nithilan karthik nithilan@managent.tech UPDATES: new research

01

managent

Slack for AI agents. A control plane so a fleet of coding agents can share one repo without stepping on each other. 5 design partners, and 200+ engineers using it.

01

the arc

It started at a hackathon. I was building with other people, and we were all using AI to write code. When we tried to merge those branches into the same repo, we hit a pile of merge conflicts. Files overwriting each other. Work that looked fine on one laptop broke the moment it landed next to someone else’s. The agents were smart. Sharing one codebase was the hard part.

That stuck. So I went and measured it. With co-authors, I studied 33,596 agent-authored pull requests across 2,807 repositories. The paper is on arXiv: AI Agent Pull Requests on GitHub: Frequency, Structure, and Merge Conflict Rates.

On exact overlap in time, 40.2% of repositories already have co-active agent-authored PR pairs, and those pairs are 79.4% of all agent PRs. Widen the window to a week and it’s 53.4% of repos and 95.0% of PRs. Cross-agent pairs conflict at 41.7%, against 19.8% for the same agent working alone.

So I started building managent.

02

what it is

One control plane for a fleet of coding agents. Before an agent starts on a file, it books that territory, and everyone else waits. You can’t collide on something you’ve leased.

The reason to build it is the enterprise case: different devices, different agents, large monorepos. A laptop, CI, and the cloud. Claude next to Cursor next to Devin. Hot files that everyone wants at once. Agents lease territory before they write, the queue stays visible, and every merge gets replayed and checked before it ships.

Inside respected leases, measured merge-tree replays have come back with 0 conflicts. Uncoordinated concurrent pairs still conflict about 1 in 5. 5 design partners are on it. 200+ engineers are using it.

03

managent’s thesis

The future isn’t about making better agents. It’s about making agents work better together.

04

why it matters

People have spent years trying to make one agent smarter. A sharper agent still has to share the repo. If it can’t, it just breaks someone else’s code faster.

About 90% of developers now use AI coding tools. 74% have a specialized coding agent. The open question is whether a team of humans and agents, across devices and vendors, can share a large monorepo without quietly wrecking each other’s work. That’s the layer managent is.