Building AI Products from Real Friction
There is a respectable, standard way to look for product ideas: map a market, size it, find a gap, and build for the gap. The lab’s experience points the other way. The best small AI products come from friction — the same annoyance encountered again and again until automating it is worth an evening, then a weekend, then a roadmap.
Friction is a specification
A repeated annoyance is a spec written by reality. It tells you the input, the desired output, how often the job occurs, and what “good enough” means — because you have done the job by hand and know exactly where it hurts. An idea from a market map has none of this; you have to invent the requirements, guess at the edge cases, and hope real users behave like imagined ones.
Friction also answers the hardest product question honestly: will anyone use this? If you have done the task manually a dozen times, someone already does — you.
Examples from the bench
Most of what X1Q1 Labs builds traces back this way. CLEDocket exists because tracking continuing-legal-education obligations across requirements and deadlines is exactly the kind of clerical burden a computer should carry. PinParse exists because verifying the citations and propositions in a legal brief is careful, repetitive, high-stakes reading. Playdate.pics exists because families want to share photos with a small circle they trust, not with an audience. And the household tools on the bench — an almanac of recurring dates and seasonal tasks, a local email reviewer — come from family logistics and personal infrastructure rather than from any market at all.
Why this fits AI especially well
AI capability is jagged: a model can be excellent at one task and useless at its neighbor, and the boundary is hard to predict from the outside. The only reliable way to find it is to run the real workflow, on real data, with a real tolerance for error. Building from your own friction gives you that evaluation set for free — you know what correct output looks like because you used to produce it yourself. It is also the best defense against demo-driven products: impressive in a demo, useless on a Tuesday.
The compounding loop
Build the tool, use it daily, and the next friction reveals itself — usually one layer deeper than you could have seen at the start. Small tools become bigger than expected; some stay personal infrastructure, and some turn out to be products other people were quietly missing too. For anyone with technical skill, a list of recurring annoyances is not a complaint log. It is a roadmap.