I'm Back
Almost five years ago, when I built my first data app, I said, "in ten years, I don't want ten projects. I want one project that is ten times as awesome."
I wrote that on this blog in 2023, in a post about focus, and learning. I promised, "I'll share more on this project soon."
Soon turned into a while.
So here's the update. It's less about where these ideas went and more about where they're going. And it starts with a confession about the bet itself: I was wrong about what "one project" would look like.
Success requires focus. And I thought I would on one thing. An app, a platform, some single artifact I could point to and say, there, that's the one. But that's not what happened. The one project turned out to be an ecosystem — a way of doing science that ties together everything I build. It has a name now; it's a name I like... it's Heureka Labs.
Here's what's on the bench.
The data. Underneath everything is a set of pipelines that stitch biological databases — genes, proteins, compounds, cell lines — into something you can actually reason over. Most biological knowledge is scattered across hundreds of sources that don't talk to each other. The pipelines make them talk. This is the discovery engine: connect the data, and you can ask questions nobody's been able to ask.
The tools. I'm started collecting ideas for this post with an AI agent running in my terminal. That sentence would have been science fiction when this blog started; today it's Tuesday and I decided to restart this writing project. The way I build changed completely. I write code with agents, wrangle data with agents, and more and more the tools I depend on don't have a screen to click at all. This isn't an AI-written blog (ed. note: I'm still reading and editing by hand). And I've written before about where AI falls short. But the tooling continues to improve, so we continue to build and explore.
The bench. The newest piece is Heureka Bench. We're making a bet on local: an electronic lab notebook for researchers, with an AI agent built into its core. Your protocols, data, code, and notes live on your own machine; the agent helps you author, analyze, and search the whole record. It's where the data and the tools meet an actual scientist at an actual desk (or bench). It's the part of the ecosystem that augments scientists, not replace them.
Three things, one system. The pipelines feed the bench, the agents run through all of it, and the whole stack points at a single question I've been chasing since the first post on this site: how do you come up with a new idea?
That question is why the writing is coming back.
During the year-plus, I kept a running list of things I wanted to think through: how we produce information faster than we can absorb it; the technical debt we carry in our own lives, not just our codebases; what happens to creativity when the tools stop having interfaces; whether discovery can be taught or only caught. Some of these will become posts. Some won't. That's the experiment, same as it ever was.
My sabbatical next year means more time to think. I'm not promising a weekly cadence. I'm promising to write when there's something worth saying, and lately there's a lot.
Five years ago I made a bet on one project. We're at an inflection point where it's beginning to pay off, just not structured in the way I predicted. I'm back to share with you what I find.
Let's discover together.
