When Ritwik first made me aware of this problem, the part that stuck with me was not the hours wasted. It was that in 2026 the PM is doing this work manually, in the same year coding agents are writing production code. The customer context sits on Slack, Jira, Confluence, Meetings, Calls, every team’s tool stack.
PMs are the human glue that holds it all together, and even after so much effort, the spec they hand over to engineering still doesn’t have the codebase context an agent needs to build the thing right the first time.
So we created the layer that holds it together.
The core of Radic is a knowledge graph that reads signals and your codebase. Not vector retrieval, not just clustering, a real graph where customer evidence directly touches modules and dependencies. Getting it running reliably with real unstructured data, real customer Slack threads, a real codebase, was the hardest and most satisfying thing I’ve ever created. It’s been four years building production AI at Deloitte, then agentic systems with Citigroup and Enterprise Ireland, plus my own AI agent marketplace, and this feels like the bet that matters most.
Cursors, VS Code and MCP integration into Windsurf is the part I’m most excited about. The specifications are nail-biting. But the real opportunity comes when the product context stops living in a silo. Every customer signal, every architectural decision, every specification connected, queryable, executable by an AI agent.
We are just getting started. Come talk to us in the comments 🙌
<a href