The best AI strategy is sometimes no AI

· Essay

Sometimes the right AI strategy is to leave AI out.

A client asked me to add AI to one of their workflows. We sat down and mapped it out: the inputs, the rules and the exact shape the output had to take.

About halfway through I knew what I was going to tell them. The rules were fixed and nothing in the workflow needed judgement, so the same input always had the same correct output. That's an if statement nobody had got round to writing yet.

Don't use AI here.

It was the opposite of what I'd been brought in to say. But a few lines of ordinary code would be faster, cheaper and more reliable than a frontier model in the middle of the workflow, and a function never decides to get creative when nobody asked it to.

Jobs that need a model

The models are very good now, and I think that's the cause. When a tool can plausibly do anything, the instinct is to run everything through it.

The hammer got very good, so everything started to look like a nail.

A model earns its cost when there's ambiguity to resolve, context to weigh, or a rule nobody has written down yet, where the right answer depends on reading the situation.

When the rules are known and the answer has to be exact, the model is overhead. It's slower, more expensive and less predictable than the five lines it replaced, and much harder to debug. You had something that either worked or didn't, and now you have something that mostly works and can't tell you why.

Keeping code deterministic

Plenty of problems need good engineering and nothing else. It's tempting to treat determinism as a limitation you grow out of once the model is big enough. I'd protect it, because a function that returns the same answer every time is working correctly.

Audit trails, exact reproduction and failures you can trace all go the moment a model enters the loop. Sometimes that's a fair price, but it should be paid on purpose, for something you actually need.

For me most of the work is deciding which problems need a model, and saying so when one doesn't.