Experiments in the future of software
What happens when the cost of developing software collapses, and AI Chat becomes the primary interface?
What is Ultraniche
Ultraniche is a place to test ideas about software by building it. Each experiment ships real apps that anyone can use, and each one is written up openly: what held, what didn’t, and what it changed about the next one.
The questions
- The interface. What’s left of an app when an AI assistant handles the set-up and the questions, and only the screens that need to exist remain?
- The economics. If AI makes software far cheaper to build, how small can an app’s audience be before it stops making sense?
- The architecture. Is MCP a useful boundary between general-purpose assistants and specialised apps?
How it started
Claude can’t tell you when the next train is due because it can’t read the operators’ timetables; National Rail publishes live times as a feed. Handing that feed to Claude became Traintrackr, and it was our MCP breakthrough. If Chat is the primary interface, what else is possible?
Tally came from the opposite direction: can we rebuild an app we already use just to fix a tiny cosmetic issue we have with it? Only after we’d built it did it become clear that MCP was the killer feature.
The experiments
- Project Titan, running now. Ten single-purpose apps in ten weeks, each built as close as possible to a single prompt. Why Titan exists
Titan is the first. More will follow.