Before you build the AI thing: what problem are we actually solving?

Somewhere right now, a competitor put "AI-powered" on a landing page, an executive saw it, and a slide was born. The slide says "AI." It does not say what for. And now three engineers are being asked to bolt a chatbot onto a problem that a dropdown solved in 2011.

So, at the risk of being the least fun person in the roadmap meeting, the one question I ask before writing a line of code: what business problem are we actually solving?

The question nobody wants asked

Not "what can AI do here." What outcome do we need, for whom, measured how? If we can't answer that in one sentence without the word "innovative," we're not ready to build anything — least of all something that bills per token.

Things that are loudly not AI problems

• "Our search is bad." — Might be. Have you tried, and I'm just spitballing here, an index? Before you embed all your data into a vector database, make sure the boring version doesn't already work.
• "Users can't find the button." — That's a design problem wearing an AI costume. A chatbot that explains where the button is, is a worse button.
• "Reports take forever." — A SQL query and a scheduled job. No agent. No prompt. No 2 a.m. page because the LLM decided to get creative with your numbers.
• "We want to look innovative." — Congratulations, that's a marketing problem. Please do not ship marketing to production.

When AI actually earns its seat

None of this is anti-AI — I build this stuff for a living. AI earns its place when the input is genuinely unstructured or ambiguous, when you need real language understanding, or when the volume of judgement is beyond what humans can do by hand. Summarising a thousand support threads a day? Yes. Deciding whether 4 is greater than 3? Please use an if-statement.

The cost of skipping the question

Skip it and here's the sequel: six months, a cloud bill with a comma in it, an on-call rotation babysitting a model, and a feature a well-placed form would have shipped in a sprint. "AI-first" should mean you reached for AI because it was the right tool — not because it was the word in the room.

Being an AI engineer means knowing when the answer is AI. It also means being honest when the answer is a dropdown.

I love building AI. I love not wasting your money slightly more. Ask the question first — and if AI really is the answer, we'll build it properly.