There is a curious, private moment most of us have had in the last year. You are sitting at a keyboard, or in a meeting, and a part of the work that used to feel like yours slips away. It is not dramatic. It is not one meeting where you notice your role is redundant. It is a slow erosion: a draft you no longer need to write from scratch, a report that arrives half formed, a decision memo that gets to the right place faster than you could have justified the meeting.
You notice the time you used to spend producing has been carved out. You wonder whether that makes you safer, or suddenly less necessary.
The memo and the message
If that sounds familiar, you are living the transition Jack Dorsey made explicit today, 26 February 2026, when he announced Block would cut over 4,000 people: nearly half its workforce. Not because the business is failing. Because, in his words, “intelligence tools have changed what it means to build and run a company.” He chose a single deep cut over repeated rounds, and the stock surged 22 percent in after hours trading. That letter matters because it names the pattern we are all seeing: smaller, flatter teams doing more because they are pairing human direction with machine production. It is not a moral panic or a scandal. It is what operational change looks like when product economics, new tooling, and simpler organisational coordination meet in the same quarter.
You felt the effect before you read the memo. The memo simply made it explicit.
There is another framing doing the rounds: the Thiel line that AI will soon solve all Math Olympiad problems and that the real casualty will be the “math people.” There is a useful provocation buried there, but it is misleading as a career framework. The Olympiad framing conflates solving bounded problems with intelligence itself. That conflation is precisely what leads people to panic about the wrong things. This is not primarily about who can compute or who can write. It is about two capacities we should name explicitly.
Execution and judgement.
The split
Execution is what is being automated fastest. It is the relentless, repeatable cognitive work: drafting copy, generating code skeletons, running analyses, producing variants, answering routine questions. These are activities where pattern matching at scale, with high fidelity, is extraordinarily efficient. When you strip work back to its atomic pieces, a great deal of it is production: controlled, predictable transformations of information.
Machines are getting very good at those transformations.
Judgement is different. Judgement is taste, selection, prioritisation, the ability to see which version matters, and the courage to act on an imperfect understanding. It is the meta skill of asking the right questions and assembling the right constraints for a problem that is not yet well defined. It is what turns an array of plausible outputs into a coherent direction.
That is not immune to tooling. Judgement will be influenced, expanded and sometimes corrected by tools. But it is not reducible to pattern replication in the same way execution is.
This difference is why the math versus words frame is directionally right but incomplete. Math problems are tidy, bounded tasks. Language tasks are messier: they carry context, history and values. But the real axis of risk is not discipline. It is whether the role’s core value is turning raw information into reliable outputs, or making choices about what to produce in the first place.
If your job is mostly about turning up the volume on production, the machine will outpace you. If your job is about choosing which volume matters, the job changes. It does not vanish.
What moving fast actually looks like
I have spent the last year building systems that orchestrate intelligent agents for organisations navigating exactly this transition. The places that have moved fastest share a pattern: they treat tools as repeatable infrastructure for execution, and then invest the human time they save into better judgement and faster experiments.
That is what Dorsey meant when he wrote that “a significantly smaller team, using the tools we are building, can do more and do it better.” It is not mystical. It is operational: smaller teams with tighter feedback loops, where people spend less time creating and more time interpreting, aligning and deciding.
There is a subtlety here worth naming. The people who are thriving in these environments are not the ones with fixed good judgement, as if judgement were a permanent attribute you either possess or lack. They are the ones who continuously recalibrate their judgement against new evidence, including evidence generated by machines. The feedback loop between human decision and machine output is the real competitive advantage. You make a call, the machine produces options at speed, you evaluate the results, and your next call is sharper because of what you learned. That cycle, repeated daily, compounds into something no static expertise can match.
What nobody is ready for
That does not make the transition comfortable.
The skills that survive are the ones we have never learned to measure. Taste. Curatorial sense. The ability to hold competing hypotheses without collapsing into indecision. When a machine generates fifteen plausible strategies and your value is recognising which one fits the context nobody else has articulated, that is not a competency that fits neatly into a performance review. But it is the competency that matters.
Organisations are optimised for measured outputs, not judgement. Compensation and career ladders favour production metrics, which will skew incentives as output becomes cheaper.
The person who ships four deliverables a day will be rewarded by the old system. The person who identifies which single outcome actually moves the business will be rewarded by the new one.
We have not yet built the structures to recognise the second kind of value. And until we do, the people doing the most important work will be the hardest to evaluate by the metrics we currently trust.
The window
How quickly? Dorsey himself wrote today: “Within the next year, I believe the majority of companies will reach the same conclusion and make similar structural changes.” From where I sit, the practical window to reshape your role is about 18 to 24 months. Not in rhetoric but in daily tasks. That is short enough to force choices, but long enough to act if you are deliberate.
Early adopters will crystallise new roles, new workflows and new standards of what “good” looks like. Once those standards are baked into how customers and markets evaluate work, catching up becomes harder than starting from behind. Those who practise judgement in public, who make decisions with machines and iterate visibly, will set the new norms. Those who delay will be judged by outcome metrics set by early movers, not by intent.
The reframe
A useful reframe for anyone worried: stop asking “is my job safe?” and start asking “what decisions do I make that AI cannot replicate?”
Break that down. List the things you decide where no single output is obviously correct: trade offs you tolerate, constraints you enforce, audiences you intuitively understand. Then spend your reclaimed time sharpening those decisions. Practise arguing for them, expose them to disconfirming evidence, and make them accountable in measurable ways.
Use machines to produce options quickly. Use the time you save to test, measure and choose among those options. The ability to choose well is the new scarce resource.
If you are an organiser: product manager, partner, lead designer, founder. Your most valuable work becomes a mixture of curation and orchestration. Curate the signals you trust. Orchestrate the right people and tools toward a single judgement.
If you are a maker: engineer, writer, analyst. Your most valuable skill becomes the tasting of your own outputs. Can you tell the one variant that will nudge behaviour versus the ten that will not? That is not glamorous. But it is where influence is reallocated.
Start now. Start differently.
Here is the tangible thing most people are not doing: learning to work alongside AI. Not the chatbot version. Not the “ask it a question and get an answer” version. That era is already behind us.
What is emerging right now is agentic AI: systems that do not just respond but execute. Claude CoWork sits on your desktop and works through tasks autonomously. Perplexity Computer, released this week, operates your machine on your behalf. OpenClaw orchestrates multi step workflows. Manus builds and deploys. These are not tools you type into. They are digital colleagues that take direction, execute independently, and return completed work for your review.
Read that again. They take direction, execute, and return work.
That changes the skill you need to build. You are not learning to write better prompts. You are learning to manage, direct, and evaluate a new kind of collaborator. Think of it less like using software and more like onboarding a remarkably fast, endlessly patient colleague who happens to be extraordinarily good at execution and needs your judgement to do the right work.
Start here: ask an AI agent to interview you. Tell it your role, your industry, your challenges. Let it ask you questions about where you spend your time, which tasks drain you, which decisions only you can make. It will surface patterns you have not noticed. It will organise your own thinking and hand it back to you as a set of instructions for how it can help. That single exercise will teach you more about the execution versus judgement split in your own work than any article, including this one.
Then go further. Give it a real task. Not a test. A task that matters. Let it produce the first draft, the first analysis, the first ten options. Do the work that comes after: evaluate, select, refine, decide. The muscle you are building is not how to use AI. It is how to lead alongside it. That is a skill. It compounds. And right now, almost nobody is training it deliberately.
Where you stand tomorrow
Some roles will evaporate, and transitions will be messy. That is the honest reality. But the tools free time. The scarcity is how you use that time. Practise judgement publicly, codify your decisions, and make selection the work you are judged on. That is how you stay relevant in a world where production is cheap and decision making is the currency.
Where you stand tomorrow will be the result of the choices you make now.
If this landed, subscribe. The next piece unpacks how to build the daily practice of judgement in a world of infinite machine output, and why the people who master that practice will define what “good” looks like for the next decade.
Craig Hepburn is an AI strategist and builder, Perplexity Fellow, and former Chief Digital Officer at Art Basel and UEFA. He works across technology, business, and system design to shape how AI operates responsibly in the real world.
Deeply encouraging. Well done!