Harness launches Agent-Ready Code Repository and AI Code Review to manage AI-generated code

According to the company, AI agents now generate code at a pace that exceeds a team's ability to review, test and deploy it. The shift is moving the challenge to storing, reviewing, approving and shipping code without breaking down. Harness co-founder and CEO Jyoti Bansal said that current code management systems were designed for human-scale workflows, with developers writing code and opening pull requests that colleagues would adjust and approve over hours or days. That model is being overwhelmed by AI agents that can produce in minutes what would have taken a team days or weeks.

Bansal stated that "software delivery is going through its biggest shift since the move to the cloud," and that existing systems were built for a different scale and a different kind of user. He argued that simply attaching an AI agent to a current repository would not work, since most repositories were written over 15 years ago without consideration for machine readability or rapid request resolution. He added that the entire software development lifecycle must become autonomous, with repository, review, pipeline and governance operating as a unified system.

The new Code Repository provides source control that can handle thousands of pull requests and commits opened at once, allowing large teams of agents to work without blocking. Search, history and comparisons run at volume. Each agent inherits permissions from the human who triggers it, down to specific repositories, branches, projects or environments, keeping humans accountable. The system is designed to work with Model Context Protocol and command-line interfaces, enabling programmatic workflows such as finding a review by author email, pulling all pull requests into one place, and resolving comment threads without a browser. This also reduces AI token costs.

AI Code Review operates at large scale and allows agents to read code requests. It checks code at merge time, letting teams set mandatory AI checks per account or per project. Changes that fail checks are rejected and returned. Feedback from rejections highlights what is at stake and includes suggested reviewers and labels for one-click remediation. Harness emphasized that humans still decide what ships, and the review tool informs teams about production readiness.

Harness has used both capabilities internally for months. From early testing, teams saved an estimated 10,000 hours over the last month.