GitHub, an AI developer platform, on Wednesday unveiled its intent-driven repository solution, GitHub Agentic Workflows, in a technical preview on GitHub Actions.

The company said GitHub Agentic Workflows was designed with user environments in mind for work on GitHub, ranging from individual developers who need single-repository automation to enterprise or open-source scale operating organisations.

If a developer writes the desired outcome in Markdown and adds it to a repository as an automated workflow, a coding agent carries out the work in GitHub Actions.

GitHub Agentic Workflows uses the capabilities of the latest coding agents, including GitHub Copilot CLI, Claude and Codex, as key drivers for repository automation.

GitHub Next, GitHub's in-house research organisation, said it began developing Agentic Workflows with the question of what repository automation with strong guardrails should look like in the era of AI coding agents. It said that by combining automated coding agents with GitHub Actions, the centre of scalable repository automation, it laid the groundwork for use across millions of repositories. It added that it was designed so developers are given key decision-making authority, including when and how widely agents are used.