A start-up raises $28.5M to build AI-run “autonomous companies” — the question that matters is not autonomy, it is who decides what counts as “sensitive”
On 6 August, Naïve Labs raised $28.5 million in Series A funding (Nexus Venture Partners, with Y Combinator, Zetta, Liquid 2) for infrastructure that lets AI agents run a company end to end — legal incorporation, virtual cards, email, compute, memory — from a single configuration. Notable for a product selling total autonomy: the platform itself ships capability policies, an immutable audit log and mandatory human validation before any “sensitive action”.
Naïve's product is not yet another enterprise chatbot: a unified serverless infrastructure meant to carry an entire company — legal incorporation, virtual card provisioning, email and phone, compute, memory and multi-agent coordination — driven from a single API. Abhishek Sharma, of lead investor Nexus Venture Partners, sums up the ambition bluntly: “The next decade is about autonomous companies. Naïve gives millions of entrepreneurs and small businesses the turnkey infrastructure to build and run autonomous companies.” CEO and co-founder Sean Dorje targets token efficiency rather than autonomy for its own sake: “making every token do more, so that autonomous companies become a profitable reality.” What no headline in the release highlights, but the product sheet does confirm: the platform itself enforces capability policies, an immutable audit log, and requires human validation before an agent executes an action deemed “sensitive”.
One could read this as vindication of the choice we made from the start — even a product selling “the autonomous company” builds a human validation gate. But the honest comparison stops where Naïve's release also stops: from the outside, we do not know who defines what counts as “sensitive” on their platform — an agent, a configuration threshold, the customer? What we can verify, however, is our own code. In AppManager, it is never the agent that judges an action harmless enough to skip approval — the very structure of the product imposes it: a quote stays a draft until the manager clicks “send”; an Automations rule, fixed or defined by the customer themselves, only ever opens an event to handle, never an order or an email executed on its own (the same check we ran on 7 August against Epicor Prism, same principle).
For AppH
- The bet on mandatory human validation, which we have defended from the start, is no longer a niche position: even a product explicitly designed for “the autonomous company” builds that gate before any sensitive action. The market is converging on what we already have.
- Our approval gate is not a platform setting that an agent or a configuration threshold could one day loosen for “more autonomy” — it is the structure of the product itself (draft quote, rule that opens an event), verifiable in our code today, not merely asserted in a brochure.
Against / the honest limit
- This is not a like-for-like comparison: Naïve aims to run an entire company from scratch (incorporation, bank cards, telephony) — a far broader problem than AppH's, which helps an existing SMB operate faster. Pitting them against each other directly would be dishonest.
- We have not tested Naïve's product ourselves — everything we know comes from a funding announcement, not an independent audit. Their human validation may be every bit as rigorous as ours; we simply cannot verify it from the outside, and that should be said plainly rather than implying otherwise.
The easy headline would be “even the sellers of total autonomy prove us right” — but that would be taking a press release for an independent audit, exactly the shortcut we refused to take yesterday with Langflow. What is true and verifiable is that human validation is no longer discussed as a brake on adoption: it now appears in the very pitch of products selling total autonomy. The real question that remains, for any vendor — us included — is not “is there human validation?” but “who has the power to decide what triggers it, and can you verify that yourself without trusting a brochure?” At AppH, the answer lies in code we can show, not in a policy we could change one day without saying so.
Reviewed by a human at AppH