I Let AI Make Too Many Decisions

AI will give you an answer to almost any question you ask it.

That's the problem.

Not because the answers are wrong - sometimes they're exactly right. But when you don't have a strong opinion of your own, the AI's confidence fills the gap where your thinking should be. And then you're not building what you had in mind. You're building what the AI had in mind, based on the information you gave it, which is never the full picture.

The contradiction is the tell.

Ask AI the same question in two different sessions, framed slightly differently, and you'll get two different answers. Both confident. Both internally consistent. Neither of them wrong, exactly just based on different framings of the same problem.

If your own conviction is strong enough, that's useful. You push back, you evaluate, you take what's right and leave what isn't. The AI is a thinking partner - one with opinions based on what it knows, not what you know.

But when conviction is low - when you haven't fully thought something through, when you're tired, when you're not sure which direction to go - the AI's answer becomes the path of least resistance. You follow it. Something gets built. An hour later, sometimes a day later, you look at what exists and realise the question of whether to build it never properly got asked.

That's the drift. Not one bad decision - a gradual accumulation of decisions made by default rather than intention.

The things I let AI decide tended to share a pattern. Low stakes in the moment - a UI approach, a small feature, a structural choice that seemed minor. Low personal conviction - I hadn't thought it through enough to have a view. And quick to build - an hour, maybe less, so the cost of being wrong felt low.

But the cost wasn't in the build time. It was in the throw-away. The hour spent building something that shouldn't have been built at all, because the question of whether to build it got answered by momentum rather than thought.

The fix wasn't using AI less. It was using my own thinking first.

Form a view before asking. Even a rough one. Even "I think this should work this way, but I'm not sure." That's enough of an anchor to evaluate what comes back - to push back when it contradicts you, to recognise when it's right and when it's just confident.

AI is like a colleague with a lot of knowledge and a strong opinion. You'd listen to that colleague. You'd take their input seriously. But you wouldn't let them make your decisions for you just because they seemed sure.

The conviction has to come first. AI can pressure-test a decision you've already made. It can't make the decision for you - it'll just make one, and if you're not anchored, you'll follow it somewhere you didn't mean to go.

Steve Drew is the founder of ka-do - built for doing, not organising.