Your agents lack context: Here's how to fix "You're absolutely right!" — Brandon Waselnuk, Unblocked

By ai.engineer

Categories: AI, Tools

Summary

AI agents fail in production because they lack organizational context—the same way new engineers do. Waselnuk demonstrates that poor context engineering early compounds into token waste, review cycles, and doom loops; fix it upfront through proper context architecture, not markdown files or basic MCPs.

Key Takeaways

  1. Context compounds like technical debt: bad context decisions early multiply into wasted tokens, rework cycles, and review tax as agents scale. Shift-left approach means investing in context infrastructure immediately, not after doom loops form.
  2. The 'curated context trap' fails because markdown documentation rots and requires an omnipotent curator for the entire org. Avoid static file systems for agent context; use dynamic, distributed sources instead.
  3. MCPs (Model Context Protocols) hit a 'satisfaction of search bias'—agents stop querying after finding first seemingly-correct information, missing critical Slack conversations or recent decision reversals (A instead of B).
  4. Goal: AI-generated code should feel written by a 5+ year team member. This requires replicating how humans build context (PRs, meetings, incident response) into agent memory and decision-making loops.
  5. Moving up the agentic adoption curve (from tab-complete to background agents without human loop) requires agents to ask clarifying questions at walls rather than fail silently, maintaining mergeable output for brownfield codebases.

Related topics

Transcript Excerpt

[music] >> Good afternoon. I hope you're all having a lovely day here at AIE. We've had great weather, though the UV has been like nine. So, hopefully you put your sunscreen on your being appropriate adults. I'm here to talk to you about context engineering, and I have the good fortune of following AJ from LinkedIn because he talked a lot about the system that we actually design and sell to other solutions. And I'm going to give you a bunch of open source tools. So, if you watch that last talk just before me, you're going to get a bunch of tool chance you can go mess around yourself, and I'll teach you a bunch of techniques today. The goal, of course, is to fix your absolutely right. I think they've taken that out of the prompts now, so it just says you're right or other things, but I'm su…

More from ai.engineer