Problem: LLMs are good at code. Commercial codebases are big and messy. The gap between "understands code" and "understands the business" is where AI coding stalls. Closing that gap — automatically — is the opportunity.
AI coding tools today operate with amnesia. Every session starts from zero context. The developer provides what the agent needs to know, or the agent re-discovers it by reading files. Neither scales.
Commercial codebases carry decades of accumulated decisions: why this pattern, not that one. What this field means to customers. Which systems break if you rename that event. This knowledge lives in people's heads, scattered docs, tribal lore, and implicit conventions. It's the reason a senior engineer at the company ships 10x faster than a senior engineer on day one — not skill, but context.