How GitHub Deals with 17 Million Pull Requests a Month

TL;DR

• 1

  Agent traffic has gone vertical: Kyle Dagel says GitHub saw 17 million agent-created pull requests in March alone, and after reporting 1 billion commits for all of last year, the platform is on track for roughly 14 billion this year if growth stayed linear.

• 2

  GitHub thinks the PR bottleneck is fixable with more automation, not less AI: Dagel points to agentic code review and agentic merge, where Copilot can find vulnerabilities, respond to comments, wait for CI and policy checks, and finish the tedious last mile of pull requests.

• 3

  Open source maintainers need control, not a one-size-fits-all standard: Instead of forcing a universal workflow, GitHub is building controls so communities can decide who gets to contribute and under what conditions, citing Mitchell Hashimoto's vouch system as one approach that not every project wants.

• 4

  The real pricing problem is model choice and runaway agent usage: Dagel argues that keeping a $200 coding subscription from turning into $2,000 will depend on better model routing, task-intent based selection, and using cheaper models for simple steps like renaming variables instead of defaulting to the most expensive model every time.

• 5

  GitHub's long-term bet is personalized agents with memory and context: He says the durable theme from ChatGPT to OpenClaw has been personalization, meaning agents that understand your coding habits, preferences, and unfinished thoughts without you manually spelling everything out.

• 6

  One of the most practical AI use cases here is brutally personal: Dagel reveals he runs a recurring loop where an agent reads his emails, Slack messages, and interviews, then tells him what he keeps saying badly and whether he actually followed its feedback from the previous week.