GitHub Agentic Workflows GA: forecast the spend first

GitHub Agentic Workflows is now generally available. You write a Markdown file that describes the job in plain language, for example "triage new issues" or "explain why CI failed". GitHub compiles that file into standard Actions YAML, and a coding agent does the work inside your Actions runners. GitHub announced GA on October 28 at Universe. It is the most direct way yet to put an AI agent in your repo on a schedule, and the bill comes from your AI credits.

What actually happened

Per GitHub's GA announcement:

• Availability: github.com on GitHub Free, Team and Enterprise. The default engine is the GitHub Copilot CLI, which requires a Copilot license. You can also bring your own coding agent and pick the model provider per workflow.
• Use cases GitHub names: issue triage, CI failure analysis, documentation updates and test writing.
• Safeguards: agents run with read-only permissions by default, inside a sandboxed container behind the Agent Workflow Firewall. Proposed changes go through a "safe outputs" validation step, and a separate threat-detection job scans them before they apply. Agents can also run in a Cloud Hypervisor or NVX micro-VM.
• Cost tools: gh aw forecast is no longer experimental. It now shows monthly ranges of AI credit spend, and it includes threat-detection credits.
• Self-hosted: support for Actions Runner Controller (ARC) runners with Docker-in-Docker.

Prebuilt examples are in GitHub Next's agentics repository.

Why it matters for your business

This is the agent pattern we recommend to clients, packaged by GitHub: the agent reads, proposes and waits. It does not write freely. Read-only defaults and a validation step before any change are what make an agent safe to leave alone in a repo.

There are two catches.

First, the cost. Each run consumes AI credits, and the threat-detection scan consumes more. A triage workflow that runs on every new issue in a busy repo adds up. Run gh aw forecast before you turn a workflow on, and set a Copilot budget. The forecast is a range, not a quote, so check real spend after the first week.

Second, the Markdown file is now code. It decides what an agent does with your repo. Review changes to it the same way you review a workflow YAML change, and keep it in CODEOWNERS.

Start with one low-risk workflow, such as CI failure analysis on a private repo. Measure it for two weeks. Then add the next one.

Sources: GitHub Changelog, GitHub Next agentics repository.