AI is a tool, not a strategy
In short: Adding AI to a product is a tactic, not a strategy. Before any AI build, answer four questions out loud - what job is slow, wrong, or expensive; what does today's solution look like; what does measurably good look like; and what failure mode can you live with.
Every week a founder tells me they're "adding AI to the product."
Cool. What problem are you solving?
Silence.
This is the most common conversation I have right now, and it tells me everything. Adding AI to a product is not a strategy. It's a tactic. Strategy is the answer to "why does this exist and who is it for?" — and if AI can serve that, fine. But it has to earn its place, the same way a database schema or a payment processor does.
The four-question diagnostic
Before I let a team write a single LLM prompt, I make them answer these out loud:
- What is the existing job-to-be-done that's slow, wrong, or expensive? If you can't name it in one sentence, you don't have a problem worth solving.
- What does today's solution look like? Even if it's "a person doing it manually" or "a spreadsheet." If you don't understand the current process, your AI version will just replicate its flaws faster.
- What does "good" look like, measurably? Latency under X. Accuracy above Y. Cost-per-task below Z. If you can't define it, you can't ship it.
- What's the failure mode you can live with? Every AI feature gets things wrong. If the failure mode is "user is mildly annoyed" — ship. If it's "regulator fines us" — don't.
I've watched teams skip these four questions and build six months of nothing. I've also watched teams who answered them honestly kill an AI project on day one and ship something better instead.
The pattern
The pattern I see across the projects that actually work: AI is the last thing they added, not the first. They built the product, found the friction, and used AI to remove it. The teams that lead with AI almost always end up retrofitting an actual product around the model — and it shows.
Use AI. Use it aggressively. But make it answer for itself first.
FAQ
Is adding AI to a product a strategy?
No - it's a tactic. Strategy answers "why does this exist and who is it for?" AI has to earn its place in service of that answer, the same way a database schema or a payment processor does.
What should I ask before building an AI feature?
Four things: what existing job-to-be-done is slow, wrong, or expensive; what today's solution looks like; what "good" looks like, measurably; and what failure mode you can live with. If a team can't answer all four out loud, the build isn't ready.
When does an AI feature actually make sense?
When it removes friction you've already found in a product that already works. In the projects that succeed, AI is the last thing added, not the first.