AI Pulse · AI Trends Pulse

Y Combinator just open-sourced the software it uses to run its own operations. It’s called QM, and it’s a multiplayer agent harness that lives inside Slack. Different employees and rooms get scoped memory and permissions, and the system spans accounting, legal, events, and engineering work. The code is MIT-licensed and pulled 638 upvotes plus 150 comments on Hacker News the same day it dropped.

This matters because YC is a credible peer, not a vendor pitching vaporware. They built this to solve their own problem, ran it long enough to trust it with real company functions, and now they’re saying the quiet part out loud: you can wire agents into the actual operating rhythm of a business. Not a chatbot for one task. A layer that touches multiple departments, remembers context, and respects who should see what.

The reaction on Hacker News tells you the category is real. Builders are asking how permissions work, what the memory model looks like, whether it scales past a certain team size. Those are implementation questions, not “is this even possible” questions. The shift is already happening.

For operators, the takeaway is straightforward. If you’re still thinking about AI as a side project or a single tool, you’re behind. The companies moving fastest are treating it as infrastructure, something that sits across workflows and learns as the team uses it. That’s exactly the kind of system we build into an AI command centre, tailored to how your business actually runs. QM proves the approach works. Now the question is whether you build it yourself, stitch together open-source pieces, or get something purpose-built that doesn’t need a team of engineers to maintain.