Stop babysitting your agents...
summary
A concrete A/B: naive MCP access runs for 2.5 hours 20.9 million tokens compiles but requires multiple human corrections and would have broken production versus a context engine that runs in 25 minutes 10.8 million tokens and gets a senior engineer approved with one nitpick merge. The engine builds a research packet by reasoning across codebase Slack PR patterns and org structure resolves conflicting truth claims enforces permissions and returns token‑optimized context so agents execute mergeable work instead of prototypes.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:45Curated context layer
- 5:09Myths about agent context
- 7:15Access isn’t enough
- 8:41Effective context engine features
- 11:07Context engine differentiators
- 12:51Lessons building context engines
- 13:55Use cases for context engines
- 14:58Context engines improve agents
- 15:20Demo: social graph tool
- 17:27Demo: Zendesk integration
- 18:29Q&A
related talks
49mSolving Context Bloat: Semantic Tool Routing in Multi-Server MCP Environments
40mHow To Tame Your Agents?
33mBuild Reasonable Software
06mYour Agent Is Wasting Tokens and You Don't Know It
32mWhat Agents Do When You’re Not Looking
31mAI Coding Agents Don't Need Bigger Models. They Need Better Tools.
01h 28mAI Coding Agents and how to code them
13mSWE-Marathon: Evaluating Coding Agents at Billion-Token Scale