Back to All
Zero

Zero deploys 2x faster with Avrea

Faster GitHub Actions workflows let Zero keep running the full test suite on every change as agents produce more code. Engineers wait less, and the team can ship quickly while building a reliable product.
A beautiful Finnish sunset over a body of water.

Results snapshot

One PR typecheck: waiting time cut from five minutes to one
Roughly 2x faster API deployments, with no application code changes
Slowest staging deployment: p90 duration down by two-thirds
More than 2x production shipping frequency over a few months

About Zero

Zero builds an AI-native CRM that brings lead generation, signal tracking, outreach sequencing, and customer management into one workflow driven by agents. Its engineering team uses agents to build the product, too.

The application has a Node API layer with Zero’s own ORM and workers that run agent workloads. Everything ships as Docker images to Google Cloud, with Cloud Run for the API and GKE for the workers.

  • Industry: CRM and sales software
  • Stack: TypeScript monorepo, Node.js, Docker, GitHub Actions, Google Cloud Run and GKE

A growing test suite put more work on CI

The majority of Zero’s code now originates from agents. Code review has also become agentic, and most of the test code is written by agents. Zero’s test suite has grown more than tenfold in a few months.

Every change runs the full suite. As the team created more pull requests, it also needed more CI runs to validate them. The speed of those runs determined how soon developers could get feedback on their changes.

“The thing we care about is how fast a pull request gets feedback,” Olli says. “All of that work now lands on CI, so CI is what determines how quickly a PR gets an answer.”

Zero wanted faster feedback from its existing GitHub Actions workflows. The team looked at Avrea because it could change runners without reworking its build and deployment process.

A runner change across sixteen workflow files

Zero migrated with one pull request, changing ubuntu-latest to avrea-ubuntu-latest across sixteen workflow files. The team kept its existing actions and Docker builds and added no cache configuration. “Avrea promised a drop-in replacement and a 2-3x speedup, and that’s what we got,” Olli says.

“Faster is only worth it if it is also boringly reliable,” Olli says.

“We also asked what the exit looks like before we moved. With Avrea, it is the same one-line change in reverse. We have never needed it, but knowing the door is open is part of why we moved everything at once.”

Faster checks and more frequent shipping

API deployments became roughly twice as fast without changes to the application code, and admin deployments improved by a similar factor. The p90 duration of Zero’s slowest staging deployment dropped by two-thirds.

The difference was particularly clear on a typecheck job that ran on every pull request. Developers had been complaining about the wait on GitHub-hosted runners. Moving that job to Avrea cut the wait from five minutes to one.

Zero’s production shipping frequency has more than doubled, while CI runs have more than tripled over a few months. Its workflows have become faster over the same period.

Zero runs the full test suite on every change as the volume grows. Fast, affordable CI makes it easier to keep that validation routine when working with agents.

“When CI is cheap and fast, every change runs the full suite without anyone thinking twice”

Olli Jokinen, Zero

More case studies: