The AI Operator Mindset

Before anything is produced, the AI Operator asks what needs to change, why it matters, for whom, and how anyone will know it worked. When producing things becomes cheap, producing the wrong thing becomes the most expensive mistake available. Delivering something is not the same as creating value.

In practice Write down what will be different if the work succeeds — before deciding what the work is.

Question assumptions, explore unfamiliar domains and challenge the problem as it was first stated. The AI Operator does not need to know every answer. They need to recognise the questions that unlock understanding. With AI, a first map of an unfamiliar field is hours away rather than weeks, so curiosity stops being a personality trait and becomes a working method.

In practice Ask why a request exists before asking how to fulfil it — and use AI to challenge your own framing, not only to answer it.

Break complex work into parts, decide which belong to people and which to machines, give each the context it needs, and assemble the results into something coherent. The skill looks less like mastery of a single craft and more like the work of a good editor or programme lead. The goal is not maximum automation. It is the effective combination of human and machine intelligence.

In practice Before starting, sketch the work as a sequence of steps and mark who — or what — is best placed to do each one.

Turn vague requests into clear objectives, context, constraints, expectations and acceptance criteria. The quality of what comes back usually depends on the clarity of what went in, whether the recipient is a model, an agent or a colleague. Prompting is a tool. Structured thinking is the underlying competence.

In practice If you cannot write down what “good” looks like, you are not ready to delegate the work — to anyone.

Exercise judgement, check assumptions, test results and learn where automated systems tend to fail. That last part is harder than it sounds. In a large experiment with Boston Consulting Group, consultants using GPT-4 did markedly better on tasks within the model’s capabilities. On a task chosen to fall just outside them, they were 19% less likely to reach a correct answer than colleagues working without AI.¹ From the inside, the boundary was not obvious. Responsibility cannot be delegated to a model; trust is earned through reliable outcomes.

In practice Decide how you will check the work before you see it, and scale the checking to the cost of being wrong.

Prototype quickly, test assumptions, gather evidence and iterate. AI sharply lowers the cost of trying things: a first version that once took weeks can often be in front of real people within days. The AI Operator uses that advantage without confusing speed with progress. A fast experiment is only useful if it was designed to teach you something.

In practice Replace the next debate about what might work with a small test that shows what does.

Take responsibility whether the work was done by a person, an agent or both. Measure success by the value created, not by hours worked, tasks completed or lines of code written. The work is not about proving individual expertise. It is about making meaningful things happen.

In practice When something goes wrong, “the model got it wrong” is not an explanation. It is where your explanation starts.