The Best AI Feature Is Sometimes No AI at All

This one's speculative — a thought experiment about design principles, not a report on something I've built.

The hammer problem

Once a team has a working AI pipeline, the instinct is to route more and more decisions through it, because it's already there and it's flexible. That instinct quietly erodes a useful distinction: some decisions genuinely need judgment, and a much larger share than people assume are actually simple lookups or fixed rules wearing a judgment costume.

A simple test worth asking first

Before reaching for a model, ask whether a human could write the exact rule down in one sentence and have it be right every time. "If the order total exceeds this number, flag it" is a rule. "Does this message sound urgent" is judgment. Only the second one benefits from a model; routing the first one through an LLM adds latency, cost, and a new failure mode — a wrong or inconsistent answer — to a problem that a single if-statement already solved perfectly and for free.

Deterministic code is a feature, not a limitation

A fixed rule is auditable, instantly explainable, and behaves identically every time it runs — properties an AI call cannot promise. For anything where predictability matters more than flexibility, that's not a consolation prize for not using AI. It's the actually correct engineering choice, and reaching for a model instead is adding risk in exchange for capability the task never needed.

The honest reason teams skip this question

Routing everything through the model is often just easier to build than sitting down and writing out the actual rule — one code path instead of two, one system to maintain instead of a hybrid. That's a real cost, but it's a development-time cost being traded for a production-time one: more expensive, less predictable, harder-to-debug behavior running indefinitely, to save an hour of design work once.

I'm Jesse Myers — Marine veteran, 32 years in enterprise IT, now building production AI systems. This site is where I write about what I've actually built, and occasionally about ideas I haven't built yet but think are worth taking seriously.