AI agents fail without organizational context: the case for context engineering

Brandon Waselnuk argues that the bottleneck for AI coding agents isn't model capability but organizational context. When agents lack knowledge of codebase history, Slack conversations, rollout procedures, and team dynamics, they produce code that compiles but breaks production.

This context gap compounds as agents scale. Without it, teams hit doom loops of corrections, wasted tokens, and review tax. Common fixes like curated markdown repos rot, and MCP tools suffer from satisfaction-of-search bias where agents stop at the first plausible answer.

Unblocked's solution is a context engine that ingests real-time data across engineering tools, resolves conflicts between stale docs and recent Slack threads, and outputs token-optimized context per workflow. Their benchmarks show 50% fewer tokens and faster triage.

The presentation includes three open-source tools: a social commit network mapper, a repo rules aggregator, and a relational context engine workshop. Waselnuk claims use cases extend beyond code generation to support and sales.