GitHub Agentic Workflows Now in Public Preview

GitHub Agentic Workflows Reaches Public Preview

GitHub Agentic Workflows moved from technical preview to public preview on June 11, 2026. Originally introduced in February 2026, the feature is now accessible to all GitHub users.

What it does

Agentic Workflows lets development teams automate complex, judgment-intensive repository tasks using coding agents running inside GitHub Actions. Unlike deterministic CI/CD scripts, agentic workflows can reason about context: triaging issues by reading descriptions and applying labels, analyzing failing CI jobs and proposing fixes, updating documentation from code changes, and performing routine maintenance.

The defining feature is its configuration model. Rather than writing YAML, users describe automation goals in natural-language Markdown files in .github/workflows/. The gh aw CLI compiles these into standard GitHub Actions YAML, reusing existing runner groups and policy constraints.

Multi-Agent Support

The system supports three underlying agent engines: GitHub Copilot, Claude Code, and OpenAI Codex, so teams are not locked into a single AI provider.

Security Architecture

Agents run with read-only permissions by default, and all content access is filtered through integrity rules. Execution happens in a sandboxed container behind the Agent Workflow Firewall, a dedicated threat detection job scans agent output before changes are applied, and outputs pass through a safe-output validation process. Early adopters including Carvana and Marks & Spencer report multi-repository changes and reusable workflow catalogues that complete in minutes.