Harness Launches Agent-Ready Repository and AI Review for High-Volume Code Changes

Harness has launched an agent-ready code repository and AI code-review product built for a delivery problem that emerges after code generation: agents can create changes faster than conventional pull-request workflows can absorb. Its answer combines high-volume repository operations, inherited access controls and merge-time checks, while keeping people responsible for what ships.

The constraint shifts from writing code to processing it

Harness Code Repository and AI Code Review launched August 27. Chief Executive Jyoti Bansal said established code-management routines assume a person writes code, opens a pull request, then waits for colleagues to adjust, test and approve it over hours or days. He argues that model is strained when agents generate in minutes or hours work that once took developer teams days or weeks.

That framing distinguishes the launch from a new code-writing tool. Harness is aiming at the repository, review and approval path that follows generation, with agent access tied to the human who started the work.

Throughput comes with a permission boundary

The repository is designed for thousands of simultaneous pull requests and commits, with search, history and comparisons operating at that volume. Each agent inherits permissions from its human trigger, scoped to a repository, branch, project or environment. Harness says the human initiator remains responsible in the audit trail.

How agents can work in the system

• Use Model Context Protocol and command-line interfaces to run delivery workflows programmatically rather than through a browser.
• Find reviews by an author’s email, collect pull requests across repositories, and create, reply to or resolve comment threads through the command line.
• Operate only within the repository, branch, project or environment permissions inherited from the human trigger.

A merge gate, not autonomous deployment

AI Code Review checks code at merge. Teams can make selected AI checks mandatory across an account or tune them by project. A failed check rejects the change for updates, while suggested reviewers and labels are intended to speed remediation.

Harness describes that as a decision-support gate, not a handoff of production authority. AI Code Review is meant to identify changes ready for staging and those needing further action; a human still decides what ships.

An internal efficiency figure

Harness says it used both capabilities internally for months and estimates teams saved 10,000 hours in the last month. The figure is a company-supplied internal estimate.