I've reviewed about 30 products in the last six months that prominently advertise AI features. Here's the pattern I can't stop noticing.
For a large proportion of "AI features," I ask: could this be done by a lookup table or a few if-statements? If yes — it's not actually using AI's core capability. It's using a very expensive API to do something that a junior dev could have hardcoded in an afternoon.
Examples I've encountered: "AI suggests the next status for your ticket" (there are four possible statuses); "AI auto-categorises your expenses" (the categories are fixed and defined by the customer); "AI checks for required fields" (this is form validation).
The product team wants to ship something with an AI angle. The cheapest route is to wrap an LLM call around a task that didn't need it. This creates a feature that's slower, less reliable, more expensive to operate, and harder to debug than the deterministic alternative — but it has "AI" in the changelog.
AI features genuinely earn their keep when the input space is unbounded and pattern-matching over training data is the only way to handle it: free-text classification across hundreds of dynamic categories, drafting content that requires stylistic judgment, synthesising information from heterogeneous sources. These are hard to fake with lookup tables.
The tell: if you can enumerate all the outputs, it's probably not an AI problem.
Comments
Loading comments…