Why we're building Tanagram

In 2019, as a new engineer at Stripe, I built a feature for our team's top user. I added tests, rolled it out successfully, and our user was happy.

I also set myself up for disaster.

Two years later, some other team discovered a data anomaly, and it blew up into a huge incident that took six months to clean up.

As it happened, I'd updated our main system, one that powered the API and user-visible aspects of this feature. Another team, millions of lines of code away, owned a different system that handled reporting. Neither of us knew of each other's changes, or that one side depended on the other.

This example plays out everywhere. Once you have more than a couple of people, codebases become a primordial soup glued together with tribal knowledge, neglected documentation, and random Slack pings. There's too much of it for you to read and remember everything.

In 2019, we didn't have a good solution to this problem.

Now, everything is changing. AI agents are tantalizingly close to agentically building entire backlogs. Give them enough tokens, tools, and guardrails, and you can almost taste the AGI.

But they need feedback and guardrails — rules telling them what's good and bad, what to do and what not to do. The tribal knowledge, documentation, and random Slack pings that were a nice-to-have last decade become table stakes this decade.

That's what we're building at Tanagram: something that takes your engineering team's collective knowledge and makes it clearly visible and enforceable for people and agents.

Tanagram exists to accelerate agentic coding. It's early days, but if you want to find out where this is all going, give Tanagram a try or join our team.

— Feifan Zhou, Cofounder