Descriptions:
Peter Werry, founder of Unblocked, delivers a conference talk at the AI Engineer summit explaining why context delivery — not model capability — is the primary bottleneck for AI coding agents, and demonstrates how Unblocked’s context engine attempts to solve it. The core argument is that agents behave like expert engineers who are perpetually onboarding: they reset their knowledge with every task and must rediscover codebases, architecture decisions, and organizational conventions from scratch. Wikis and claude.md files help but still leave agents prone to “satisfaction of search” — finding one relevant artifact and stopping before surfacing the full picture.
Unblocked ingests code, pull requests, Slack conversations, and architecture documents to build a continuously updated organizational knowledge graph. Werry demonstrates the system live, showing how an agent using Unblocked identifies nuanced optimization opportunities in an internal component called the Source Mark Engine by drawing on historical PR discussions and Slack threads that a raw Claude Code session would never surface. He also shows the system answering developer questions in Slack and auto-generating architecture diagrams from live code.
The talk frames the problem along what Werry calls an “AI maturity curve,” from autocomplete tools through Cursor to MCP-equipped agents, ultimately pointing toward fully automated software factories. The session is particularly relevant for engineering teams evaluating how to give autonomous agents durable, up-to-date understanding of large, evolving codebases.
📺 Source: AI Engineer · Published August 27, 2026
🏷️ Format: Hands On Build
