing models, and the companies keep announcing that AI is handling more of the work, while the engineers using that AI are somehow expected to be more productive, more adaptable, and more current than they were twelve months ago.

That is a tax. It’s just not called one.

The case study nobody is sharing

Here’s a specific version of how this plays out.

A senior engineer at a mid-sized SaaS company, eight years in, solid performer, trusted on the team, spent roughly four months in late 2025 learning Cursor, then Claude Code, then an internal agent workflow their CTO introduced in January. Each tool required real time. Not tutorials. Actual integration into daily work, figuring out where the output was reliable and where it wasn't, building the mental model for what to trust and what to check.

During that same period, their team was restructured. Two junior engineers were let go. The senior engineer absorbed a portion of their workload. No pay adjustment. The efficiency gains from the AI tools were cited in the restructuring announcement as evidence the team could operate leaner.

The engineer is still there. Still learning the next tool. Still absorbing the work. Still being evaluated partly on AI fluency in quarterly reviews, a metric that didn't exist two years ago.

The company’s headcount cost went down. The engineer’s workload went up. The tools that enabled the restructuring are the same tools the engineer is now required to demonstrate proficiency in to keep their job.

That’s the tax. Paid by the individual. Collected by the company.

The identity layer

Software engineers didn't just build a career. They built an identity around a specific kind of competence: the ability to understand complex systems deeply, to reason through hard problems, to know what the code actually does rather than what it's supposed to do.

That identity is under pressure in a way that’s hard to name precisely.

The market keeps moving the target for what counts as that skill. Moving it on a schedule nobody published, using metrics nobody agreed to, in service of an efficiency story that benefits the companies selling the tools more than the engineers using them.

The engineers most worn down right now aren't the ones who fell behind. They're the ones who stayed current, who adopted everything, learned the workflows, added the resume lines, and still feel like the ground keeps shifting under them.

That feeling is accurate. It’s pattern recognition, not anxiety.

I want to be honest about something.

The framing that bothers me most isn't "AI will replace engineers." It's the subtler version, the one where it doesn't replace them all at once, it just keeps raising the cost of remaining competitive until the math stops working for a specific group of people. Quietly. Without a single dramatic announcement.

I don’t have a clean resolution to that. I’m not sure one exists yet. But I think naming it clearly is worth something, even without the answer.

Why Some Engineers Aren't Worried

They own something specific.

Not a tools list. Not a certification in whatever the current platform is. A result. Something with their name on it when it breaks.

The distinction matters here: a process job executes known steps to produce a deliverable. An outcome job owns a result you’re held accountable for when something goes wrong. AI does the process work efficiently. It cannot be accountable for whether it worked. It cannot get on a call with an angry client at 11pm and explain what happened. It cannot make the judgment call about whether to delay the release or ship and patch.

The engineers navigating this most effectively have found a version of outcome ownership, a system, a relationship, a decision, that doesn't reduce to a prompt. They're not immune to the pressures described above. But they have leverage that shows up when the next round of budget conversations happens. Their name is attached to something that would visibly fall apart without them.

The ones most exposed are the ones whose entire value proposition is speed. Speed in a world where the tools get faster every six months is not a moat. Speed is what you have before someone builds a faster tool. Judgment is what you have after.

The three moves that actually matter

Not frameworks. Actual moves, this week and this month.

1. Find the thing the AI output gets wrong, and build the judgment to catch it.

The METR finding points at something real. Engineers who used AI tools without developing the ability to evaluate their output shipped problems they wouldn't have shipped before. The skill that survives this environment isn't generating code faster. It's knowing when the generated code is wrong before it costs something, before it hits production, before the client sees it, before the 2am call.

This week: take one thing the AI generated that you nearly shipped without checking. Understand specifically why it was wrong. That’s the judgment you’re building. It’s also the thing you can talk about in an interview that nobody else can fake.

2. Put your name on an outcome, not a workstream.

Owning a feature is not the same as owning what happens when that feature breaks in production for a specific customer at 11pm. One is a process. One is an outcome. The market is beginning to price that distinction correctly, even if the job descriptions haven’t caught up yet.

This month: identify one result, not a task, a result, that would fall apart without you specifically. A system that only you fully understand. A client relationship that runs through you. A technical decision that has your reasoning behind it. If you can't name one, that's the first problem to solve. Not the tools list. This.

3. Stop treating the tools list as the strategy.

By Q3, vibe coding is on 40,000 resumes. Cursor, Copilot, Claude Code, n8n, these are table stakes, not differentiators. The interview question that actually matters isn't which tools you use. It's what you did when the tool was wrong. What you caught. What you decided. What you own.

Build the story around that question. Write it down. Practice answering it with a specific example. That’s the actual differentiator in a market where the tools list has become the noise.

The honest version

The companies adopting AI tools are not thinking about what it costs the engineers to keep up with them. That's not cynicism. It's just not in their model. The efficiency gains accrue to the company. The adaptation cost accrues to the individual.

That asymmetry is the tax.

The engineers who navigate it best aren’t the ones who pretend the tax doesn’t exist, and they’re not the ones who refuse to pay it. They’re the ones who decide what they’re building while paying it. Outcome ownership. Judgment over speed. A result with their name on it.

The tools list is just the cover charge. The moat is what you can do that the tools can’t catch when they’re wrong.

You already know the ground is moving. The only question is whether you’re building on it or running across it.

It had a dramatic opening about the second wave of layoffs. A few data points about headcount math at Series B companies. A killer closer about calendar invites.

You’ve read that issue. I’ve written it nine times with different headlines.

This one is different. This one is just: here’s what I’d actually do, this week, if I were you.

The question nobody is answering honestly

Every piece of career advice right now falls into one of two categories.

The first is doom. Here’s how bad it is. Here’s the data. Here’s the pattern. Good luck.

The second is relentless optimism. Learn AI. Add value. Upskill. Position yourself. It’s full of phrases that mean nothing specific and require nothing specific.

I’ve been guilty of the first. The second makes me want to close the tab.

What’s missing is the middle thing. The honest, specific, slightly uncomfortable answer to: what do I actually do with this information?

That’s what this issue is.

Start with the honest audit

Before any move, you need to know what you’re working with.

The framework I keep coming back to, and have referenced enough that regular readers will recognize, is the process/outcome split.

Process work is executing known steps to produce a deliverable. Writing the report. Building the model. Drafting the copy. Running the meeting. AI does this. Not perfectly, not always, but well enough and getting better every quarter.

Outcome work is owning a result you’re accountable for when it breaks. When the campaign fails, when the product ships wrong, when the client churns, someone’s name is attached to that. AI assists with outcome work. It doesn’t own it.

Most roles are a mixture. The question is the ratio.

Sit down this week and actually map it. Take your last two weeks of work. List everything you did. Mark each item: process or outcome. Be honest. Don’t round up toward outcome because it feels better.

If you land at 70% process or higher, that’s not a crisis. It’s information. And information is what you make a plan with.

The three moves that actually matter

I’m not going to give you a framework with eight steps. Here are three things. Do them in order.

Move one: Own one result before Q3

Not a workstream. Not a project. A result with your name attached when it breaks.

This is harder than it sounds because most mid-level roles are deliberately structured to distribute accountability. That’s not an accident. It protects the organization and it quietly makes you easier to cut.

Volunteer to own the next thing that has a clear success metric and a clear failure state. The PM who owns the product launch metric is harder to eliminate than the PM who contributed to it.

In the UK, we’ve seen this play out clearly at the wave of fintech restructurings over the last 18 months. The people who survived weren’t always the most technically skilled. They were the ones with a result attached to their name.

Move two: Map your exposure honestly

Go through your role and identify the three tasks you do most often. Then ask: could a reasonably configured AI tool do an adequate version of this in six months?

Not perfect. Adequate.

If the answer is yes for two of the three, that’s where the risk sits. Not because you’ll be replaced tomorrow, but because when headcount pressure arrives, “this function can be partially automated” is exactly the language that shows up in the restructuring memo.

You want to know this before your manager does.

Move three: Find out who actually makes headcount decisions

This sounds obvious. Almost nobody does it.

Your direct manager has less influence over your continued employment than you think. In most companies above 50 people, headcount decisions run through finance, through the CFO’s model, through whoever owns the department P&L.

That person probably doesn’t know your name. They know a number.

This week, figure out who that person is at your company. You don’t need to do anything with that information immediately. But you need to know where the decision actually gets made.

I want to be honest about something.

Writing specific guidance is harder than writing analysis. Analysis lets you be right at the macro level while being vague about the individual. Guidance requires you to commit to something that might not apply to every person reading it.

I almost made Move One more hedged. “Consider owning a result, if that’s appropriate for your role and your organization’s culture.”

That sentence is useless. So I cut it.

Some of this will be wrong for your specific situation. That’s the trade-off. Vague advice that applies to everyone produces nothing. Specific advice that applies to most people and annoys a few is the more useful document.

What this is not

This is not “learn to use AI tools.”

That advice is not wrong exactly. It’s just so incomplete that it functions as wrong. Everyone is learning to use AI tools. The people who survive the second wave won’t be the ones who learned fastest. They’ll be the ones who figured out which part of their job couldn’t be handed off without something actually breaking.

That’s a different skill. Nobody is teaching it because it requires looking at your own role without flattering yourself.

This is also not “network more.” Networking advice in a displacement wave is like being told to make friends on a sinking ship. The relationships that will matter are the ones built on demonstrated judgment and real accountability, not LinkedIn connection requests.

The timeline question

People keep asking me when.

My honest answer: the second wave hits late 2026 into early 2028. The companies most exposed are the ones that announced AI restructuring between 2023 and 2025 without completing the actual deployment. They cut headcount on the promise of AI productivity. When the productivity numbers don’t close, they’ll cut again. Different language. Same math.

That gives you roughly two to three quarters. It’s enough time if you start now. It’s not enough time if you’re still deciding whether to start.

The X thread on this is running live at @MangoAggro —> I’m tracking the prediction publicly. If I’m wrong, I’ll say so there first.

This week

One audit. One result to volunteer for. One conversation to figure out who owns headcount at your company.

That’s it. Not a course. Not a pivot. Three things before Q3.

The ladder got pulled up. That’s real. But the people who build their own footing don’t need it to come back down.

You already know something is coming. The only question is whether you’re the person who did something about it before it arrived.

Graduates booed.

Not at a politician. Not at a controversial speaker. At the mention of AI. At commencement addresses that told them the future was bright while they already knew their entry-level job offers had quietly evaporated.

They weren’t being dramatic. They were being accurate.

And the people already in the workforce who watched those videos and felt something uncomfortable. That discomfort is worth paying attention to. Because what happened to that graduating class is a preview, not a separate story.

What actually happened to entry level

The entry-level job market for knowledge workers didn’t collapse all at once. It thinned. Over 18 months, the postings got fewer. The application-to-interview ratios got worse. The roles that did exist got more competitive, with more experienced candidates applying down because mid-level was thinning too.

A LinkedIn analysis found entry-level tech job postings dropped over 30% between 2022 and 2025. The roles didn’t disappear from the org chart. They just stopped being filled when they turned over.

Klarna is the cleanest example. They froze hiring, let attrition run, and watched headcount drop from 5,000 to under 3,500 over two years. No layoff announcement. No severance story. Just positions that stopped coming back. The company framed it as AI efficiency. The reality was simpler: they decided not to replace people who left, and AI gave them the language to make that sound like vision rather than cost-cutting.

That pattern of attrition plus hiring freezes, with no headline, is now the standard playbook. It’s cleaner than a mass layoff. No WARN Act exposure. No LinkedIn backlash. No CEO having to explain it on CNBC. The roles just quietly stop existing.

And the people who were supposed to step into those roles never got the chance.

In the UK the picture is sharper than most people realize. Graduate employment in tech roles fell for the second consecutive year. The roles that did open were increasingly demanding experience the candidates couldn’t have, because the internship pipelines had also thinned. A closed loop that nobody designed deliberately but that everyone contributed to, one small hiring decision at a time.

The pipeline problem nobody is calculating

Here’s what isn’t in any earnings call.

Senior engineers exist because someone let them be junior engineers first. Let them ship something slow. Debug something wrong. Own a small failure that was real enough to hurt but not catastrophic enough to end their career. That’s how the instinct gets built. Not through a course. Not through a certification. Through being wrong in front of people, repeatedly, until you develop the judgment that can’t be shortcut.

You cannot skip that process and expect to manufacture senior talent on demand five years later.

The companies that quietly eliminated their junior hiring over the last two years did not eliminate the need for senior talent. They eliminated the pipeline that produces it. The senior engineers those juniors would have become in 2028 and 2029 simply don’t exist yet. They’re not behind. They’re not underqualified. They just never got the entry point that would have started the clock.

I’ve been sitting with this longer than I expected to. There’s something genuinely uncomfortable about watching an entire cohort of people do everything right: the degree, the internships, the portfolio, the networking, only to arrive at the starting line and find the race reorganized while they were training for it. That’s not inefficiency. That’s a broken promise at scale. And the companies that made that promise implicitly, through years of graduate hiring programs and intern pipelines, mostly haven’t acknowledged breaking it.

The knowledge gap that creates is already locked in. You can’t un-do 2023 and 2024 junior hiring decisions in 2026. The compounding that would have happened is gone. The question now is what that means for the people currently in mid-level roles who are going to need that pipeline behind them and find it thinner than expected.

What mid-level should take from this

This is where most coverage of the entry-level story stops. It treats the graduate job market as a problem for graduates. It isn’t. It’s a leading indicator for everyone currently employed.

The logic that made entry-level roles dispensable, that this work can be done more cheaply and reliably by AI, is already being applied to the layer above it. Mid-level roles that are primarily about translating between teams, coordinating timelines, summarizing information upward, and executing defined processes are being evaluated by the same math that made junior roles disappear.

The difference is the timeline. Entry-level went quietly over 18 months. Mid-level will go the same way, slightly slower, because the coordination work is more complex and the relationships take longer to replicate. But the evaluation is already happening in quarterly planning meetings. Headcount reviews. Backfill decisions that keep getting pushed to next quarter.

The companies that cut junior roles didn’t announce a policy change. They just stopped posting when roles turned over. The companies evaluating mid-level right now aren’t announcing it either. The signal isn’t a memo. It’s a backfill request that keeps getting deprioritized. A reorg that leaves your role reporting to someone who can’t quite explain what you do. A skip-level that gets cancelled twice in a row.

Those are the signals. Most people are trained to wait for the explicit signal: the announcement, the meeting invite, the email at 6am. By the time the explicit signal arrives, the decision is already months old.

The process versus outcome audit, applied honestly

Most people reading this believe they sit on the outcome side of the line. The honest audit usually finds something more uncomfortable.

The test is simple but most people avoid running it because they already sense what the answer will be.

For every task in your last two weeks of work, ask one question: if this task had never been done, what would have actually broken in a way someone would notice and care about? Not “would someone have had less context.” Not “would the meeting have been slightly less prepared.” What would have broken. A product. A revenue number. A customer relationship. A system that someone gets paged for at 2am.

If the honest answer is “not much,” that task is process work. And process work is what gets compressed first, regardless of what it says on your job description.

The uncomfortable finding for most mid-level workers who run this audit honestly: more of the list is process than they expected. Summarizing for leadership. Coordinating between teams who could coordinate directly. Translating technical information into formats that AI now generates adequately. Attending meetings to represent a function rather than to change an outcome.

None of that work is worthless. It’s just compressible. And compressible is what the second wave targets.

The entry-level erasure as an opportunity

There’s a version of this story that ends with something actionable rather than just alarming.

The pipeline thinning created a real gap. Organizations that eliminated junior roles now have fewer people building the instincts that come from doing early-career process work: the debugging, the documentation, the small ownership moments that build judgment over time.

That gap is an opportunity for anyone willing to do deliberately what junior engineers used to do by default: own small failures, take accountability for things that could break, build the track record through repetition.

The move this week is specific. Find one thing in your organization that is being handled by a process rather than owned by a person. Something where if it degraded, someone senior would notice. Volunteer to own it. Not manage the process around it. Own the outcome, meaning your name is on it when it breaks at an inconvenient time, and you’re the one who fixes it.

That’s the thing AI can handle the execution of but cannot hold the accountability for. That’s where value is concentrating. Not in the doing of the work. In the ownership of what happens when the work goes wrong.

The window to claim that territory is shorter than most people think. The second wave isn’t coming for people who own outcomes. It’s coming for the layer that sits adjacent to them: the coordinators, the translators, the process managers who are useful but not accountable.

I post this kind of thinking on X in its raw form, before it turns into an argument, sharing the observations before they’re fully formed. If you’re not following there you’re seeing the finished product without the process. @MangoAggro

The entry-level erasure isn’t a story about graduates. It’s a story about what the workforce looks like in three years, who survives it, and how little time there is between now and then to choose which side of that line you’re on.

The ladder got pulled up. The people who build their own footing don’t need it.

There is a specific vibration in corporate Slack channels right now. It is the sound of leadership teams frantically trying to figure out how to frame their upcoming headcount reductions as “strategic pivots toward AI-driven efficiency.”

We are currently living through the Great AI-Washing.

If you have been in the tech industry for more than fifteen minutes, you know that corporate jargon is a shield. It is used to hide mistakes. And the biggest mistake of the last four years was the mindless, cheap-money hiring spree of 2021 and 2022. Companies over-hired because everyone else was doing it. Now, the bill is due.

But saying “we mismanaged our budget and hired three times as many middle managers as we needed” makes a CEO look incompetent. Saying “we are restructuring to leverage the transformative power of Generative AI” makes them look like a visionary.

The problem is that for the person sitting at their desk wondering if they have a job on Monday, this isn’t visionary. It is gaslighting.

Intelligence Analysis

The data is finally catching up to the theater. A new survey of over 1,000 hiring managers reveals a staggering gap between corporate PR and operational reality.

The Stat: 59% of companies are currently blaming AI for their layoffs because it “plays better with stakeholders and investors.”

The Reality: Only 39% of those same companies actually have an AI implementation strategy.

This means that a massive chunk of the “AI displacement” we are seeing in the news is an accounting trick. Companies are using AI as a financial shield to cover up structural failures.

Why does this matter? Because if you think you’re losing your job to a chatbot, you might feel resigned. You might think “the tech is just too good.” But if you realize you’re losing your job because your VP of Sales didn’t hit their numbers and needs a convenient excuse to cut 20% of the team, you realize the game is still about people.

We are seeing a repeat of the 2008 financial crisis “efficiency” narrative, but with a Silicon Valley coat of paint.

The Junior Death Spiral

The Massey University study released this week confirms where the pain is being felt most acutely: the entry-level roles.

These are the “gateway” positions. The roles where you learn how a company actually functions. The study shows that while AI isn’t necessarily replacing these people’s creative output yet, it is being used to justify not hiring them in the first place.

Management is making a short-term bet. They are betting that they can use AI to bridge the gap left by junior staff.

The result? Senior engineers and project managers are being turned into high-priced administrative assistants. They are spending 30% of their day doing the “grunt work” that used to be done by entry-level staff, under the guise of “AI-enhanced productivity.”

It is a productivity trap. You aren’t being more efficient. You’re just being used as a more expensive version of the person they just fired.

The Management Purge

Here is what happens next. This is my prediction for the remainder of 2026.

We are currently in Phase 1: The Scapegoat Phase. AI is the excuse for every layoff, every budget cut, and every missed bonus.

Phase 2 (Months 3-6 from now): The Productivity Plateau. Shareholders will start asking for the “AI gains” they were promised. If you fire 20% of your staff and claim AI will do the work, your remaining staff’s output should eventually skyrocket. But it won’t. Because, as the 39% stat shows, most companies don’t actually know how to use the tools. They just know how to say the words.

Phase 3 (Late 2026): The Management Purge. This is the big one. When the board realizes the “AI efficiency” was just a PR sticker on a failing engine, they won’t go after the workers again. They’ll go after the middle and upper management who sold the lie.

The people who “AI-washed” their departments without actually building a strategy are the most vulnerable people in the tech ecosystem right now. If you are a manager who has been using “AI” as a reason to cut your team but you haven’t actually spent 40 hours in the tools yourself, you are a target.

Are You Being AI-Washed?

If you are worried about your role, use this checklist to determine if your company is actually innovating or just managing their stock price through gaslighting.

The PR Gap: Leadership mentions “AI” at least five times in every all-hands meeting, but your department hasn’t been given a single Enterprise seat for a professional AI tool.

The Admin Leak: Your senior staff are spending increasing amounts of time on manual data entry or basic research that used to be handled by juniors who were recently “automated.”

The Budget Ghost: There is no line item for AI training, but the “marketing and PR” budget for “AI-driven initiatives” has doubled.

The Vague Mandate: You have been told to “look into how AI can save us time,” but nobody has defined what success looks like or provided any guardrails.

The Performance Shield: A department head who missed their KPIs by 40% is suddenly the biggest advocate for “AI restructuring.”

The Tool Lockout: Your company’s IT department has blocked all generative AI tools while the CEO is on LinkedIn talking about being an “AI-first organization.”

The Junior Freeze: All entry-level hiring is frozen “pending AI evaluation,” but the manual workload for those roles is still being distributed to existing staff.

The Consulting Influx: The company has hired a “Big 4” consultancy to “oversee the AI transition” instead of letting internal teams build solutions.

The Narrative Shift: Layoffs are announced as “rebalancing for the AI era” despite the company still having record-high manual backlogs.

The Strategy-less Stutter: When asked for a three-year AI roadmap, leadership provides a list of buzzwords instead of a technical deployment plan.

The Actionable

If your company checks more than five of these boxes, they are in the 59%. They are AI-washing their failures.

Do not try to save them. Do not try to be the person who “finally makes AI work” for a leadership team that doesn’t want it to work. They want the narrative of it working.

Instead, do three things:

Audit the Financials: Look at the company’s quarterly reports. Are they actually growing, or are they just cutting? AI-washing is almost always a sign of a stagnant or shrinking business.

Document the Manual: Keep a log of every “manual” task you are doing that is supposedly being handled by “AI.” When the audit comes (and it will in Phase 3), you want to be the one who can prove where the bottleneck is.

Look for the 31%: Start networking with companies that are actually building. These companies aren’t talking about “displacement.” They are talking about “deployment.”

They aren’t replacing you with a machine. They’re replacing you with a line item that looks better on a PowerPoint.

Don’t be the person left holding the bag when the board realizes the PowerPoint was a lie.

One gets sent to investors. One gets said at the all-hands. They are not the same document. Most workers are only reading one of them, and it’s the wrong one.

The investor version

When a CEO talks to Wall Street about AI, the language is specific and the math is explicit. Headcount efficiency. Cost per unit of output. Productivity multipliers. Reduction in force as a percentage of revenue saved.

These aren’t euphemisms. They’re line items. The investors need the real numbers because they’re making decisions with real money. The company can’t afford to be vague with them.

Salesforce told investors in early 2024 that AI would allow them to grow revenue without growing headcount. They did. Then they cut 700 people from their sales team, a team they’d said AI would augment, not replace. The augmentation story ran for 18 months before the restructuring story replaced it.

Block ran the same sequence. So did Duolingo. So did IBM, who announced 7,800 jobs would be replaced by AI two months after telling employees that AI would make their work more meaningful.

The pattern isn’t a coincidence. It’s a playbook.

The employee version

The all-hands version is different. It’s about opportunity. About how AI frees people up to do higher-value work. About collaboration between humans and machines. About the company’s commitment to its people during a time of change.

This version isn’t exactly a lie. The higher-value work does exist. Some people do move into it. The problem is that the all-hands version never specifies how many people the higher-value work requires and that number is always smaller than the current headcount.

The augmentation story and the displacement story are the same story. They just have different audiences and different timelines.

What the timeline actually looks like

The pattern has enough data points now to be predictable.

Phase one: AI assists the team. Productivity up. Morale decent. Leadership points to efficiency gains in earnings calls while telling employees this is about doing more, not doing it with fewer people.

Phase two: AI allows us to do more with less. Headcount flat. Some roles stop getting backfilled when people leave. The official explanation is “restructuring for agility” or “evolving our operating model.” This phase lasts 6 to 18 months.

Phase three: restructuring for efficiency. The announcement. HR has the script. The severance packages are already calculated.

Most companies in the tech sector are currently somewhere between phase one and phase two. The gap between phases is compressing. What took three years in 2020 is taking 18 months in 2026.

The honest read on your position

There’s a question worth sitting with this week. Not the comfortable version, the specific one.

Which phase is your company actually in? Not what the last all-hands said. What does the backfill rate say? When someone at your level left in the last six months, did the role get posted? Did it get filled? Or did the work get absorbed?

That answer is more accurate than anything your manager has told you about your team’s future.

The second question is harder. If your company is in phase one or two right now, what does that mean for your specific role? Your specific function. Your specific deliverables.

If most of what you produce is process work (steps executed to deliver a defined output) the math is not in your favor regardless of how well you execute. The companies cutting right now aren’t cutting bad performers. They’re cutting expensive process workers. Those are different categories.

The move, specifically

This week. Not someday.

Find the problem your company has that keeps coming back regardless of how many times someone addresses it. Not the task. The problem underneath it.

Then put your name on that problem. Not your deliverables against it, the problem itself. Own the diagnosis. Own the measurement of whether it’s getting better. Own the conversation about why it matters.

That shift from function to problem is the only positioning that survives what’s coming. Skills have half-lives. Problems don’t disappear. The person who owns a problem that keeps arriving is harder to restructure away than the person who produces a deliverable that AI can generate in thirty seconds.

You already know which one you are right now. The question is whether you move before someone else runs the math for you.

They don’t announce it when the math changes. They announce it when the decision is already old.

It is.

But it’s also doing something else at the same time:

It’s making fewer analysts necessary.

The mistake most people are making

They’re asking:

“Can AI do my job?”

That’s the wrong question.

The real question companies are asking is:

“How many fewer people do we need to do this now?”

And that shift changes everything.

The pattern already happening

A year ago, a typical workflow looked like this:

• Stakeholder asks a question

• Analyst writes SQL

• Builds dashboard

• Interprets results

• Presents findings

Now?

• Stakeholder asks a question

• AI generates query

• AI builds draft dashboard

• AI suggests insights

The analyst is still there.

But instead of doing everything…

They’re reviewing, correcting, and validating.

That sounds like leverage

And it is.

But it’s also compression.

Because once one person can do more…

You don’t need as many people.

This is where most analysts get it wrong

They think:

“I’ll just use AI better than everyone else”

But that’s not a moat.

That’s the baseline.

If your work mostly looks like this

Be honest:

• Writing SQL queries

• Cleaning data

• Building dashboards

• Generating reports

• Answering ad hoc questions

You are exposed.

Not because those tasks disappear…

But because they no longer justify multiple people doing them.

The uncomfortable shift

AI is absorbing the analysis layer.

And most analysts live inside that layer.

The split that’s already forming

There are now two types of analysts.

1. Execution Analysts (shrinking fast)

• Query writers

• Dashboard builders

• Report generators

• Data responders

They are still useful.

But there will be fewer of them.

2. Decision Analysts (becoming critical)

• Define what should be measured

• Frame the problem before analysis begins

• Influence product and business decisions

• Tie data directly to revenue, cost, or risk

These roles don’t just survive.

They become more valuable.

The real question you need to answer

Not:

“Am I good at analysis?”

But:

“Do I influence decisions — or just produce inputs?”

Most people avoid this question

Because they already know the answer.

And it’s uncomfortable.

What I would do if I were you

Stop doing this:

• Measuring your value by output (queries, dashboards, reports)

• Acting like a service layer for the business

• Waiting for questions instead of shaping them

Start doing this:

• Insert yourself earlier into decision-making

• Ask: “What decision will this drive?” before doing the work

• Tie every piece of analysis to a business outcome

Ignore this:

• “Learn more tools” advice

• Endless dashboard optimization

• Surface-level AI usage

That’s not where leverage is moving.

Immediate action (today)

Open your last piece of work.

Ask:

Did this change a real decision?

Not:

• Did someone read it

• Did someone like it

• Did it look clean

Did it change something?

If not…

You’re operating in the execution layer.

Structural move (next 3–6 months)

You need to reposition into one of these:

• Product analytics (closer to decisions)

• Growth analytics (direct revenue impact)

• Strategy / embedded roles (not support functions)

No middle layer.

No “reporting specialist” positioning.

That layer is shrinking.

The part no one will say directly

Companies are not planning to remove analysts.

They’re planning to:

Need fewer of them over time

And AI is how they get there.

Final reality

You’re not being replaced overnight.

You’re being out-leveraged gradually.

And most people won’t notice…

Until headcount starts to quietly shrink.

If you’re an analyst reading this:

You don’t need more information.

You need to decide:

Stay in execution → get compressed
Move to decisions → become necessary

You won’t be replaced all at once.
Just slowly… until no one notices you’re gone.

They start with silence.

No backfill.

No replacement.

No announcement.

Someone leaves. The role just… doesn’t come back.

At first it looks like discipline. Cost control. A temporary pause.

It isn’t.

It’s the first phase of removal.

The pattern nobody calls a layoff

In the last cycle, layoffs were loud and visible.

Headcount dropped. Then it came back.

This time is different.

Companies aren’t cutting first. They’re compressing.

The work still exists. The outcomes still matter. But fewer people are attached to them. The gap gets filled by a mix of AI tools, automation, and whoever is left on the team.

From the outside, nothing looks broken.

Inside, everything is tighter.

That’s why most people don’t see it until it’s too late. They’re waiting for a layoff signal that never comes.

The roles that disappear first

It’s not random.

It’s not “low performers” or “non-technical roles.”

It’s work that follows a pattern.

Work that can be described.

Work that can be repeated.

Work that produces a deliverable, not a result.

That’s process work.

Here’s what that looks like in practice.

Internal reporting roles

People who gather data, format it, summarize it, and send it upward. AI does this faster and without fatigue. The company doesn’t need more of it. It needs less.

Mid-level product management

Not all PMs. The ones translating between teams, writing specs, coordinating timelines. The work exists. The number of people doing it shrinks.

Support and operations layers

Anything that routes requests, answers known questions, or follows a playbook. AI doesn’t replace all of it. It removes enough that fewer humans are needed.

Translation roles

The layer between technical and non-technical. Turning one form of information into another. These roles feel important because they connect things. They’re also the first to be compressed.

None of these roles vanish overnight.

They thin out.

One person leaves. Not replaced.

Another leaves. Still not replaced.

Then a team that used to be five is three.

No announcement. No headline. Just less room.

The test most people avoid

You don’t need inside information to know where you stand.

Look at your last two weeks of work.

Be honest.

How much of it was:

- writing documents

- updating tickets

- summarizing information

- coordinating between people

- following a known process to produce a deliverable

If most of your time fits that list, you’re exposed.

Not because your work isn’t valuable.

Because it’s compressible.

Now flip it.

How much of your time was tied directly to an outcome that breaks if you disappear?

Revenue slows.

A system fails.

A product stops improving in a measurable way.

That’s different.

That’s harder to compress.

What actually protects you

Not your title.

Not your years of experience.

Not even how “technical” your role is.

What protects you is ownership.

Ownership of something that matters when it breaks.

The shift happening right now is simple.

Companies are moving from:

- people who execute processes

to:

- fewer people who own outcomes, supported by AI

If you stay on the process side, you compete with tools.

If you move toward outcomes, you use the tools.

What to do about it

This is the part most people skip.

They see the pattern. They feel the risk. Then they wait.

Don’t.

Start with this.

1. Attach your work to a measurable outcome

If you can’t point to a metric that moves because of your work, you’re floating. Find one. Tie yourself to it.

2. Reduce pure execution work

Anything repetitive, document-heavy, or coordination-heavy. Automate it. Delegate it. Shrink it. Don’t build your identity around it.

3. Move closer to failure points

Where does the company lose money, users, or reliability? Get closer to that. Those areas keep humans longer.

4. Make your absence felt

If you disappeared tomorrow, what would break? If the honest answer is “not much,” fix that before the company does it for you.

Most roles don’t get eliminated in a single decision.

They get made smaller until they don’t need to exist.

The people who leave first aren’t always the weakest.

They’re the ones attached to work that quietly stopped needing them.

The role doesn’t disappear all at once. It just stops coming back.

Not a meeting invite. Not a conversation with their manager. An email from “Oracle Leadership” sent before most of the country had finished their first coffee. By the time they tried to log in to ask a colleague what was happening, their systems were already locked.

That’s how it ends now. Not with a conversation. With a DocuSign link.

What Oracle actually did

Oracle is cutting up to 30,000 people - roughly 18% of its 162,000-person workforce - to free up an estimated $8 to $10 billion in cash flow. The money goes toward AI data center construction. Oracle has $553 billion in remaining performance obligations, almost entirely AI contracts. Revenue hit $17.2 billion last quarter, up 22%.

The company isn’t struggling. It’s redirecting.

The people who got that 6am email weren’t laid off because Oracle was failing. They were laid off because Oracle needed cash to build the infrastructure to fulfill contracts it had already signed. The headcount was the most controllable line item. So the headcount went.

TD Cowen analysts called it before it happened. Cutting 20,000 to 30,000 employees could generate $8 to $10 billion in incremental free cash flow. The math was always going to land here. The employees just didn’t know they were the math.

The part nobody talks about

After 20 years at a company, a generic email from “Oracle Leadership” is what you get. One manager wrote on LinkedIn that the cuts affected “some of the most talented, dedicated, and high-performing people I’ve had the privilege of working alongside.” He found out the same way they did.

People noticed their Slack acting strangely the day before. A glitchy white screen, an alphanumeric error code. They wondered if something was wrong. Nothing in their professional training prepared them to recognize that as the sign it was.

This is the thing I want to sit with for a minute. Not the number. The experience.

You go to sleep employed. You wake up to an email. Your access is already gone before you’ve had a chance to process what you’re reading. There’s no room in that sequence for dignity.

Why this keeps happening this way

Companies don’t tell employees in advance because advance notice is a liability. People download files. They contact clients. They call lawyers. The 6am email with immediate system lockout is not cruelty for its own sake. It’s a risk management protocol.

That’s somehow worse.

The people who designed that process were not monsters. They were HR professionals and legal teams doing their jobs, optimizing for a clean separation. The humans on the other end of it were a variable to be managed.

Some of the Oracle employees on the forums are noting the company may have violated the WARN Act - the law that requires 60 days notice for mass layoffs at qualifying sites. If Oracle classified affected workers as remote to avoid the threshold, those workers are potentially owed 60 days of back pay. That’s worth knowing. That’s worth asking an employment attorney about.

But I don’t want to skip straight to the practical. Because before the practical, there’s the human.

What I actually want to say

If you got that email Tuesday morning, or you got one like it from a different company in a different year, I want you to know that the speed and coldness of it is not a reflection of your value. It’s a reflection of how the system processes humans when the math changes.

You are not the math.

The fear that a lot of you carry right now, the checking the calendar, the reading the CEO memos too carefully, the noticing when a skip-level meeting gets cancelled - that fear is not paranoia. It is accurate threat perception. The people who felt safe at Oracle two weeks ago were not safer than you are. They just hadn’t gotten the email yet.

You’re not imagining it. And you’re not alone in feeling it.

What I know is this: the people who come through this wave intact are the ones who stopped waiting to find out from an email and started moving before the email came. Not in panic. Deliberately. Building the visibility, the network, the outcome ownership that makes you harder to reduce to a line item.

That’s not a pep talk. That’s the pattern in the data.

You deserve more than a 6am email. Start acting like it.

The email is already written. The only variable is when they decide to send it.

That’s the story nobody in the mainstream wants to say plainly. So let me say it.

A Duke University survey of 750 CFOs, run with the Federal Reserve Banks of Atlanta and Richmond, found something that should make you furious. Companies are reporting AI-driven productivity gains averaging 1.8% in 2025. When researchers checked those claims against actual revenue and employment data, the gains weren’t there. The CFOs believe AI is paying off. The numbers say it isn’t. Not yet.

Goldman Sachs found the same thing. Their senior economist noted that across the broader economy, there is still no meaningful relationship between AI adoption and productivity. Fewer than 20% of U.S. businesses are using AI for any core function at all.

And yet the headcount is gone.

The 59/9 gap

Here’s the number that deserves to be on the front page of every business publication. A ResumeBuilder survey of 1,000 hiring managers found that 59% of companies deliberately emphasize AI’s role in layoffs because it “plays better” with investors, media, and the public. Only 9% said AI had actually replaced roles at their company.

59% blame AI. 9% replaced by it.

That is not a rounding error. That is a 50-point gap between the story your company is telling and what is actually happening inside the building. Even Sam Altman, the person whose company made all of this possible, told Fortune that companies are engaging in AI-washing. Using AI as cover for cuts they were going to make anyway.

The company isn’t lying about AI making them more efficient. They’re lying about AI being the reason you’re gone.

Why this matters more than the layoff number

The standard framing is: AI is coming, jobs are going, adapt or die. That framing is real but incomplete. It treats displacement as a clean technological story. This isn’t that.

What’s actually happening is a three-layer cut. First, companies over-hired during 2020-2022. Second, AI gave them the political cover to reverse that without looking incompetent. “We’re not correcting a mistake, we’re investing in the future.” Third, the actual displacement is real but smaller and slower than the narrative suggests. The productivity gains haven’t arrived. The job losses have.

You are paying the price for a transformation that hasn’t finished arriving.

That’s the power story. Not the tech story.

Block cut more than 4,000 people, nearly 40% of its workforce. WiseTech Global cut 2,000. Atlassian cut 1,600. All framed as AI-driven efficiency. The Duke CFO study suggests a one-year lag between AI investment and realized revenue gains. Which means these companies cut the people before the AI was ready to replace them. They bet on the future. You absorbed the cost.

What the productivity paradox actually tells you

Economists are calling this Solow’s paradox, named for Robert Solow, who in 1987 observed that computers were everywhere but invisible in productivity statistics. The same pattern is playing out now. AI is everywhere in the announcement. Invisible in the revenue line.

This has two implications nobody is saying clearly.

First: the displacement wave is real but the timing is different than the narrative suggests. The companies cutting now are cutting ahead of the AI actually working at scale. The real displacement, when the productivity catches up to the promise, is still coming. The early cuts are the prelude, not the event.

Second: the gap between the narrative and the reality gives you a window. Right now, in most organizations, the AI is still in the pilot phase. Still being evaluated. Still being integrated. The roles being cut are the easy ones. The most process-heavy, the most clearly measurable. Roles where someone owns an outcome are being preserved because the AI can’t own that accountability yet.

The process/outcome divide is not a framework I invented. It’s what the data is showing. Process work is going. Outcome work is staying.

What to actually do

Pull up your last two weeks of work. List every task. Now ask: was I executing a known process, or was I accountable for a result that breaks when it goes wrong?

If most of what you do is the former, you are in the first wave. Not because your company said so. Because the math says so.

The window is still open. The productivity paradox tells you that most organizations are still in the lag phase. AI investment without AI results. That is your gap. Use it to move from process work to outcome ownership before the results close the distance.

The CFOs are optimistic. The economists are skeptical. The workers are gone.

Don’t wait for the revenue line to catch up.

The productivity gains are coming. Your job already left to make room for them.

That advice is six months out of date and the people giving it know it.

Prompt engineering was a skill in 2023 when you had to carefully construct inputs to coax useful outputs from early models. The models are better now. The skill of crafting prompts is being automated by the same tools that made it valuable. The job category that barely existed two years ago is already contracting.

So let’s talk about what’s actually happening to skills right now. Not what the LinkedIn influencers are saying. What the data shows.

The skills the data says are contracting

Computer programmers sit at the top of Anthropic’s exposure list. Not junior programmers. Not entry-level developers. Programmers generally. The research doesn’t soften this. Writing code, translating a requirement into working syntax, is the task AI has gotten best at, fastest. The job-finding rate for workers aged 22 to 25 entering AI-exposed occupations has fallen 14% since ChatGPT launched. That number is a leading indicator not a lagging one.

Customer service representatives are second. This one surprises nobody. Klarna’s AI handles the workload of 700 humans. That’s not a forecast. That’s a published number from a company that doesn’t need to impress anyone with it.

Data entry workers are third. Also unsurprising. What is surprising is that data analysis, the step above data entry, is following close behind. Not because AI can do analysis. Because AI has made the line between data entry and data analysis much harder to draw.

The University of Sydney published research this week confirming these three categories are most exposed. The researchers also confirmed something worth sitting with: even within the most exposed occupations, AI use is still limited. The mass displacement hasn’t arrived yet. What has arrived is the leading edge. The job-finding rate declining 14% is what a leading edge looks like in the data.

The skills the data says are appreciating

Judgment under uncertainty. The ability to look at something AI produced and know, not from a checklist but from depth, that it’s wrong. Not wrong in an obvious way. Wrong in the specific way that costs the company $200,000 if it ships.

This skill cannot be built by studying AI. It’s built by years of being in rooms where things went wrong, being held accountable for those failures, and developing the instinct that comes from having skin in the game repeatedly. AI is producing more output than ever. The humans who can evaluate that output accurately are becoming the scarcest resource in tech.

Stakeholder translation. The ability to sit across from someone who is frustrated and vague, understand what they actually need rather than what they said, and turn it into something buildable. AI can assist this conversation. It cannot replace the person who knows that “it feels slow” means “we’re about to churn.” That gap between human ambiguity and technical precision is still human work.

Accountability ownership. The willingness to own a result completely. To be the person whose name is on it when it breaks at 2am. This isn’t a technical skill. It’s a positioning decision. And it’s the one most tech workers are actively avoiding because it’s uncomfortable. The market is about to pay a significant premium for people who lean into it instead.

The contrarian read nobody is saying

Citadel Securities published a rebuttal this week arguing that actual AI adoption is slow, expensive, and nowhere near fast enough to meaningfully displace workers. They pointed to software engineering hiring being up 11% year on year.

They’re not wrong about the current data. They’re wrong about what the current data means.

Hiring up 11% for software engineers while companies simultaneously cut entry-level programmers from their payroll means the pyramid is changing shape. Not shrinking. Reshaping. Senior engineers who can own outcomes are in demand. Junior engineers who execute processes are not. The average is fine. The distribution is brutal.

The people reading Citadel’s “don’t worry” report and feeling reassured are the ones most at risk. Not because displacement is happening to everyone. Because it’s happening selectively. And the selection criteria are exactly the process versus outcome split.

What to actually do with this

The skills audit is the only honest starting point. Not “what new tool should I learn.” That’s a proxy move. The real question is where you sit on the process to outcome spectrum right now.

Pull your last two weeks of actual work. For each task ask: am I executing a defined process or am I owning an outcome someone cares about when it breaks?

If more than half your tasks are process work, the skill building that matters isn’t prompt engineering. It’s moving toward accountability. Find one thing in your organization that nobody clearly owns when it fails. Own it. Make the result your name. Do this before your company does the same audit and draws its own conclusions about your role.

The skills that are actually safe right now aren’t the ones everyone is talking about learning.

They’re the ones that require being wrong in front of people, repeatedly, until you build the instinct that can’t be taught.

The skills AI can’t replace are the ones that hurt to develop.

Most CEOs don’t say it that plainly. They use language designed to insulate them from the human cost of the decision. Dorsey said it out loud, which made him sound honest. It wasn’t honesty. It was strategy.

When a CEO says “AI is why we cut 40% of our workforce,” two things happen simultaneously. The stock goes up — investors read it as operational efficiency and bet on margin expansion. And the remaining employees get the message without anyone having to deliver it personally. You are next. The only question is when.

Atlassian just did the same thing. 1,600 jobs. Ten percent of the company. CEO Mike Cannon-Brookes said AI is reshaping how teams collaborate — which is, after all, Atlassian’s entire business. The CTO left the same day. The stock held steady. The market has already priced in the math.

The thing nobody is saying about these announcements

Anthropic published a study this week that should be making more noise than it is. The researchers introduced something they call “observed exposure” — a metric that compares what AI is theoretically capable of doing against what it’s actually being used to do in professional settings right now.

The gap is enormous.

AI is barely scratching the surface of what it’s technically capable of doing. The researchers found that actual adoption is just a fraction of feasible capability. Most organizations are still in early deployment. The tools exist. The infrastructure is being built. The workflows haven’t been redesigned yet.

Here’s what that means in plain language: everything that’s happened so far — Block, Atlassian, WiseTech, eBay, Pinterest — is the slow part. The companies cutting now are the early movers. They’re doing it while the majority of organizations are still in the “AI strategy” phase, still in the pilot program, still deciding which workflows to restructure first.

The displacement we’ve seen so far is not the peak. It’s the setup.

Who benefits from telling the story this way

Microsoft AI chief Mustafa Suleyman said this week that all white-collar work could be automated within 18 months. Anthropic CEO Dario Amodei has been saying half of entry-level white-collar jobs could disappear within five years. These are not independent observers. These are the people selling the tools.

There is enormous financial incentive to make AI sound more capable and more inevitable than it currently is. Every time a CEO cites AI as the reason for layoffs, it validates the vendors’ pitch. It generates demand. It moves stock. Block’s shares surged 15% the day the cuts were announced.

This doesn’t mean the displacement isn’t real. It is. But the narrative is being shaped by people who profit from the fear as much as from the technology. The workers losing their jobs are the proof of concept that makes the next sale easier.

Nobel economists Daron Acemoglu, Simon Johnson, and David Autor — the same researcher who documented the devastation of the China Shock on American manufacturing — published a warning this week. They called AI a “pure automation technology” that does the opposite of collaborating with workers. It commodifies human expertise. It renders it less valuable. Then superfluous.

These are not doomsayers. These are the people who built the frameworks we use to understand how technology reshapes economies. And they’re saying this time could genuinely be different.

The number that isn’t making headlines

75% of people displaced by AI never apply for unemployment benefits.

Some don’t know they qualify. Some expect to find another job quickly. Some are embarrassed. Some don’t understand the system.

The result is that every official statistic about AI displacement is an undercount. When economists say “the labor market is holding up fine,” they’re reading data that structurally cannot capture what’s actually happening. The people most affected by this wave are the least visible in the numbers used to assess its severity.

The real scale of what’s happening is larger than any headline suggests. It always has been.

What you actually do with this

The Anthropic study is the most useful thing published this week, not because it’s alarming — though it is — but because it gives you a framework for thinking about your own position.

The relevant question isn’t whether AI can theoretically do your job. It’s whether AI is currently being deployed to do the specific tasks that make up your job. The gap between those two things is your window.

If your role is heavy on process work — executing known steps, producing defined outputs, following established workflows — that gap is closing fast. The infrastructure being built right now is designed specifically for that category of work.

If your role involves owning outcomes, making judgment calls when things break, holding relationships that require trust built over time — that gap is still meaningful. Not permanent. But meaningful.

The move worth making in the next 90 days isn’t learning prompt engineering. It’s auditing your own role honestly. Pull your last two weeks of work. For each task ask: is this process or outcome? If more than half are process — the window is open now. It won’t stay open indefinitely.

The displacement is real. The narrative around it is being shaped by people who benefit from the fear. And the statistics are undercounting the actual damage.

You are not paranoid. You are paying attention.

The tools already exist. The organizations just haven’t finished deploying them yet.