Business judgment on AI products
Every employee gets an isolated assistant, and the same one can still be driven together in a channel, with memories that do not leak.
01
Start inside the user's day · Public facts + observable behavior · 2026-08-28
Every employee gets an isolated assistant, and the same one can still be driven together in a channel, with memories that do not leak.
Public materials do not yet show how users complete this job today or what they replace.
The product targets friction in this job, but public user evidence does not yet show the cost, frequency, or consequence of leaving it unsolved.
Separate accounts scatter memory; one shared bot is too scary to grant real permission. The trend is assistants becoming company colleagues, not personal aides. The entry is per-person isolation inside a company, used in-house first then opened. Pricing undisclosed.
Its public repository has 13,884 stars and 1,648 forks, showing developer attention; repeat use and payment are not yet verified.
① Whether the fork-to-star ratio holds above 10% in three months — a drop means the attention; was spectating, not deploying; ② Whether any company outside the YC ecosystem says publicly that it runs this — this decides; whether it is an industry standard or an internal tool; ③ Whether third parties…
Worth dissecting. Its public repository has 13,884 stars and 1,648 forks, showing developer attention; repeat use and payment are not yet verified.
if a multi-agent video workflow has to go from "you use it" to "the team uses it," qm's scoping structure ports directly — one isolated workspace per person, one shared skill library, admin-defined model allowlist. That combination is far simpler than building a permission system and gets you the same outcome.
Not disclosed. MIT licensed, no pricing page, no hosted offering. ① Whether the fork-to-star ratio holds above 10% in three months — a drop means the attention; was spectating, not deploying; ② Whether any company outside the YC ecosystem says publicly that it runs this — this decides; whether it is an industry standard or an internal tool; ③ Whether third parties…
Every employee gets an isolated assistant, and the same one can still be driven together in a channel, with memories that do not leak.
Its public repository has 13,884 stars and 1,648 forks, showing developer attention; repeat use and payment are not yet verified.
An English validation note will follow from the public evidence.
The product claims to help users complete: “Every employee gets an isolated assistant, and the same one can still be driven together in a channe”. User evidence has not yet verified pain intensity or the cost of doing without it.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
02
Market comparison
The Chinese–English market comparison is not complete yet. A conclusion follows only after its coverage and verifiable evidence are recorded.
03
The call and next move come first; the full read retains the evidence and counterevidence.
Not another company-wide ChatGPT, but an agent scoped per person: every employee gets an isolated workspace, and in shared channels everyone still drives the same one.
Y Combinator's own engineering team (yc-software/qm, MIT). The product runs at
qm.ycombinator.com. They used it internally first and open-sourced it after.
Read: YC coordinates thousands of portfolio companies, hundreds of partners, and a pile of internal process every day. "Get the whole company onto one agent" is a more real pain for them than for most startups. Used-in-house-first is more credible than any pitch deck.
What it deliberately does not do: it is not a general assistant for individuals. The whole design assumes a company with many people in it. Alone, you carry the complexity for nothing.
Until now a company that wanted AI in real work had two options, both awkward:
qm fills in the missing middle: scope it per person instead of rationing permissions. Companies have been taping this together themselves — several months of one engineer's time, per company.
Not disclosed. MIT licensed, no pricing page, no hosted offering.
Read: YC is not trying to make money here. The return is the standard. If thousands of portfolio companies run the same agent substrate, YC owns the default position in that layer, which is worth considerably more than subscription revenue. This is open source used the classic way.
| Dimension | Call |
|---|---|
| Founder-product fit | Built for their own use first. The pain is their own — the strongest form of fit |
| Product insight | Saw that "personal assistant" and "company system" are different animals, and solved it by scoping rather than by permissions |
| Execution quality | Already in production; every substrate (harness, session store, sandbox, memory) sits behind an interface |
| Timing | Right on. Companies just finished trying AI and are now asking how to put everyone on one thing |
This is a textbook open-source land grab, aimed at the layer that is hardest to displace later.
13,257 stars and 1,547 forks in fifteen days is extraordinary for enterprise-facing infrastructure, which normally has no viral surface at all. There is one explanation: the people adopting it were already waiting for it. The fork count carries the signal — an 11.7% fork-to-star ratio means these are not applause, they are checkouts being configured for a deployment.
The transferable rule: when every company is taping the same thing together by hand, turning the tape into an open-source product wins you the standard, not just users. The barrier is not technical — multi-tenant isolation, Slack integration, sandboxes are each unremarkable. The barrier is who gets to be the default.
The cost is just as direct. Freedom to swap harnesses is bought with an abstraction layer, and every additional harness adds a convention the core can no longer change. When one harness makes a breaking change, qm either follows or loses those users. Worse, it only serves companies with a lot of people in them; for a team under ten it is a net liability. The design itself amputates the low end of the market.
The bigger question: does the agent-platform layer end up winner-take-all like Slack, or does every company roll its own like CI? qm is betting on the former, and YC's position in the ecosystem earns it the right to make that bet.
① Whether the fork-to-star ratio holds above 10% in three months — a drop means the attention was spectating, not deploying ② Whether any company outside the YC ecosystem says publicly that it runs this — this decides whether it is an industry standard or an internal tool ③ Whether third parties contribute non-default harness adapters — that is the test of whether "not tied to any vendor" is real
Product logic: if a multi-agent video workflow has to go from "you use it" to "the team uses it," qm's scoping structure ports directly — one isolated workspace per person, one shared skill library, admin-defined model allowlist. That combination is far simpler than building a permission system and gets you the same outcome.
Pricing structure: none. There is no pricing.
Strong pick. Not because the execution is exceptional, but because it demonstrates a path: use open source to take the default position in a layer that is just now forming — a layer that will certainly be collecting rent three years from now.
05
Use these searches when the official site is missing or the current link is only a lead.