GitHub tackles vibe coding chaos with a control center for AI coding agents – News

Microsoft Corp.’s GitHub unit is trying to get a grip on the growing mass of uncoordinated artificial intelligence coding agents, launching a new control center known as the Agent HQ. It’s meant to serve as a kind of “mission control”, where software developers can manage multiple third-party coding agents in one place.

These coding agents are tools that are used by developers and software engineers to complete programming tasks autonomously with only minimal supervision. They’ve become incredibly popular in the last couple of years, giving birth to the concept of “vibe coding”, where human developers simply prompt AI to write code on their behalf and review it afterwards to try and accelerate their productivity.

Numerous companies, including GitHub itself, have released coding agents, and the result is that some teams are managing several of them at once. But when users have multiple agents all working on different tasks, it takes a lot of effort to keep track of them all, and that’s what GitHub wants to help with.

The GitHub Agent HQ gives developers a place to manage AI coding agents from companies including OpenAI Group PBC, Google LLC, Anthropic PBC, xAI Corp. and Cognition Inc., along with GitHub’s own Copilot, all in one place.

GitHub Chief Operating Officer Kyle Daigle told CNBC that the idea is to try and bring a bit more order to the “chaos of innovation” that has arisen from juggling so many agents. “With so many different agents, there’s so many ways of kicking off these asynchronous tasks, and so our big opportunity is to bring this all together,” he explained.

Agent HQ provides users with a command center from where they can assign new tasks to agents, steer them and monitor their work. They’ll be able to use it to observe what each agent is working on in real time, and correct its course if they see it going off track or doing something it shouldn’t. It will also help developers to see which agents are more effective. For instance, they can assign each supported agent the same coding task and have them all do it in parallel, so they can compare the results.

There are security benefits to having a command center too. Whereas standalone coding agent implementations require users to grant broad repository access, the Agent HQ allows granular controls to be implemented at the platform level. In this way, the third party agents will inherit the enterprise security and governance of GitHub Copilot. This means they’ll run within sandboxed GitHub Actions environments secured by firewalls and operate under strict identity controls, the company explained. Even if an agent goes rogue, the firewall will stop it from accessing external networks or stealing data.

GitHub said Agent HQ will launch initially for Copilot Pro+ subscribers later this week, with support for OpenAI Codex. In the coming months, it will add support for the other coding agents.

In addition to Agent HQ, GitHub announced a new Plan Mode in Microsoft’s Visual Code Studio platform that uses GitHub Copilot to create step-by-step plans for tasks before it goes ahead and executes them, so users can review it first. It’s also added a code review feature to Copilot that allows it to access tools such as CodeQL and evaluate its own code before passing it to a human developer.

Image: GitHub