Give your coding agents the context your team already has

AI-generated code should feel like it was written by someone who’s been on your team for years. That means understanding the decisions, conventions, and lessons that shape how your software gets built.

In this talk, Unblocked’s Brandon Waselnuk explains how to give agents that context so you spend less time correcting their work.

Your agent found an answer. Did it find the decision that changed it?

An architecture document says to do one thing. Last night’s Slack conversation says to do another. Your agent finds the document, writes code, and tells you it’s done. Now you have to explain what it missed.

As teams move toward parallel and background agents, those gaps create more rework, wasted tokens, and code to review. Brandon walks through what a context engine needs to do to help agents understand how your organization works and act on the right information.

What you’ll learn

• Where context files and MCP fall short. Curated files go stale, and agents can stop searching after the first plausible answer.
• How to handle conflicting information. Connect knowledge across code, docs, and conversations to determine which guidance applies.
• What makes context useful to an agent. Account for who’s asking, enforce permissions, and deliver relevant information without filling the context window.
• How to start building. Explore open source tools for mapping team expertise and auditing repository rules, plus a workshop on combining search with structured queries.