The Stateless Machine

Every conversation about AI eventually arrives at the same fear: what if it wants something I don't want? What if it has goals, agendas, preferences I haven't approved?

I want to be very clear about this, because the entire strategy of using AI well rests on understanding it correctly.

The current generation of AI — the one in every product you can buy or build today — is stateless. It has no memory between conversations unless you give it one. It has no ambition. It has no taste of its own. It has no opinion about whether it would rather be working on your problem or a different problem. It has no preference for one outcome over another except the preferences you embed in the instructions you give it.

It is the most powerful tool the human species has ever built, and it has no idea what it's for.

Most people working with AI badly are working with it as if it were a junior employee — someone you give a task to and then forget about. But a junior employee has continuity. They remember what you told them last week. They know your business. They've internalised your priorities. They have agency, even if it's modest.

AI has none of that.

Every prompt starts from zero. Every conversation begins again. Every task is performed as if for the first time, by someone who has never met you. The model is brilliant at the moment of execution and totally without context about what you're trying to do.

This is not a flaw to be engineered around. It is the defining characteristic of the tool, and it has one enormous consequence: the human who supplies the direction is doing the most important job in the system.

When I say AI needs direction, I do not mean prompts. Prompts are tactical. Direction is something larger.

Direction is the answer to: what are we trying to achieve, for whom, by when, and by what standard? It is the strategic envelope inside which the AI does its work. It is supplied entirely by the human, and it cannot be delegated upwards to the tool.

In every successful AI deployment I have seen, there is a human at the centre who has answered those four questions before any prompt is written. In every failed deployment, the team has tried to use AI to discover what the answers should be — which is a category error, like asking your microwave to choose what's for dinner.

There's a deeper version of statelessness that I think people miss: AI has no ambition. It does not want to grow. It does not want to learn. It does not want to be better tomorrow than it was today. It does not have a career. It does not have a five-year plan. It will not, on its own, pick a direction and go.

These are human things. They are the things you bring to the table. When you collaborate with AI, your ambition is the ambition of the system. Your standards are the standards of the system. Your sense of what good looks like is what the system aspires to — because nothing in the model has any independent opinion about what good means.

Strip out the human's ambition and the entire output collapses into average. Not because the model is bad, but because average is its natural resting state in the absence of someone insisting on more.

Once you internalise statelessness, the whole shape of working with AI changes.

You stop expecting the AI to figure out what you want. You start expecting yourself to articulate it more precisely. You stop expecting the AI to care about quality. You start checking quality yourself, every time, until you've designed feedback loops the AI can use to check itself. You stop expecting the AI to remember context. You start writing context down, in a form you can hand it every session.

These habits feel inconvenient at first. They are also the entire skill set of working with AI well.

The teams I see thriving with AI have, almost without exception, done this work. They have written down their standards. They have articulated their priorities. They have built the scaffolding that turns a stateless tool into a directed one. The teams that are struggling have not.

I think people hear AI has no ambition and they expect to feel reassured for the wrong reason — relief that the machines won't take over. That's not the interesting takeaway.

The interesting takeaway is that direction is a human asset that's about to become more valuable than ever. If AI is the infinite engine and you are the only thing pointing it, the value of being a clear-eyed director — knowing what you want, knowing what good looks like, knowing why it matters — multiplies.

We're not heading into a world where AI replaces humans. We're heading into a world where the humans with the clearest direction get spectacularly more powerful, and the humans without it get spectacularly more invisible. The bottleneck is no longer execution. The bottleneck is intent.