Every model release quietly resets part of what you know. The prompt that pulled a brilliant answer out of 4.2 can wreck the response in 5.5. Each model has its own specialisation, its own quirks, its own idea of what a good instruction looks like — so the technique you sharpened last year is already going dull. The models adapt, and so must you.

Which leaves an honest question sitting on the table: if what was impossible last month is routine today, and what I learned yesterday is redundant now — what do I actually know that stays valuable? And if the answer is "not much," what does that do to how we build the organisations we work in?

Hold that second question. It's where this is going.

What Actually Stays Scarce

Stay on top of the technology, yes. But the durable edge lives upstream of any model: the ability to define a problem cleanly and strip out ambiguity. Cassie Kozyrkov put it well across her latest videos — such as rest in peace, prompt engineering. A good prompt was never a skill in its own right. It was a symptom of someone who had already done the hard thinking. One model update and the "prompting edge" is gone, because the thing that made the prompt good was never the prompt. The model itself is becoming a commodity, and what gets commoditised is everything most humans can do reliably at scale. Right now that feels dazzling. Dazzling isn't the same as scarce.

So where does that leave experienced people? My own read — and here I'm arguing, not citing — is that seniority holds its value not because senior people are better with the tools, but because of what the tools still can't reach: the relationships built over years, the context of the business, the undocumented rules and guardrails that live inside an organisation and never make it into a document. That's what experience is. A model can draft the memo; it can't tell you which of three stakeholders will quietly kill it.

The Job Quietly Changed

This shifts what the work actually is. Where you used to lay out every step for the machine to follow, now you state the target, the constraints, and the problem — and leave room for the model to work out how. You stop managing execution and start setting direction.

Which is, more or less, the job of a leader.

Good leaders already do this. They know which problems are theirs to own. They read strategy and intent well enough to judge which problems have the highest chance of real impact. Then they hand a hard, well-framed problem to a team that executes. So here's the uncomfortable version of the question: with this technology, are we all leaders now — each with a world-class execution team on tap?

If We Are, the Bottleneck Moved

If everyone commands an elite execution team, execution stops being the scarce thing. The scarce thing becomes the right to decide what gets executed.

In traditional organisations, that right is bolted to titles and budgets. The person sitting on the problem has to pitch it upward — scope it, sketch a solution space, and then wait while it loops through the org until someone with the authority rules on what to focus on and who does the work. That loop was tolerable when execution was the slow part. It isn't anymore. Now the decision cycle is the slow part, and it's strangling teams that could otherwise move.

So the question I keep turning over: how do you rebuild an organisation so decision-making runs at the speed execution now runs at — without losing safety, scale, or reliability? Because when the execution power sits in the models and the problem-context sits in the people, progress still stalls the moment every real decision has to climb all the way up to the CTO or CIO.

My bet is that the organisations that win this shift are the ones that can make direction, strategy, and ownership explicit — clear enough that people, agents, and tools can all act on them without stopping to ask permission. That's harder than it sounds, and a little frightening. It means taking the implicit guidance system that currently lives inside leaders' heads and putting it somewhere both humans and agents can reach.

Things are changing fast. So must the way we build organisations to run.