AI EngineerVideoSource Linked

How to Generate Mergeable Code with a Context Engine — Peter Werry, Unblocked

Agents often stop after finding plausible code context. Werry argues for task-specific retrieval that also surfaces decisions, conventions, PR history, and expertise before planning or review.

AI Engineer · Aug 27, 2026

Open Source Open Markdown Open JSON

Source Summary

Unblocked’s context engine combines code with organizational knowledge such as decisions, conventions, architecture rationale, Slack discussions, and **historical pull requests**. Its open-source query engine ingests repository PRs, synthesizes a schema, and supports agent queries.

Practical Implication

Feed agents **task-specific context** instead of attaching an entire wiki or codebase. The demo’s contextual planning took about **one minute versus about two minutes** without Unblocked, while review guidance reused prior expert comments and team relationships.

Agent-Ready Context

Unblocked’s context engine combines code with organizational knowledge such as decisions, conventions, architecture rationale, Slack discussions, and **historical pull requests**. Its open-source query engine ingests repository PRs, synthesizes a schema, and supports agent queries.

Feed agents **task-specific context** instead of attaching an entire wiki or codebase. The demo’s contextual planning took about **one minute versus about two minutes** without Unblocked, while review guidance reused prior expert comments and team relationships.

Those timings come from a demonstrated task, and the stated **50% token reduction** is a customer claim rather than a controlled evaluation. The talk does not establish how retrieval quality or maintenance effort changes across organizations.

Connected Context · Feed7 Judgment

Unblocked broadens coding context from repository structure to the decisions, conventions, discussions, prior reviews, and expertise that determine whether code is mergeable. It reinforces bounded, task-specific retrieval and gives a concrete PR-history implementation, but the demonstrated timing and customer token claim do not establish retrieval quality, maintenance cost, or reliability across organizations.

Building Agents Is Trivial Now, Context Is the Next Frontier — Jeff Ng, UnblockedThe outage example explains the failure mode Unblocked targets: repository evidence can support technically plausible advice that conflicts with later organizational decisions or incidents.The Last Human Code Review: Building Trust in AI-Generated Code — Itamar Friedman, QodoQodo treats inspectable team rules, architecture, incidents, and contracts as prerequisites for automated review; Unblocked describes a retrieval engine for supplying much of that organizational context during planning and review.Does Runtime Topology Context Improve LLM-Generated Kubernetes Security Patches?KuTIE provides controlled evidence for the same selection principle: bounded context helps when it contains relationships relevant to the task, while unrelated added context does not necessarily help.RLM: Recursive Language Models for Large Codebases - Shashi, Superagentic AIBoth reject loading an entire codebase upfront; RLMs bound evidence through recursive repository inspection, while Unblocked retrieves task-specific evidence across code and organizational history.

Context Map

contextcoding#context-engineering#retrieval#agent-reliability

Uncertainty

Those timings come from a demonstrated task, and the stated **50% token reduction** is a customer claim rather than a controlled evaluation. The talk does not establish how retrieval quality or maintenance effort changes across organizations.