Introduction
Agent-based workflows have become a dominant trend in software development, with GitHub leading the charge by integrating agentic capabilities into its platform. One notable experiment is an accessibility agent designed to support developers in creating more inclusive user interfaces. This article shares the successes and lessons learned from building this general-purpose tool, offering valuable insights for teams aiming to enhance their accessibility practices.
Two Core Goals of the Accessibility Agent
GitHub is piloting an experimental agent with two primary objectives:
- Provide just-in-time guidance: Deliver reliable answers to accessibility questions directly within the GitHub Copilot CLI and VS Code integration, helping engineers resolve issues as they code.
- Automate remediation of objective issues: Identify and automatically fix simple, indisputable accessibility problems before code reaches production.
To achieve the second goal, the agent automatically evaluates any changes that modify front-end code.
Performance and Common Issues Found
To date, the agent has reviewed 3,535 pull requests with a 68% resolution rate. The top five issue types, in order of frequency, are:
- Making structure and relationships clear to assistive technologies
- Providing clear and concise names for interactive controls
- Ensuring users are aware of important announcements
- Ensuring text alternatives for non-text content
- Moving keyboard focus through pages and views in a logical order
Each of these represents a barrier automatically removed, improving the experience for people who rely on assistive technology.
Understanding LLMs and Agents
This post assumes basic familiarity with large language models (LLMs) and agents. If you need a refresher, explore these resources:
- A guide to deciding what AI model to use in GitHub Copilot
- GitHub for Beginners: How to get LLMs to do what you want
- How to write a great agents.md: Lessons from over 2,500 repositories
- Multi-agent workflows often fail. Here’s how to engineer ones that don’t.
The Mindset Behind the Agent
The social model of disability teaches that access barriers—and therefore impairment—are created by how environments are built. The same principle applies to digital experiences. With this agent, GitHub is not trying to “solve” accessibility in isolation. Instead, it augments developers’ efforts to remove barriers that arise from how GitHub’s UIs are constructed.
The agent is not a silver bullet for every scenario. By clearly defining its scope—focusing on objective, automatable issues—the team accelerated the experiment’s launch and gained broader buy-in.
Lessons Learned from Past Accessibility Efforts
Previous accessibility initiatives at GitHub provided crucial context. The agent builds on earlier manual reviews and automated tools, but with a key difference: it operates contextually within the development workflow, catching issues at the pull-request stage rather than after deployment. This proactive approach reduces friction and embeds accessibility into the engineering process.
Conclusion and Next Steps
GitHub’s general-purpose accessibility agent demonstrates that agentic AI can meaningfully improve digital inclusion. With a 68% resolution rate across thousands of pull requests, the experiment shows promise. The team plans to expand the agent’s capabilities, refine its detection of more nuanced issues, and share more detailed metrics. For organizations on their own accessibility journey, this case study highlights the value of focused, automated support—not as a replacement for human expertise, but as a powerful ally.