What we work on
A few things at a time, built properly.
A few things at a time, built properly and run by us. Here's how we split the work.
Software
The things we build are our own — small web and mobile experiments. Small team, so we do the whole thing ourselves rather than hand it off, and we ship what works under real conditions, not what performed well in a demo.
AI
Applied AI, not AI theatre. Agents that do actual work. LLMs that earn the inference cost in what we build. Automation that removes a real chore from a real workflow. If the only reason a feature exists is to say we use AI, it doesn't ship.
Ideas
A pipeline with a clock. Each experiment gets a bar — a metric, a user count, a date. Clears the bar: we invest more. Misses it: we pause or wind it down and write up what we learned. Nothing accumulates quietly as a stale roadmap item.
How we decide what survives
Every experiment gets a concrete bar set before we start — not a feeling, a number. An active user count. A revenue threshold. A date by which signal has to appear. If it clears, we go deeper. If it doesn't, we stop, write the post-mortem, and move on. That discipline is the actual point. The software is just what it produces.