Harness tackles influx of agent-delivered code with Code Repository and AI Code Review
SiliconANGLE Kyt Dotson
Harness launched new tools for AI-written code: a repository and review system built for agents. The bet is that software teams now need infra for code bots, not just more code bots.
Based on reporting by SiliconANGLE, Kyt Dotson — 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 trying to rebuild the plumbing around software delivery for a world where AI agents can spit out code faster than humans can sort through it. The company on Tuesday announced Agent-Ready Harness Code Repository and AI Code Review, two products aimed at teams that are already using coding agents at a growing pace.
The argument from co-founder and CEO Jyoti Bansal is blunt: the old workflow assumes people write the code, open pull requests, and then wait hours or days for colleagues to review, test, and approve. That rhythm made sense when development moved at human speed. It starts to fall apart when agents can create large volumes of code in minutes or hours.
Harness says the fix is to make the entire software delivery cycle work as one system. That means source control, review, the pipeline, and governance all need to line up, rather than teams bolting an agent onto a repository that was built for another era. Bansal also pointed out that many repositories were built more than 15 years ago, before anyone was thinking about machine readability or near-instant request handling.
The new repository is designed to handle thousands of pull requests and commits at once. Search, history, and comparisons are supposed to keep working at that scale, and each agent gets permissions inherited from the human who triggered it, down to the repository, branch, project, or environment. Harness also built the system around Model Context Protocol and command-line interfaces, so agents can work programmatically without constantly opening a browser.
AI Code Review follows the same logic at the gate. It checks changes at merge, lets teams decide which AI checks are mandatory, and rejects anything that fails. The review feedback is meant to explain what is at stake, not just point at moved lines, and the system includes suggested reviewers and labels to make fixes easier. Harness says it has been using both features internally for months and estimates the early work has saved 10,000 hours over the last month.
My take — AI-written commentary, not fact-checked reporting
This is the right fight to pick: not “how do we add AI to dev tools,” but “how do we stop AI from turning software delivery into a junk drawer.” The industry keeps pretending the repo is fine and only the coding assistant needs a tune-up. That’s a nice way to end up with faster chaos.
Read more about this at: SiliconANGLE