Key Takeaways

  • Your engineering team's comfort making frequent changes directly links to CI speed; slow CI creates “monster changes” no one wants to review.
  • AI agents supercharge this effect: they generate changes around the clock, amplifying the bottleneck of a slow Continuous Integration pipeline exponentially.
  • Ryan Nystrom, an engineering manager at Notion, highlights the difference between a CI loop that takes an hour and one that takes three minutes. The latter unlocks orders of magnitude more output from both humans and AI.
  • With AI agents capable of generating thousands of pull requests weekly, Claire Vo argues that a slow CI means you might as well be “throwing all those PRs in the trash.”

Your Old CI Speed Won't Cut It Anymore

Think about how your engineers currently approach changes. If your CI pipeline takes forever, they probably group many small updates into one giant, scary pull request. Notion's engineering manager Ryan Nystrom explains this human behavior plainly. “If it's slower, then you're building up these like monster changes and you're going to be even slower about like judiciously like reviewing every little like tiny thing.” No one wants to touch those. They sit, they gather dust, and they often lead to more complex merge conflicts and bugs down the line.

Nystrom argues that the faster your CI completes, the more comfortable engineers feel making and pushing changes. It's a psychological shift. They know they can quickly get signal on their work, whether pushing to dev or production. This rapid feedback loop encourages smaller, more frequent commits, leading to a much healthier codebase and less cognitive overhead for individual developers.

AI Agents Turn Slowness Into a Catastrophe

Now, add AI agents to the mix. What was once a bottleneck for human engineers becomes an absolute choke point. AI agents don't get tired. They don't need coffee breaks. They can work on a VM while you're sleeping, churning out code and fixes. Nystrom calls this the human problem, but “on steroids.”

Imagine an AI agent generating hundreds, even thousands, of potential pull requests a week. Each one needs to pass through CI to get validated. Nystrom paints a vivid picture: “If I've got a CI loop that takes an hour to run, that, your agent's just going to sit there and spin for like an hour waiting for results to like do something. If it takes three minutes to run, like holy crap, how much more stuff are you you as a human and then especially as your like little swarm of agents going to be able to get done?” The difference isn't linear; it's exponential.

Claire Vo drives this home: you cannot integrate AI agents effectively if your CI is slow. It's not just inefficient; it’s catastrophic. “You might as well be throwing all those PRs in the trash,” she says. The sheer volume of AI-generated work means that every minute of CI delay is multiplied across potentially thousands of autonomous changes, effectively nullifying the productivity gains AI promises.

What to Do With This

This week, measure your median CI time for feature branches. Don't just look at the average; find the typical experience. If it's over 10 minutes, set an aggressive, short-term goal to cut it in half. Assign an engineering lead the sole responsibility for this mission, giving them direct access to resources. This isn't a nice-to-have; it's the foundational investment required to unlock any meaningful AI-powered engineering gains.