Anthropic just dropped a report on AI and labor markets. It’s useful. It’s also going to get misread by almost everyone who has a career to protect.
Here’s the headline most people will pull: “No systematic increase in unemployment for highly exposed workers.”
They’ll exhale. Put it in a deck. Move on.
Wrong move.
The report introduces something called “Observed Exposure” — a blend of what AI can theoretically do, and what it’s actually being used for in real workflows. In fields like Computer Science and Math, they estimate 94% of tasks are theoretically feasible for AI to perform. Current observed usage? Around 33%.
That 61-point gap isn’t comfort. That’s a loaded gun.
The signal you’re missing is in the hiring data.
They find no unemployment spike. But bury the lead much?
The same report flags a canary: hiring of younger workers into exposed occupations has slowed. For workers aged 22–25, there’s an estimated 14% drop in job-finding rate in highly exposed roles. Barely statistically significant — but the direction is ugly.
Displacement doesn’t announce itself with pink slips. It announces itself with silence. Fewer entry-level openings. A req that gets reposted and then quietly closed. A team that doesn’t backfill.
Oh — and the February jobs report just came in 150,000 jobs off from estimates. That’s not noise. That’s signal.
The silence is already here.
The real story isn’t capability. It’s context.
Yann LeCun — Meta’s chief AI scientist — has been publicly critical of GPT architecture limitations for years. He’s not wrong about the architectural constraints. But here’s what the critics missed: you don’t have to solve architecture when you solve context.
Proprietary data — your CRM, your support tickets, your customer behavior, your pipeline signals — fed directly into these models makes them remarkably more intelligent. Not because the model got smarter. Because it finally knows what you know.
This is the dimension shift people aren’t pricing in.
Think about what open source did to software competition. It didn’t make proprietary software disappear — it added an entirely new dimension of competitiveness that changed the rules for everyone. MCPs (Model Context Protocols) do the same thing for AI deployment. They’re not a feature. They’re a new axis.
When any company can connect a model to their internal data stack in days instead of months, the moat isn’t architecture anymore. It’s context. And context is suddenly available to everyone.
Here’s why this matters more than you think.
Until very recently, somewhere between 65 and 85% of every AI project budget went into data engineering. Getting data clean enough, structured enough, accessible enough to even run the model. That was the wall.
That wall is coming down.
When integration friction collapses — and it is collapsing — deployment goes from “pilot project” to “this is how we run the business.” The constraint shifts from can we build it to do we have the nerve to deploy it.
Companies with nerve will deploy. Their competitors will follow. The ones still running on gut feel and quarterly slide decks won’t see it coming.
Now here’s the part nobody’s saying.
I’ve been thinking about this enough to lose sleep over it. Not because it scares me — because it excites me.
Once you get through the five stages of grief about your job never being the same again — and it is already gone, I’m sorry — the doors start to open.
Automating drudgery doesn’t just save time. It forces you to think differently. You stop asking “how do I do this faster?” and start asking “should this exist at all?” You stop minimizing waste and start building decision pipelines. You start designing self-healing feedback loops that generate capacity — actual new capacity — instead of just plugging holes.
The software does the drudgery better than you did anyway. It doesn’t complain. It doesn’t have an off day. It doesn’t forget the step you added in Q3.
And when you’re free from it? You get to do the thing that AI still can’t: make the call. Connect the dots. Build the relationship. Decide what matters.
The best marketing right now is a product that genuinely does something remarkable. I find myself more energized about that than I have been in a while. That’s not a PR line — that’s just what happens when you actually have time to think.
What to do with this.
If you’re in one of these exposed roles: get curious before you get scared. The question isn’t will my job change — it already has. The question is what do I build next.
If you’re leading a team: stop treating integrations as plumbing. Every system connection you build is compounding. Every one you delay is compounding for your competitor.
And if you’re still waiting on a pilot to prove out ROI — the Anthropic report is telling you the gap between capability and deployment is enormous. First movers in deployment win. Not first movers in prompting.
Traffic stopped a few exits back. Get your operating system ready.
Hit me up if you want to talk through it — this stuff is genuinely fun and fascinating once you stop fighting it.
