Articles · The Shift
9,000 AI Jobs Cut in 5 Weeks. The Stock Price Went Up.
3 April 2026 · 10 min read
In the last five weeks, four companies cut nearly nine thousand jobs. Not because they were failing — revenue was up. Because AI made those roles unnecessary. Block's stock went up eighteen percent the day they announced.
Block — four thousand people. Forty percent of the company. Revenue was up twenty-four percent. Atlassian — sixteen hundred people, over nine hundred from engineering and R&D. WiseTech Global — two thousand people, twenty-nine percent of their entire workforce. Meta — seven hundred more, plus billions shifted into AI.
Every board in every industry is watching. The question is not whether this is coming to your organisation. The question is: who are the people who stay?
The CEO said they were hiring. Five months later, they cut 1,600.
In October 2025, one of these CEOs went on a podcast and said — quote — “Five years from now, we'll have more engineers working for our company than we do today.” He said they were hiring more graduates than ever. Five months later, he cut sixteen hundred people. Nine hundred from engineering. The CTO left the same day.
This is moving so fast that even the people running these companies cannot see more than a few months ahead. If you're waiting for your company's leadership to warn you, that warning may arrive as a restructuring memo.
Jensen Huang's line in the sand
NVIDIA's CEO said one sentence that cuts through everything: “If your job IS the task, you're highly going to be disrupted.”
He is drawing a line. On one side — process work. Data entry. Scheduling. Report formatting. Compliance checks. Work with clear inputs and clear outputs. If that is your entire role, you are on the wrong side of the line. Block just eliminated four thousand roles of it.
On the other side — the judgment layer. Strategic direction. Problem definition. Client relationships. The knowledge of what good looks like. AI cannot replicate this.
AI has been superhuman at reading medical images since 2020. And there are more radiologists working today than in 2020, not fewer. Because the job was never “read the image.” The job is diagnosis — patient history, judgment calls. The reading is the task. The diagnosis is the judgment layer. AI replaces the task. The people who only did the task are the ones who are vulnerable.
The 47% premium
Anthropic published a study that found something most people don't expect. People earn forty-seven percent more when they work alongside AI and apply their own judgment to what it produces. Not younger workers. Older ones. People who have enough experience to know when the output is right — and when it is not.
The Dallas Federal Reserve published a framework that draws the line clearly. Process work is replicable, teachable, documentable. It can be extracted from a person and turned into a checklist. It can be automated. Experience — knowing what to look for, knowing what feels wrong before you can prove it, understanding what happens downstream — cannot be written down fully. It cannot be automated.
When Block went from ten thousand to six thousand, which people were let go? The ones doing process work. The ones whose jobs could be written into a checklist. The value doesn't disappear from the organisation. It concentrates. Fewer people. Same revenue. The ones who stay become dramatically more valuable.
The Catch
BCG and Harvard studied what happens when people use AI at work. The ones who just accepted whatever AI gave them, without questioning it, performed twenty percent worse than people who didn't use AI at all. Using AI without judgment makes you worse, not better.
Three skills that determine who stays
The people who stay have three things. Not technical skills. Not AI certifications. Three skills that only come from experience — and that AI makes more valuable than ever.
1. Specification skill
The ability to describe exactly what needs to be built. Jensen Huang said: “The definition of coding is now specification. We just went from thirty million coders to probably one billion.” Being able to describe a problem clearly, spell out what you need, and explain what good looks like — that is now the most valuable skill in the economy. Requirements documents. Project briefs. Problem statements. RFPs. You've been doing this your entire career. You just didn't know it had been renamed.
2. Interpretation judgment
The ability to look at what AI produces and know when it is wrong. You cannot learn this in a weekend workshop. It comes from years of seeing what good looks like and what failure looks like. AI can produce a perfect-looking financial model, legal brief, or strategic plan. Only someone who has seen a hundred real ones knows when the perfect-looking one is wrong. Google's head of research said it directly: when he hires, he's not looking for technical skills. He's looking for judgment.
3. Deep domain knowledge
The experience that tells you whether the specification is good or garbage. AI can produce an answer to any question. Only someone who has lived in that field for twenty years knows whether you even asked the right question. This cannot be downloaded. It cannot be fast-tracked. It cannot be automated. The Dallas Fed calls it tacit knowledge. Process work can be copied. Experience cannot. What you've built over decades — that is not a liability in the age of AI. It is the advantage.
Domain depth multiplied by AI capability equals real leverage. That is the formula. And the depth part — you already have it.
The window is open now
Jack Dorsey said most companies will reach the same conclusion and make similar structural changes within the year. The restructuring memo does not check which sector you work in. Finance, operations, legal, marketing — it is coming to every department in every industry.
It is better to hear this now — clearly, calmly, with the research in front of you — than in eighteen months when the memo arrives and you haven't positioned yourself.
The next time you delegate something to AI, ask yourself: could I explain to someone with no context in my industry exactly why this output is right or wrong? If you can — that is your experience in action. That is the thing that cannot be automated. If you can't — that is the gap to close. And closing it is not a technology problem. It is a depth problem.
The window to build these three skills — specification, interpretation, domain knowledge — is open right now. It will not stay open forever.