YC's QM Got 12k Stars in a Week. It Still Doesn't Solve the Problem.

On August 1st, 2026, YC launched QM, their internal multi-agent harness, and it hit 12k stars within a week. I installed it on a VPS for my team, and here's the thing: it looked eerily similar to what we've been building with Zodus.

This post is my take on what QM got right and what it got wrong.

First, a preface: QM is a genuinely good tool and you should try it. It's an agentic harness that gives everyone in an org their own OpenClaw-like agent at a fraction of the cost, with the added benefit of shared context.

The story behind it is simple: managing that many OpenClaw instances got too hectic (and probably too expensive) for YC, so they built their own tool to do it better.

The problem is pretty straight forward.

Every OpenClaw instance needs its own machine. Sure, you could run multiple agents on one box, but then:

• How do you manage memory?
• How do you keep files isolated?
• How do you make sure an agent doesn't "escape" its sandbox? (I'm looking at you, OpenAI, Anthropic and Meta 👀)

Even if you're a 1000x agentic engineer, custom plumbing is something you have to maintain. And with the release cadence of OpenClaw and Hermes - good luck.

The solution is straightforward in theory: build your own primitives around the agentic loop, then compose those into something that scales at the org level. That's exactly what YC did. They built everything a cloud harness needs: a cron system, skill storage, file storage, connectors to major tools (Slack, GitHub, Notion) and made it work with established harnesses like Pi, OpenCode, and Codex.

So what's the problem?

Two things.

Nobody knows what QM actually is. It launched a week before Buzz, and the perception stuck that QM was, like Buzz, a venue for agents to hang out. The reality is the opposite: QM is the host agents live on, not the place they gather. One launch post wasn't enough to make that distinction land.

QM is not proactive, and it doesn't solve the context problem. It only works if you keep working it, inside its web portal, or by actively tagging it in Slack. The agent waits for you. That's the gap that matters. Agents that wait to be asked are tooling; agents that notice are leverage. Closing that gap means persistent understanding of what the company is doing, and agents willing to act on it without being summoned. That's the hard part, and it's still unsolved by QM, and honestly, by everyone.

For what it's worth, QM is built right by people who know the problem well. I'm excited to see where it goes. But as of today, it has a long way to go. And so does the category.