Meet GitHub Agentic Workflows: How to Automate Your Entire Repository Using AI Agents

Modern artificial intelligence (AI) has already changed how developers write code, but what if it could also manage your entire repository on its own? That's no longer a hypothetical. GitHub recently launched Agentic Workflows in technical preview, marking one of the most significant changes to developer tooling since the introduction of GitHub Actions. If you are part of a software team drowning in issue backlogs, flaky CI pipelines, and repetitive maintenance tasks, this release couldn't have come at a better time.

Agentic Workflows is a rethink of what automation means in a modern software development context, where, rather than writing complex YAML configurations to instruct machines how to do things step by step, developers can now describe what they want accomplished in plain Markdown and let an AI agent figure out the rest. This is part of GitHub's broader philosophy of Continuous AI, where the integration of AI into the software development lifecycle improves automation and collaboration, much as Continuous Integration/Continuous Deployment (CI/CD) practices transformed how teams build and ship software.

What Exactly Are GitHub Agentic Workflows?

GitHub Agentic Workflows allow developers to automate repository tasks using AI agents that run natively within GitHub Actions. The feature is a collaboration between GitHub, Microsoft Research, and Azure Core Upstream, and the entire implementation is open source under the MIT license.

The mechanics are straightforward. Instead of drafting a YAML file, developers can add a plain Markdown file to their .github/workflows/ directory describing their automation goals in natural language. Running gh aw compile via the gh aw CLI extension then converts that Markdown into two files:

• A standard GitHub Actions workflow, and
• A corresponding lock file (.lock.yml), which is what GitHub Actions actually executes.

The workflow runs using a coding agent of your choice, which is GitHub Copilot CLI by default, but also Claude Code or OpenAI Codex, depending on your configuration.

Importantly, GitHub positions Agentic Workflows as an addition to existing CI/CD pipelines, not a replacement for them. They are designed for subjective, judgment-heavy, repetitive tasks that traditional YAML workflows struggle to express, not for build, test, or release pipelines.

Key Features and Capabilities

Here's what makes Agentic Workflows stand out from conventional GitHub Actions automation:

• Natural language workflows: Describe what you want automated in plain Markdown, the AI agent interprets your intent, and determines how to carry it out.
• Security-first by design: Workflows run with read-only permissions by default. Write operations require explicit approval through "safe outputs" pre-approved, reviewable GitHub operations, such as creating a pull request or adding an issue comment. Execution also happens in a sandboxed environment with tool allowlisting and network isolation.
• Pull requests are never merged automatically: A critical safety boundary baked into the system where humans must always review and approve any proposed code changes before they are merged.
• Support for multiple AI coding agents: Works with GitHub Copilot CLI (the default), Claude Code, or OpenAI Codex. The same Markdown workflow format runs across all engines, avoiding lock-in to any single AI provider.
• Deep GitHub integration: Being a GitHub product, you get native access to repositories, issues, pull requests, Actions, and security tooling through the GitHub MCP Server, with additional support for browser automation, web search, and custom MCP integrations.
• Flexible trigger mechanisms: Workflows can respond to issue and pull request events, run on scheduled intervals, be dispatched manually, or be invoked via commands in comments.
• Agentic authoring: Developers can create, edit, debug, and optimize workflows using AI agents directly in VS Code, GitHub.com, or their preferred coding environment.

Real-World Use Cases: Where This Gets Interesting

GitHub identifies six primary categories of repository automation that Agentic Workflows make possible, tasks that would be difficult or impossible with traditional YAML workflows alone:

• Continuous triage: Automatically summarize, label, and route new issues as they arrive.
• Continuous documentation: Keep READMEs and documentation aligned with code changes.
• Continuous code simplification: Repeatedly identify code improvements and open pull requests for review.
• Continuous test improvement: Assess test coverage and add high-value tests.
• Continuous quality hygiene: Proactively investigate CI failures and propose targeted fixes.
• Continuous reporting: Generate regular reports on repository health, activity, and trends.

For teams looking for a practical starting point, Peli's Agent Factory — named after co-author Peli de Halleux serves as a guided tour through different specialized agentic workflows, with practical patterns that teams can adapt, remix, and standardize across their repositories.

In Conclusion:

The broader implications of Agentic Workflows are ones the entire tech industry should pay attention to. For decades, automation in software development required developers to explicitly program every conditional, every trigger, every exception. Agentic systems introduce something different: the capacity for software infrastructure to exercise judgment within boundaries you define.

GitHub's approach to that responsibility is notably careful. The security model is read-only by default, sandboxed execution, approved write operations only, and a firm rule that pull requests are never automatically merged shows that the team is taking seriously what it means to hand autonomous agents the keys to a production repository. The guardrails are intentional, not an afterthought.

• If you are a business leader, the value proposition is that engineering hours spent on repetitive operational work can be redirected to higher-impact product development.
• If you are a developer, GitHub Agentic Workflows promise to help you spend more time on building and less time maintaining.

GitHub Agentic Workflows are still in technical preview, meaning community feedback is actively invited through GitHub's Community discussion board and the GitHub Next Discord. But the foundation being laid here is open source, security-conscious, model-agnostic, and built on natural language. GitHub Agentic Workflows look less like an experiment and more like the scaffolding for a new normal in software development.