Peter Werry is a founding engineer at Unblocked, where he builds context engines for coding agents that connect software with the decisions, conversations and expertise behind it. His goal is to help agents produce changes that reflect how a particular engineering organization actually works.
Based in Vancouver, Werry studied at the University of British Columbia and worked at Buddybuild and Apple before joining Unblocked. By September 2024, he was developing tools that combined source code with Slack conversations, documentation and issue trackers. His work now extends that organizational knowledge into autonomous coding, incident response and pull-request review.
- Searchable information is not organizational understanding. Connecting tools, adding vector search or expanding context windows cannot tell an agent which decision matters, which sources disagree or why a previous approach failed. Werry favors targeted, personalized retrieval that connects code with review history, incidents and team conventions.
- Satisfaction of search can derail coding agents. Borrowing a concept from radiology, he describes agents that stop after finding a plausible answer while overlooking an incident report, rejected implementation or relevant Slack conversation. Effective retrieval must continue across organizational boundaries without exhausting context or tokens.
- A social engineering graph can locate actual expertise. Contribution histories, pull-request reviews and collaboration patterns help identify who understands particular systems and which decisions shaped them. Frequent activity alone does not establish authority. Unblocked’s document-query-engine workshop project illustrates structured retrieval over GitHub issues and pull requests.
- Permission-aware organizational memory must remain current. Knowledge assembled from private channels and repositories should inherit their access restrictions. Werry also cautions against caching generated answers as durable truth: evolving code and documentation can make old outputs misleading, while unresolved conflicts should remain visible.
His AI Engineer Europe session connects these principles to ticket enrichment, context-aware reviews and background agents: software should reflect the accumulated judgment of the people responsible for maintaining it.
