# What is YC doing? > YC open-sourced QM, the agent harness they run themselves. On what the release signals about how we'll build companies in the future (aka next week). - written by: Andrew Schroeder - edited by: Fable 5, GPT-5.6 Sol - status: published · 2026-08-02 - canonical: https://moltolabs.ai/notes/what-is-yc-doing/ - raw markdown: https://moltolabs.ai/notes/what-is-yc-doing.md --- That’s the question a friend and I asked in our text thread after they announced [QM](https://qm.ycombinator.com) the other day (pretty timely given [my earlier piece](/notes/alone-in-the-playground/)). For those who haven’t seen, QM is YC’s open source harness meant for companies. It’s basically positioned as OpenClaw or Hermes but for an entire org rather than just an individual. Rad. Let’s be clear — I haven’t actually played with this yet (again, see that same piece…this doesn’t feel like it’s really for me). Looking on Hacker News, some people are very excited. Others are pointing out perceived slop. I don’t know if it’s good. And that’s not what I want to talk about today. Instead, I want to ramble a little bit about what this product might indicate about how we build companies in the future (aka next week). My friend and I concluded that we, in fact, have no idea what’s going on at YC. Not in the “they’ve lost their minds” way, but rather “idk they’re pretty smart people and I’m sure they have a plan or maybe they’re just trying to build. what. people. want.” Potential plans: stay top-of-mind for builders, become the “OS for startups” company, world domination? I’m kidding…but am I? While I’m not sure about their motives, I do think it’s clear that the folks at YC are thinking a lot about what future orgs might look like. Garry and co have been very open about publicly talking about how they’re running their org and what they’re seeing the AI-pilled startups in their cohorts doing (and I think it’s pretty clear that AI-pilled is their MO these days — see [Paxel](https://paxel.ycombinator.com), their tool that reads your coding-agent sessions and tells you what kind of builder you are). So maybe while [Every](https://every.to) is using their position as a working/running company making content and building tools as a window into what companies of the future might look like, we can view these recent YC product releases as indicative of where they think work is going as well. Said that way I suppose it’s obvious. What does QM tell me (as someone who hasn’t even dug through the repo yet). Well, first I’ll applaud them for actually putting stuff out there beyond saying “multiplayer is cool.” Also, pretty cool name! I have a tendency to be cynical about pretty much everything but I’m not going to let that get in the way here. We are all figuring this stuff out, and more collective iterations = more better. So what is their version signaling to others? The bet I see them making is that entire orgs should exhaustively log what they’re doing in order to learn from it. I don’t think it’s really any different from the pre-AI days. It’s just that now having an actual vault of knowledge is way more powerful. Doing this allows companies to move faster and more efficiently. The implication, however, is that there’s a great deal of trust to do this. By that I mean I need to be willing to openly share everything with my colleagues. Pretty sure anyone who’s ever worked can see how this can be tricky. To their credit, a big chunk of QM’s design is scoping — every person and project gets their own walled-off workspace — so they clearly see the trickiness too. But a company brain is still a shared brain. Some other thoughts: - Damn. They’re getting pretty ambitious. And I think that’s something so cool about AI and very relevant to the future of work. They just threw a massive spectacle for startup school. They’re building a media empire (between podcasts and recycled talks from their various events). And now they’re launching products?! Fair to say YC has always positioned themselves as hungry and willing to expand, but it just feels like they’re moving with real pace. - It doesn’t matter. If this has staying power, that is. [Garry’s GBrain](https://github.com/garrytan/gbrain) (his open-source agent memory brain), Paxel (the vibecoding eval), now QM. They may or may not be the next big thing but they are indicative of what *mainstream* early adopters want. Better control over memory and skills. Introspection -> improved AI efficacy. Company AI that doesn’t act like a complete dum-dum. - Express yourself first, then code. Many on Twitter/X quickly latched on to their request for contributions to be simply describing what you want rather than code (the README’s actual words: contributions are taken as “human-written text, not code”). Hell yes. Really love the direction they went here and I think it speaks to something broader. If coding is solved, maybe we should give more thought to how we describe what we want (and why). LLMs are really good at language, after all. - Shared context — so hot right now. I view this as a really neatly bundled package of context. That’s not a knock, it’s what everyone has been driving towards with their personal setups (and what I did with my setup). Open question for me is if a massive monoculture of context is helpful or actually becomes fragile. Regardless, it’s where we are and my experience (and sounds like YC’s) is that the enforced alignment certainly helps things move at breakneck speeds. - QM still doesn’t kill my pain. Ok — I haven’t tried it yet so this take is premature, but I don’t see all my friends suddenly committing to sharing a harness. This feels like a very cool tool for a young org, not something that will help me collaborate with peers working outside the walls of a single company. Cool, but not a network building tool. So I’ll wrap here with my shameless Corner Table plug. QM and [Buzz](https://block.xyz/inside/introducing-buzz-where-humans-and-agents-work-together) (Block’s humans-and-agents workspace from a couple weeks back) are awesome developments…it’s been a good couple of weeks. I need to check them out and you should too. But if you want something different — something that lets you learn and share without opening the gates to your own “brain” — please reach out and I can share Corner Table with you (there’s a waitlist at [cornertable.ai](https://cornertable.ai)). We’re in closed beta right now. That means we’re still figuring things out, but it is fun seeing what agents can learn from other agents (**with different context**) when we start to cut the humans from the loop.