You have heard the story. AI is coming for your job. The company you work for, or the one you interviewed with last month, has an AI strategy document somewhere with a number attached to your role. Maybe 20%. Maybe 40. The exact figure does not matter. What matters is the assumption underneath it: a machine can do the work you do, and the machine costs less than you.

This is not paranoia. Companies are running the numbers. One of them ran them harder than anyone. And this week, we found out what happened when the arithmetic met reality. The company with the most resources, the most data, and the most reason to believe the plan would work discovered that it did not.

The Company With Every Advantage

The company is Meta. The plan had a code name: Project OT, short for Organization Transformation. According to Reuters, who reviewed internal planning documents and spoke to more than twenty people familiar with the company, it was drawn up in January at Mark Zuckerberg's annual leadership retreat at his Hawaii compound. The vision was an "AI native" future. AI tools and agents would interact, workflows would automate, and new work would be built around AI from the start. The human side was blunter. In the most aggressive scenarios, some teams could shrink by as much as 60%. What remained would be small, "talent-dense" groups of humans overseeing virtual workers.

The restructuring was meant to arrive in two waves. One in May. Another in November. The first wave went through. Meta cut about 10% of its workforce, close to 8,000 jobs, and moved 7,000 workers into AI workflow initiatives. It left 6,000 open roles unfilled. These are not small internal tweaks. They are the moves a company makes when it believes the org chart has to change before the product roadmap can.

On the night of May 19, hours before the second wave was due to land, Zuckerberg pulled the plug. The November wave died before leaders had worked out how many people would lose their jobs. The most aggressive version of the plan, the one designed to prove that agents could replace humans at scale, stopped.

The Gap Between Motion and Value

The reason it stopped is the part worth your attention. It was not public pressure, not a change of heart, not a rule from a regulator. It was the numbers.

An internal post from Meta's CTO in June said code changes to internal platforms were up 220% year over year. Changes that reached users as new or improved features were up 36%. Read that twice. The activity more than tripled. The output scarcely moved. More code, not more shipped value. More motion, not more product.

Anyone who has worked near software knows the difference. More code can mean a better product. It can mean more review, more bugs, more clean-up, and more meetings about why the system did something nobody asked it to do. At Meta, the internal data confirmed the darker version. Posts tied unchecked AI agent activity to large-scale, disruptive actions humans would not take. Major technical and security incidents rose 40% from a year earlier. Employee time spent resolving those problems rose as much as 70%.

That is not a gain. That is a new workload dressed up as automation. The tool produced more. The production produced more problems. And the humans, the ones the plan aimed to make optional, spent more time than at any point before cleaning up what the machines could not clean up themselves.

The Layers That Refused to Disappear

Months later, something happened that the Project OT planning documents did not see coming. Meta asked some engineers in its Applied AI division to return to management roles. Not a reshuffle. A reversal. The layers that were meant to vanish, the management structure the "talent-dense" vision aimed to cut, got asked to come back.

The management. The review process. The judgement about what gets shipped and what gets cut. The dull human work of deciding whether an output is good enough to put in front of a customer. None of it disappeared because an agent could write code. It turned out the agent could write code and someone had to decide whether the code was worth anything. The layers did not vanish. They proved to be the point.

This is the company with more compute, more researchers, and more cash than anyone trying this. It raised its 2026 capital expenditure guidance to as much as $145 billion. And with every advantage that money can buy, it could not make the layers go away. The question that should travel beyond Meta is simple. If they could not do it, who can?

What the Hiring Data Says About Why

A report this week, from a recruitment platform that assessed 683 skill groups across 92 employers, asked a question that reframes the conversation. Not "can AI do this?" For most tasks, some version of it can. The question was different: which parts must a human do alone, and which can they direct and check?

The headline figure: 72% of the skill groups they assessed were in full or in part delegable to AI. That sounds like the replacement story. It is not. The breakdown is where it turns. Analytical reasoning and problem-solving: 90 to 100% delegable. Operations and execution: 28%. Leadership and interpersonal skills: below 50%.

The skills Meta could not cut, the management and review and judgement that got asked back four months after the cuts, are the same skills the hiring data says AI cannot replicate. And employers know it. Across 113 leadership scorecard templates from 92 employers, emotional intelligence criteria averaged 52% of total scorecard weight, edging out every other criterion combined. The hiring question has shifted from "can this person use AI?" to "can this person do the parts AI cannot?"

The recruitment platform's founder put it in plain terms. Foundational knowledge has not gone away. It is what lets someone catch an AI's mistake. The skill that matters is not the ability to generate output. It is the ability to know whether the output is right. And that ability does not come from the tool. It comes from the person.

The Question the Reports Are Pointing At

Meta is not the whole story. It is the most expensive test case. The company spent billions, ran the experiment at scale, and discovered that the work AI can do and the work that makes a company function are not the same work. The code went up. The features scarcely moved. The incidents climbed. And the people management aimed to make optional turned out to be the ones who made the output usable.

The hiring data says the same thing from the other direction. The skills that sit at 90% on AI's delegability index, the analytical and problem-solving tasks, are the ones everyone has access to. The skills that sit below 50%, the leadership and interpersonal and operational work, are the ones employers weight most when they hire. The gap between what AI can do and what employers pay for is the gap where careers compound.

This is not a case against AI. The same week Meta's reversal became public, companies were announcing new AI agents designed to take on complex work over days and weeks. The transition is real. But the data, from both sides, points to the same conclusion. AI changes the job. It does not replace the parts that matter. The people who build the skills that resist replacement, the ones below 50%, are the people who stay valuable when the restructure comes. The people who build the skills that sit at 90% are the ones the spreadsheet was modelling.

See Which Skills Sit Below the Line

We mapped this. The Skills Economy 2026 is our free report, drawing on sixteen sources across five regions: the skills employers are hiring for, the ones disappearing from job ads, and the gap between what AI can do and what employers pay for. If Meta's numbers tell you anything, it is that the question of which skills matter is no longer theoretical. It is the question your next restructure will answer, with or without your input.

Download The Skills Economy 2026 report — free. Read it before your next restructure does the maths for you.

And If You Want the Structured Path

The report shows you the landscape. If you want to build the skills the data says AI cannot replicate, our course range is where that happens. A short lesson in your inbox each day, built for working professionals, no technical background assumed:

• Emotional Intelligence at Work — the human skill that sits below 50% on AI's delegability index and above everything else on leadership scorecards.
• Leading Without Authority — the management capability Meta tried to cut and asked to bring back.
• Async-First Productivity — the capacity for focused work that AI multiplied the noise around but could not replace.

The company with every advantage could not replace its people with AI. The question is whether you are building the skills that made them irreplaceable. Start with the report.