GitHub has launched a technical preview of Agentic Workflows, a new feature developed with Microsoft Research that runs AI agents automatically inside GitHub Actions. Defined in markdown and compiled to GitHub Actions YAML, these workflows are triggered by repository events like new issues, pull requests, or comments. Supported agents include GitHub Copilot, Claude Code, and OpenAI Codex. Typical use cases include issue triage, documentation updates, test coverage monitoring, and CI failure investigation. The feature is positioned as a complement to traditional CI/CD rather than a replacement, since agentic workflows are non-deterministic. Security is a central concern: workflows run in isolated containers with read-only repo access, internet access restricted by firewall, sanitized user content, and a Safe Outputs subsystem for write operations. GitHub warns the product is in early development and should be used with caution.
15.3K Impressions1 Comment