GitHub Agentic Workflows is generally available

GitHub describes Agentic Workflows as an open-source project from GitHub Next that brings coding agents into GitHub Actions for reasoning-based tasks.[1] Its examples include issue triage, CI failure analysis, documentation updates and test writing.[1] Users write workflow definitions in natural-language Markdown, and Agentic Workflows compiles those files into standard Actions YAML.[1] Teams can use GitHub Copilot or another coding agent, and GitHub says workflows can select model providers individually.[1] A model router can also select models according to task type and difficulty.[1]

GitHub says workflows run with read-only permissions by default in a sandboxed container behind the Agent Workflow Firewall.[1] A safe-outputs process validates workflow outputs, and a dedicated threat-detection job scans proposed changes before they are applied.[1] Agents’ access to GitHub content is subject to integrity-filter rules.[1] Teams can use existing Actions runner groups and policy constraints, and Agentic Workflows can run on self-hosted ARC runners with Docker-in-Docker.[1] GitHub also says agents can run in Cloud Hypervisor or NVX micro-VMs, while enclaves can prevent data leaks in cross-repository workflows.[1]

The service is generally available on github.com for GitHub Free, Team and Enterprise.[1] GitHub says users can install the CLI extension and follow its quickstart to trigger a first workflow.[1] The GitHub Next agentics repository offers ready-to-use workflows for triage, dependency management, code review and reporting.[1] The default GitHub Copilot CLI engine requires a Copilot license.[1] GitHub’s billing documentation says running an agentic workflow incurs GitHub Actions compute charges as well as separate AI-inference charges from the model provider.[4] The command now includes threat-detection credits and monthly ranges for AI credit spend.[1]

GitHub said Agentic Workflows was in technical preview on February 13, 2026, and described the work as a collaboration among GitHub Next, Microsoft Research and Azure Core Upstream.[2] In the preview implementation, the project was open source under the MIT license.[2] A GitHub speaker reported that one of the speaker’s leads estimated saving 29 hours a week on development work, while measured velocity across the speaker’s team more than doubled on average.[1] The speaker also said AI authored 91% of the team’s commits and that the team cut production errors from about 11,000 a day to under 1,000.[1] These are reported team results, not a stated guarantee for users.