GitHub adds 4 million cores as AI agents break its services

At its yearly conference in San Francisco this week, GitHub said it will add 4 million cores, which are processor units, and 160 petabytes of storage. It will also rebuild its backend, the part of the system that users do not see. The leaders of GitHub say that AI agents use the platform in a new manner and cause failures. Days before, a previous chief executive of GitHub launched a rival, Entire.

How it works

Kyle Daigle, chief operating officer of GitHub, says failures began at the start of summer. They came in services that scaled correctly before. Some core services must be rewritten, and GitHub will divide compute from storage in its Git systems.

Jay Parikh, who leads GitHub, says an increase of 5 to 10% is sufficient for only a week or two.

The numbers

GitHub serves more than 225 million developers, says Daigle. The source gives no numbers on how many failures there are or how long the work takes.

The other side

Daigle says the world does not need a new GitHub. Daigle says rivals make a solution for only a part of the problem.

Thomas Dohmke, who leads Entire, says developers clearly want one. Entire is open source. The Microsoft investment arm M12 funds Entire.

Insight: Microsoft is on both sides of the GitHub fight

GitHub Enterprise users pay for the AI credits that agents use, and agents cause failures at GitHub. The Microsoft investment arm M12 also funds Entire, a rival. This reads as a plan for the two sides: Microsoft can get a result if agent work stays on GitHub or moves to Entire. For leaders, one host is then a risk to plan for.