Harness tackles influx of agent-delivered code with Code Repository and AI Code Review

Harness Inc., a software delivery platform provider, has introduced the Agent-Ready Harness Code Repository and AI Code Review to address the growing need for managing the increasing influx of code produced by AI agents. These new tools aim to optimize the software development lifecycle (SDLC) for teams that are rapidly adopting AI coding agents.

As AI agents generate code at a faster pace than teams can write, review, test, and deploy it, the focus is now shifting towards the storage, review, approval, and shipping of code without disrupting the system.

Co-founder and Chief Executive Jyoti Bansal highlighted that traditional code management, which expects humans to write code and open pull requests while colleagues fine-tune, test, and approve over hours or days, is being overwhelmed by AI agents that can produce large volumes of code in minutes or hours. This shift requires a reimagining of code management systems, as the previous permission and code-keeping systems, designed for hours or days of work, cannot handle the accelerated SDLC.

Harness is rebuilding its code management layer to cope with the new scale and requirements of AI agents. Bansal emphasized that most enterprise teams are attempting to integrate AI agents into their existing code repositories without proper organization or future-proofing for machine readability and rapid approval processes. The company believes that the entire SDLC must become autonomous, including the repository, review, pipeline, and governance, to achieve seamless integration and operation.

The Harness Code Repository provides a scalable solution, designed to handle thousands of pull requests and commits simultaneously, allowing teams of hundreds or thousands of agents to work efficiently without causing system breakdowns. The system offers enhanced search, history, and comparison capabilities to run at high volumes. Each agent is granted permissions inherited from the human that triggers them, preserving the human's responsibility for audit trails.

To facilitate programmatic operations, Harness has tailored the system to use Model Context Protocol and command-line interfaces (CLI). This enables the full software delivery lifecycle to run automatically, such as finding reviews by the author's email, pulling all pull requests across multiple repositories, and managing comment threads via the CLI. Using the CLI, agents can interact with the system directly, reducing AI token costs.

Code review functionalities have been streamlined for large-scale use, allowing agents to evaluate code requests similar to how a human would. The system checks code at merge time, enabling teams to define mandatory AI checks at the account level or customize them by project. Any rejected changes are promptly returned to the team for updates, with feedback highlighting the consequences rather than the specific lines modified.

The system also provides suggested reviewers and labels to simplify one-click remediation, ensuring modifications can be merged without much difficulty.

Harness has implemented these new capabilities internally and reported significant time savings for teams, with an estimated 10,000 hours saved in the last month. The company aims to support the SDLC transformation, ensuring that while AI agents can generate extensive code, human decision-making remains crucial for determining what code ships to production.