November 2026 Summaries

Sourcegraph developed an internal GitHub Actions bot to address review bottlenecks created as coding agents accelerate code production, replacing routine approval chasing with explicit review paths: Ship for changes that need no additional review, Show for post-merge visibility, and Ask for required pre-merge human review. The bot examines pull-request diffs, surrounding code, and an author’s recent, relevant contribution history, while applying rules that always require review for sensitive areas such as authentication and data exposure. Initially used only to make recommendations, the system was refined through human feedback, with emphasis on providing evidence that lets developers question its assessments and distinguish component-specific familiarity from general experience. The experiment has reduced requests for quick approval stamps, but Sourcegraph is still evaluating whether deferred reviews receive meaningful attention, how the bot’s notifications affect interruptions, and whether lower approval demand maintains review quality.

Sourcegraph will retire its AI coding assistant Cody on January 31, 2027, concluding a product launched in early 2023 that combined AI chat and code completion with Sourcegraph’s code-search infrastructure to help developers understand complex codebases. The company says it chose not to develop Cody into an autonomous coding agent and is instead focusing on retrieval infrastructure that supports developers and a range of third-party or internally built agents, including tools such as Amp, Claude Code, Cursor, and Codex. Its strategy centers on a deterministic code graph, Code Search, Deep Search, Code Finder, MCP and CLI integrations, and Agentic Batch Changes, intended to improve code discovery, AI-generated code accuracy, and large-scale changes such as migrations, dependency upgrades, and security fixes. Sourcegraph acknowledges that Cody customers may need time to replace existing IDE and API workflows and says it will work with affected teams, noting that Deep Search and MCP integrations can support many, but not all, current use cases.