Your CEO Says "We Need AI." Now What?
"We need AI." If you run anything, you've either said that sentence or had it said at you. It's the most common AI question I get — and it's also the least useful one, because it skips the only step that actually matters.
Episode 15 was our first Dear Human in the Loop: Oscar and I answering the practical questions founders, engineering leads, and developers actually ask, instead of doing another news recap. Four questions came in. Here's how I answered them, and where Oscar and I pushed on each other.
"We need AI." Where do I start?
Start with problems, not prompts.
"We need AI" is a solution shopping for a problem. My first question back is always the same: what's one painful, repetitive process that's costing you real time or money right now? Start there. Not with a model, not with a tool, not with a Head of AI — with a single workflow worth fixing. Get one closed loop working end to end, measure it, and you've learned more about where AI helps your business than any strategy deck will tell you. Oscar comes at this from the engineering side and I come at it from the "what actually ships" side, and we land in the same place: pick the problem first, and the tech choices get easy.
Should an AI engineer be my startup's first technical hire?
Hire for your biggest business problem, not the trendiest job title.
An "AI engineer" is a great hire when your core problem is genuinely an AI problem. If your real problem is that you can't ship product, you need someone who can ship product. The title on the job post should be downstream of the bottleneck, not downstream of the hype cycle. I've watched teams hire the exciting résumé and then discover the thing holding them back was plumbing nobody wanted to own. Figure out what's actually broken, then hire the person who fixes that.
My developers don't trust AI. Should I push them?
Show them real value and give them time. Trust beats mandates.
You can't mandate your way to adoption — push too hard and you get malicious compliance and quiet resentment. What works is letting good developers feel the value on their own terms: a task that used to take an afternoon done in twenty minutes, with them still in control of the output. Skepticism from strong engineers isn't the enemy; it's a signal to take seriously. Earn the trust, don't order it, and the adoption takes care of itself.
Which skills will still matter in five years?
Fundamentals. Model quirks expire; the fundamentals don't.
Every specific trick — this prompt pattern, that tool's shortcut — has a shelf life. What compounds is the boring, durable stuff: engineering fundamentals, system design, product thinking, communication, and critical thinking. Those are exactly the skills you need to point AI at the right problem and judge whether its output is any good — which, conveniently, is the same muscle every other answer here depends on. Bet on the things that were true before this wave and will be true after it.
The through-line across all four: AI is a tool, and tools serve problems, people, and fundamentals — in that order. Start there and "we need AI" turns into something you can actually ship.
Got a question for a future Dear Human in the Loop? Drop it in the comments. Watch or listen to the full episode below.
