Last week I added 7 AI buttons to my student backend
— Summary AI, Planning AI, Q&A AI. Looked impressive. After a week, users clicked fewer than 5 times. I was thinking: "Not enough entry points, right?"
Microsoft just hit the same wall
Microsoft Copilot's head Jacob Andreou recently admitted publicly: they had stuffed Copilot into several Windows apps. Traffic looked great, but users used it and left — they didn't stick around. Now they've pulled those features back.
Andreou's words hit hard: "Products have to earn their right to exist through the value they create. Just driving people to the entry point doesn't work."
My mistake was treating AI like a universal band-aid — slapping it wherever it hurt. But users didn't actually feel the "pain." Three lessons they learned, I copied too: 1) High-traffic entry ≠ valuable entry; 2) Fewer entry points, deeper usage per point; 3) AI popping up suddenly annoys people — worse than no AI.
Today you can audit your product with this same lens
Money: $0
Time: Half a day to one day (depends on product complexity)
Technical barrier: No code needed — just open a checklist
First step: Mark every spot in your product/website with "AI" / "Smart" / "Auto" in the name. Then ask yourself one question for each — "If the AI wasn't here, what task would the user fail to complete?" If you can't answer, cut it.
I listed 12 AI entry points myself, kept only 3 that were actually useful. The week after cutting, usage of those three remaining entries doubled.
How to use this at different stages
If you're just starting (no product yet): Don't think "where to add AI" yet. Think "which user action hurts so much they'd pay to fix it." AI is painkiller, not vitamins.
If you have 1-2 clients: The most important thing to do right now is ask them directly — how many times did you open my product last week? What for, each time? Write the answers down. Three weeks later you'll see the picture clearer than any analytics dashboard.
If you're scaling (have paying users): This is the stage where you're most likely to make my mistake — adding more and more features, each with AI. Stop. Look at the data. Cut the rarely-used ones first — that's ten times more important than adding new features.
One last thing: not everyone needs this advice. If you're not currently "adding AI buttons everywhere," it's fine to skip this for now. But if you are, pause. Look at that number first.