Stepsize AI

The Story

In 2013, I moved to London to live with my brother Nick and our friends Jared and Matt. We spent every evening, night, and weekend building software together out of our hacker house. One night in 2015, we prototyped an RNN that could generate code. It was rough, but it showed us something nobody else could see yet: one day, AI would build software for humans. We wouldn't be writing code anymore.

We were so pumped by that insight that we quit our startup jobs and moved into the chef's quarters of Jared's parents' Chinese restaurant in Hastings. Free rent in exchange for help in the kitchen (really it was just Jared doing the helping). That's where Stepsize started. The pitch? Natural language would become the ultimate programming language. You'd describe what you want, and AI would build it. People thought we were nuts.

What We Built

We built an issue tracker that lived inside the code editor, integrated with GitHub, Linear, Jira, and the tools developers already used. The thesis: by living where developers worked, we could build a dataset of contextualised code that AI would need to truly understand and write software.

Then we layered on AI-powered status updates that could summarize what a team had been shipping, automatically. Teams at Snyk, Amazon, and Sainsbury's picked it up because they were tired of bouncing between their editor and project management tools.