The software industry is growing quickly due to AI. 36 million new users joined Github in the past year¹, and 63% of people using AI to write code identify as non-developers². Most of these developers will never write a single line of code. They’ll describe what they want, and an LLM will write it.

Currently, developers use git to create a paper trail for the code they author. Git captures the lines of code that changed and a brief sentence why, nothing else.

That paper trail is missing context. For the first time in the history of software, that context exists.

When a developer starts a new coding session, they prompt the LLM with instructions for what they want to build. Dozens of decisions form the rest of the session, all of which we can excavate to build a new foundational layer.

This makes up the missing tribal knowledge that exists in conversations and meetings, and without it, it’s hard to understand the product.

But once you have it, you can answer convoluted questions. The reason your token expires in a day? It’s because it matches your session token. That confusing database column? It ended up being a temporary flag for an old A/B test.

Millions of new developers don’t have the clarity to be confident in what they’re building, and the current tools require an engineer to review every line of code.

The data shows the consequences: 96% don't fully trust AI code is functionally correct, and only 48% always review before merging¹. 10.3% of Lovable apps shipped with critical row-level security issues². Even the models exhibit this gap: they identify their own vulnerabilities 78.7% of the time in review mode, but generate them 55.8% of the time by default.

We need more context to ship confidently.