GitHub’s agent traffic grew 1400% in 2026. That’s not a typo. And it’s pushing the world’s biggest developer platform to its limits. GitHub COO Kyle Daigle sat down with Latent Space to explain what happens when AI agents stop autocompleting code and start shipping it. If you run a SaaS company, this is your infrastructure story too.
The short version: everything GitHub built for human developers moving at human speed is now getting hammered by machines moving at machine speed. Pull requests, CI/CD pipelines, code review, dependency management. All of it needs to change.
What happened
- GitHub COO Kyle Daigle went on the Latent Space podcast to lay out GitHub’s plan for handling the explosion of agent-generated code.
- Agentic coding on GitHub grew 1400% in 2026, creating strain across infrastructure, code review, and CI/CD systems.
- GitHub hit 1 billion commits in 2025. It’s now seeing 275 million commits per week, putting it on pace for 14 billion this year.
- GitHub Actions went from 500 million minutes per week in 2023 to 1 billion minutes per week in 2025, and it’s still climbing.
- The platform has had notable uptime issues as a result, drawing public criticism from the developer community.
- GitHub is rethinking pull requests (calling them “prompt requests”), trust models for open source, and how CI/CD works when agents are the primary committers.
The numbers
- 1400% growth in agentic coding activity on GitHub in 2026
- 275 million commits per week (up from 1 billion for all of 2025)
- 14 billion projected commits for 2026 if growth stays linear (it won’t, says Daigle)
- 1 billion GitHub Actions minutes per week in 2025, up from 500 million in 2023
- 200 million+ developers on the platform
- 14x commit growth pushing databases, monorepos, and availability to the edge
5 things SaaS operators should learn from GitHub’s agent growing pains
- Your infrastructure was built for humans, not agents. GitHub designed its systems for developers who commit code a few times a day. Agents commit hundreds of times. If your SaaS product has rate limits, queue systems, or compute allocation based on human usage patterns, those assumptions are about to break.
- CI/CD is becoming a general-purpose compute layer. GitHub Actions started as a way to run tests and deploy code. Now it’s the execution environment for agents. Daigle described it as a general-purpose compute layer. If your product plugs into CI/CD, you’re sitting on a much bigger opportunity than you think.
- Supply chain security gets way harder with AI-generated code. GitHub’s acquisitions of npm, Dependabot, and Semmle were about securing the dependency graph. When agents generate code that vendors dependencies, forks repos, and creates packages at scale, every link in that chain becomes a potential vulnerability. You need validation tools, not just scanners.
- Code review is getting reinvented. Pull requests are becoming “prompt requests.” When most PRs come from agents, the review process shifts from reading diffs to vouching for the agent and auditing the prompt. If your SaaS product touches developer workflows, this is a design problem you need to solve now.
- The definition of “developer” is expanding fast. GitHub has 200 million+ developers. AI is lowering the barrier to building so dramatically that leadership, marketing, and ops people are writing software again. Daigle himself went from COO back to active coder thanks to AI. Your user base is about to get a lot bigger and a lot less technical.
The hot take
GitHub’s uptime problems aren’t a bug. They’re a preview. Every SaaS platform that serves developers (or agents acting on behalf of developers) is going to hit this wall. The companies that survive will be the ones that treat agent traffic as a first-class infrastructure concern right now, not after their status page turns red. If you’re a SaaS operator and you haven’t stress-tested your platform against 10x agent-driven load, you’re already behind.
The Agency OS play
If you run a SaaS company, this week is the week to audit your infrastructure assumptions. Pull up your usage data and look at API call patterns over the last 90 days. Are you seeing spikes that don’t match human working hours? That’s agent traffic. Figure out what percentage of your load comes from automated tools versus humans. Then model what happens when that number doubles every quarter.
Next, look at your CI/CD pipeline. If you’re using GitHub Actions (and you probably are), start treating it as a compute platform, not just a deployment tool. Build workflows that validate AI-generated code before it hits your main branch. Add automated checks for dependency integrity. If you don’t have a policy for how agent-generated pull requests get reviewed and merged, write one this week. It doesn’t have to be perfect. It has to exist.
Finally, think about your product. If your SaaS serves developers, you need an agent-native tier. That means API endpoints optimized for high-frequency automated calls, webhook infrastructure that can handle 10x current volume, and rate limits that distinguish between a human clicking buttons and an agent making 500 requests per minute. The platforms that build for agents first will own the next wave. The ones that bolt it on later will spend the next two years playing catch-up.