Summary
Brandon Waselnuk, co-founder of Unblocked, takes the stage at AI Engineer to explain why enterprise coding agents consistently underperform even when given tool access — and what a real context engine looks like versus the local-maxima solutions most teams reach for first. His central argument: AI-generated code should feel like it was written by someone who has been on the team for years, but agents begin every session with no institutional memory.
Waselnuk identifies two common but insufficient approaches: the curated context trap, where teams dump markdown files into a virtual filesystem for agents to grep over, and RAG setups that provide information access without genuine understanding. The subtle failure is that access to information is not understanding — an agent can produce code that compiles but triggers a production incident because it missed a feature flag rollout procedure or a subtle architectural constraint buried in a year-old Slack thread.
Unblocked’s context engine ingests data from across the full engineering organization — commits, incident management tools, architecture records, Slack conversations — and applies six characteristics to deliver the right context at the right time in a token-optimized format. The system serves enterprise customers including Workday and General Motors, supporting both human engineers asking questions in Slack and machine-to-machine context delivery for autonomous background agents. Waselnuk closes with a note on next-generation models like Anthropic’s Fable 5, arguing that even the most capable models need a navigable map before they can explore territory effectively.
📺 Source: AI Engineer · Published September 09, 2026
🏷️ Format: Deep Dive
