TL;DR: GitHub released Agentic Workflows in public preview on June 11, 2026 — available to all GitHub Copilot subscribers after a four-month closed technical preview. Write automation instructions in plain English markdown, compile them with gh aw compile, and an AI agent runs inside GitHub Actions to handle reasoning-heavy tasks like issue triage, CI failure diagnosis, and documentation updates.

CI/CD Just Got a Judgment Layer

Traditional CI/CD pipelines excel at deterministic tasks — build the container, run the tests, deploy the artifact. They cannot triage a flood of GitHub issues, diagnose why a flaky test started failing on Tuesdays, or update documentation to match a refactored API. Those tasks require reading context and making judgment calls. That is what GitHub Agentic Workflows is for.

Launched in public preview on June 11, 2026, it adds an AI reasoning layer on top of existing GitHub Actions without replacing anything. Deterministic steps still run deterministically. The agentic layer handles the tasks where humans were previously the only option.

4 monthsTechnical preview period
All plansCopilot Free through Enterprise
~10 minTime to first workflow run

How It Works: Markdown to Actions in Two Steps

  1. Create a plain English markdown file in .github/workflows/ describing what the agent should do.
  2. Run gh aw compile on that file to generate a .lock.yml — a standard GitHub Actions workflow file.

The .lock.yml runs as a normal Actions workflow on a schedule or triggered by events. There is no new runtime to learn. The compiled lockfile defines how a containerized AI agent — Copilot, Claude, Gemini, or OpenAI Codex — reads the repository context, takes action, and creates a PR with the result.

New in public preview: Agentic Workflows now supports GitHub Actions' built-in GITHUB_TOKEN, eliminating the need to create and manage a separate Personal Access Token.

Prebuilt Workflow Examples

Workflow What It Does Trigger
Issue Triage Classify and label new issues automatically Issue opened
CI Doctor Diagnose failures, propose fix PR CI failure
Daily Repo Status Create daily repo health report as issue Schedule
Doc Sync Update docs to match code changes PR merge
Compliance Check Detect policy violations and report Push
💡
5-Minute Quick Start
Install: gh extension install github/gh-aw
Add a prebuilt workflow: gh aw add-wizard githubnext/agentics/daily-repo-status
The interactive wizard handles engine selection, secret setup, and workflow file creation. First run takes about 10 minutes total.

Why the Technical Preview Took Four Months

The four months were not feature development — they were security architecture. Teams in the technical preview consistently said they would never give an LLM unconstrained write access to production repositories. GitHub's answer is the lockfile and safe outputs pattern: the agent operates in an isolated containerized environment, and the lockfile strictly defines what it can modify, read, and produce. The permissions, secrets, runner environments, review gates, and audit logs are all inherited from — and controlled by — the existing Actions configuration.

The design principle: "agent decides, trusted step executes." The agent cannot escalate its own permissions.

🔗
Resources · Official Sources · Getting Started
— GitHub Agentic Workflows CLI (gh-aw) Official Repository
— Prebuilt Workflow Examples Repository (agentics)
— gh-aw Release History and Latest Version Download
  • Plain English markdown compiles to GitHub Actions YAML via gh aw compile
  • AI agents handle judgment-heavy tasks: issue triage, CI diagnosis, doc updates
  • Choose from Copilot, Claude, Gemini, or OpenAI Codex as the AI engine
  • Built-in GITHUB_TOKEN support eliminates PAT management overhead
  • Available to all Copilot subscribers (Free, Pro, Pro+, Business, Enterprise)