You don't need another AI feature (maybe)

Here's a take that probably sounds odd coming from someone who builds AI features for a living: most of the businesses rushing to add AI right now don't actually need to.

I should be clear that I'm not saying this as a skeptic. I build this stuff. Tool-calling, model routing, an agentic feature on my own site that genuinely does something a plain form never could. I like it, and I'll happily build you one. That's sort of why the next part bugs me.

Because a lot of what I see at the moment is AI getting bolted onto things purely so they look current. The "Ask AI" button tucked into a corner nobody was looking at (yes, there's one on this site too, and I went back and forth before deciding mine earns its keep). The chatbot stapled to a site that would have been better off as a clear FAQ. You can usually feel when it happened, because the feature has the texture of a roadmap meeting where someone said "we should have AI in here somewhere," rather than anyone watching a real person get stuck.

The thing that gets me about the bolted-on kind is that it's usually worse than what was already there. A confident little model standing in front of a problem that a dropdown was quietly solving fine, except now it can be wrong, and it can be wrong in front of a paying customer. That's a strange trade to make in the name of looking modern.

You probably want me to hand you the rule for this. When is AI worth adding, when isn't it. Honestly I've stopped pretending I have a clean answer, because I don't think one exists. It's a case-by-case thing with a pile of factors, and most of them only make sense once you're actually looking at the specific product and the specific person using it. There are things I instinctively weigh. Whether the problem is genuinely messy and open-ended, or just has a right answer some boring rule could hand you. Whether the thing it replaces is a person doing the same dull task five hundred times over. How much it really matters if it gets something wrong now and then. Whether it takes away a step people hate, or adds one nobody asked for. But none of that is a formula, and the moment I try to turn it into one it falls apart on the next project.

So I'm left with something almost embarrassingly basic, which is just: does the AI actually solve a problem someone has. That's the whole filter. Not whether it would look cool, not whether it plays well in a pitch deck. Does a real person walk away better off. It's the same bar I tried to hold my own little button to. When the answer is yes, AI is fantastic and I genuinely love building it. When the most honest answer you can give is "it'll look current," I'd treat that as a sign to put the idea down.

None of this is a campaign against AI. If anything I want more of the good kind, the kind that earns its spot. I'd just gently suggest that before you add the magic button, you check whether anyone actually wanted it, or whether a humble dropdown was getting the job done all along. Just worth remembering that the dropdown almost never hallucinates.