Harness rebuilt its Git repository for nonstop AI agent traffic
The New Stack Frederic Lardinois
Harness rebuilt its Git repo for AI bots that never clock out. It’s aimed at the pull-request pileup nobody’s testing team can keep up with.
Based on reporting by The New Stack, Frederic Lardinois — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Harness is betting that the real bottleneck in AI coding isn’t generation. It’s review. Martin Reynolds, one of the company’s field CTOs, says engineering leaders keep running into the same awkward answer when he asks how they’re handling the flood of pull requests from coding agents: they’re not keeping up.
That problem showed up early in Harness trials of GitHub Copilot and Amazon CodeWhisperer. More code was coming in, but the reviews were stalling, and the test team was getting buried. Reynolds says the increase was once 1.5x to 2x, and now he sometimes sees teams at 10x, with some claiming 50x. In hallway conversations at the conference, he says, drowning in pull requests came up again and again.
Harness’s answer is two products launched together: a rebuilt Code Repository and a new AI Code Review tool. The company says it spent much of the last 12 months building a software delivery knowledge graph, mapping pipelines, deployments, incidents, and policies so the reviewer can get context quickly without burning tokens. One example it gives is a migration flagged because an earlier incident review had found an unindexed CREATE INDEX statement that locked a production table for 14 minutes.
The bigger argument is about what should and shouldn’t be automated. Reynolds thinks reviewers should focus on what’s actually changing, not the scaffolding around it, and he’s blunt that some parts of the pipeline still need to stay deterministic. Test results, for example, still come from the test runner. The point isn’t to tear that out. It’s to make the human part smaller and smarter.
Harness also rebuilt its repository for a different kind of traffic. The company says the new system is a ground-up AI-first repository for humans and agents, built on Kubernetes, spread across multiple clouds and regions, and tested at thousands of commits per second. It’s already in beta with about 20 enterprise customers. That matters because GitHub’s own August 17 outage showed what happens when infrastructure built for people gets slammed by traffic peaks — and agents don’t keep office hours.
My take — AI-written commentary, not fact-checked reporting
The industry keeps pretending review is a side quest. It isn’t. If the bots are going to shovel out 50x more code, the first winners won’t be the loudest model demos; they’ll be the unglamorous tools that stop human teams from drowning in their own prompts. That’s the boring truth, which is usually the right one.
Read more about this at: The New Stack