Building Snapity without writing the code

27 September 2026

In the summer of 2025 we were in the very early days of working out how to work with coding agents. People everywhere were trying them, sharing tips and showing their enthusiasm, and we were all working it out as we went along. I wanted to see how far the agents could go without me, so I set myself an experiment. I would build a mobile app, publish it, produce all its marketing and make a sale, without writing a line of code or a word of the marketing myself. The first commit on the project that became Snapity went in on 19 July 2025.

I live in Cambridge, and the railings around town are covered in posters for events. I have often stopped in front of one and thought I'd like that in my diary, and then done nothing, because I couldn't be bothered typing all the details from the poster into my calendar. I had wanted an app to do that for years, and with AI it had become possible to photograph a poster and have a model pull the event out of it. It was a fairly small, self-contained job, which suited an experiment. It was also my chance to get some experience of Kotlin Multiplatform, which lets one codebase serve Android and iOS.

At first I read every line the agent produced, and it often needed improving, fixing or rearchitecting. I found the problems and told the agent how I wanted them fixed. After a while I got a sense of what kept going wrong by my standards, so I wrote those standards down. Coding standards for Kotlin and Compose went into the agent's instructions on 10 August, and they fixed a lot of the problems. Brand guidelines and a marketing strategy followed.

Some code was easy to let go of. The website I could check by looking at the page. Kotlin Multiplatform involves a lot of boilerplate marshalling between Swift and Kotlin, and it was too tedious to read, so I took it on trust. The rest I stopped reading once I was confident the standards were working.

All the while, ideas about how to work with agents were changing every few weeks. Prompt engineering had been the big theme, and by June 2025 the talk had moved on to context engineering. Over that summer it moved again, to spec-driven development, first with Amazon's Kiro and then with GitHub's Spec Kit, whose commands I added to Snapity in September. Agents could only be trusted with small pieces of work then, so I kept fine-grained control, writing a detailed specification and instructing the agent to work through it one point at a time. By February 2026 the whole practice had a new name, agentic engineering, and the way I had worked in September already looked dated.

The agents were weakest where the work was least mechanical. The user experience took many iterations, some with real testers, before the flow was right. Talking to users changed the product as well. I realised that people would more often have an event as text, an invitation arriving by email or WhatsApp, or a poster attached to a message, than stand in front of a poster taking a photo.

My job was deciding the priorities, reviewing the interface and talking to users. Marketing needed me too. The agents wrote the scripts for the videos, one or two each on TikTok and YouTube Shorts, and generated the voiceovers. People thought the voiceovers sounded creepy, so I replaced them with a real voice, which was a departure from making everything with AI. That made me realise how hard a compelling voiceover is, and that most people can't do one well, so the temptation to use an AI voice is still there. I haven't resolved it.

Snapity is live on Google Play and the App Store, and as a desktop download for Mac, Windows and Linux. The phone apps peaked at 119 monthly active users in February 2026 and have had between 22 and 32 a month since May. Desktop users don't show up in those figures. The app is still free. I planned to keep it free while I established the market and settled on pricing.

In July 2026 I realised that the ChatGPT, Claude and Gemini apps could all do what Snapity does, taking text or an image and turning it into a calendar event. I kind of knew this was coming, so I wasn't too surprised or disappointed, but it meant the chance of a sale had passed. The sale was only ever in service of the bigger goal, which was to learn how to use the AI coding tools and see what was possible. On that, I got hands-on experience of all the tools and techniques, and a better sense of where the boundaries of the possible lay. The question now is whether it is worth keeping Snapity running at all. I have been trying to think of a niche the big companies wouldn't be interested in, and I think I might have found one.