Harness rebuilds source control for the agent-written code flood
Generating code is no longer the hard part — reviewing, approving, and shipping the flood an agent produces is. That's the problem the delivery platform Harness is now attacking, and it means the next AI-boom battleground isn't the IDE, it's the gate.
Harness launched a Code Repository and an AI Code Review layer built for what it calls "agent-ready development," betting that the entire software delivery cycle has to become autonomous. The company's argument, laid out by co-founder and CEO Jyoti Bansal, is blunt: most repositories were designed fifteen years ago for a world where humans write code, open a pull request, and wait hours or days for colleagues to review it. AI agents that can produce volumes of code in minutes now crush that model. Search and history slow as indexing falls behind, pull requests pile up faster than anyone can read them, and a permission system built around a list of developers has no good answer for a machine that can merge code on its own.
The Code Repository is a source-control layer scale-tested to handle thousands of concurrent pull requests and commits — the workload of a team of hundreds or thousands of agents working around the clock. Each agent inherits its permissions from the human who triggers it, scoped down to a specific repository, branch, or environment, so the human keeps audit responsibility. Review runs at merge: teams set mandatory AI checks once per account or tune them per project, and anything that fails gets bounced back with suggested reviewers and one-click remediation. The whole lifecycle runs programmatically, so agents can create, reply to, and resolve comment threads without opening a browser.
The deeper move is where the review authority comes from. Harness plugs its code review into an SDLC Knowledge Graph — every policy the team enforces, everything that has failed in production before — so the gate judges a change with full enterprise context rather than a fresh set of eyes. Harness has been running both features internally for months and says its teams have saved more than 10,000 hours of manual review time per month. A human still has to decide what ships; this is infrastructure to make that decision an informed one.
Why it matters: the coding-agent story has been dominated by how you generate code — Slack Code, Cursor, Claude Code — but the constraint has shifted to what happens after. GitHub and GitLab built review and permissions for human-paced workflows; a vendor that rethinks the repository, the review, and the governance as one system is positioning itself for the moment when code itself stops being the bottleneck. Bansal says it outright: you don't solve this by bolting AI onto a fifteen-year-old repo. If autonomous delivery takes hold, the approval gate becomes the real moat.
What to watch: whether this becomes a land grab — every code-management incumbent now racing to prove it can govern agent-produced code, and whether "AI code review" earns trust or just rubber-stamps volume.
Does your team review agent-written code with any confidence, or is it all shipping on vibes? Tell us in the comments.
Sources: Harness (PR Newswire) · SiliconANGLE · SD Times
