AI lowers the cost of building features so much that teams build capabilities looking for a use. The antidote is an old discipline: start from the job the customer hires your product to do.

The filter

  • Name the job. What progress is the user trying to make? An AI feature that doesn't advance a real job is decoration.
  • Check it beats the current way. Users already get the job done somehow. Your feature must be meaningfully better, not just newer.
  • Find the friction it removes. The best AI features delete steps from an existing job rather than adding a new thing to learn.
  • Validate with the actual user, not the demo audience.

Bottom line: run every AI idea through the job-to-be-done — features that advance a real job earn their place; capabilities in search of a use don't.