Why "AI-first" is usually the wrong framing
It's almost always "problem-first, AI-sometimes." A short case for boring mental models.
Every few months there’s a new version of the same pitch: your team needs to be AI-first. Start every problem by asking how AI can solve it. Default to AI.
I find this framing consistently unhelpful, and I want to say why without sounding like someone who thinks the tools aren’t useful. They are. The framing isn’t.
The actual decision
When I’m looking at a piece of work, the question I’m actually asking is: what’s the fastest path to a reliable outcome?
Sometimes that’s AI. More often it’s a combination. Sometimes it’s just running a PowerShell script I wrote two years ago that works perfectly and requires no prompting.
“AI-first” as a directive skips the decision entirely. It assumes the answer before understanding the problem, which is the exact failure mode it’s trying to fix (replacing “we’ve always done it this way” with “we use AI now”).
Where it goes wrong in practice
I’ve seen AI-first thinking produce three specific failure patterns:
Complexity creep. A simple data transformation that could be a ten-line script becomes a prompt chain with retry logic and validation. More moving parts, more things to go wrong.
Confidence without accuracy. AI produces a plausible-sounding answer faster than a correct one. When speed is the goal, you stop checking. This ends badly.
Skill atrophy. If you always reach for AI when you’re stuck, you stop building the mental models that tell you when AI’s answer is wrong. Which is the one thing you actually need.
The framing I prefer
Problem-first. For any given piece of work: what does success look like, what do I actually know about this problem, and what’s the best tool for it?
AI is often the best tool. It’s also sometimes not. The skill is knowing which is which, and that skill requires you to actually think about the problem before opening a chat window.
Boring mental model. Genuinely better results.