Somewhere on Naïve's infrastructure, AI-generated cats and dogs are dancing and boxing on TikTok. Sean Dorje found that customer himself and told TechCrunch about it with the Series A news, which is either the most honest proof point in a funding round this year or a warning label, depending on your mood. Palo Alto's Naïve closed $28.5 million on August 6, Nexus Venture Partners leading, Y Combinator, Zetta, and Liquid 2 in the round, angels including Gokul Rajaram, Apollo.io's Tim Zheng, and former HubSpot COO JD Sherman. About $32 million total raised. Pitch: agents can ship an app by lunch. Turning that app into a company still means a human clicking through incorporation, banking, email, phone, UI built for browsers, not tool calls. Naïve wraps LLC filing, virtual cards, inboxes, numbers, memory, and compute behind one config and API, plus a governance gateway for budgets and approvals. Dorje's numbers to TechCrunch: ARR up 10× in six months into the low double-digit millions, 10 employees, 30,000+ developers since launch.
Twenty-year-old founders and an API that still needs a human
Dorje and Dennis Zax are 20, Berkeley dropouts, shipping together since 14, teenage sale of a computer-vision shop called ezML, then YC with Naïve. That biography opens doors. It also tempts people to write the product story as destiny. Ten people. Low-double-digit-millions ARR claim. Both can be true without the sequel writing itself. Nexus partner Abhishek Sharma told SiliconANGLE the bet is infrastructural: autonomous software exists. The decade is about autonomous companies. Naïve is the turnkey stack so small businesses don't have to become AI experts. Whether agent-run shops keep paying for someone else's incorporation-and-cards layer once they have real revenue is the part quotes can't settle.
TechCrunch's walkthrough: paste a Naïve prompt into Cursor, Claude Code, or Codex. The agent hits the API and can drive a U.S. LLC filing, state, NAICS-ish industry code, description, name candidates, then email, cards, phones, databases, compute, Stripe/QuickBooks hooks. Templates for AI SEO shops, SaaS, recruiting, accounting, support, even a mobile emulator. KYC/KYB don't vanish. Users still do identity checks and required payments themselves. Agent fills forms. Human shows face to the bank. The governance gateway (budgets, caps, human OK before sensitive spends) is supposed to stop silent burn. No third-party security audit of that gateway showed up in the Series A materials TechCrunch and SiliconANGLE summarized. Normal at this stage. Unfinished engineering, not a settled safety claim.
I keep noticing how often "autonomous company" slides skip the part where a state filing office or a bank compliance queue still runs on human time. Naïve's honesty about KYC is refreshing precisely because the category marketing usually isn't.
Cats, rental cars, and the gap between signups and ARR
Dorje's customer list to TechCrunch is messier than deck-speak usually allows, which is why it stuck with me. AI automation agencies (fastest-growing lane, he said, first business a lot of people start is selling agents to other small businesses), faceless TikTok/YouTube channels, an autonomously run rental-car agency, and those boxing cats. Content farm with a corporate card is not a gross-margin case study. It is a thin wall between categories.
30,000 developer signups ≠ low-double-digit ARR. TechCrunch doesn't pretend they are. How many pay recurring versus weekend free-tier? Unknown. Valuation? Undisclosed. Once the LLC exists, the expensive bit is often the agents, model calls, fat contexts, idle sandboxes. Dorje said inference and serverless agents are among the fastest demand lines, with unnamed enterprises sniffing around. Series A money goes to researchers and four bets: virtualized agent sandboxes, model routing / inference optimization, memory, governance and orchestration. Lightweight JS environments instead of a full VM per agent so you pay when something's awake. A router that picks cheaper models and caches prior reasoning. Memory so the "company" remembers which invoice it already disputed. None of that agenda is unique in 2026. Shipping it glued to the paperwork surface people already use is the hard part for a 10-person lab.
The rental-car anecdote is the one I want more detail on and don't have. How much human escalation still happens when a customer disputes damage? Coverage is thin. Until someone publishes a longer operational write-up, I'm treating it as a color story, not a playbook.
Our vision at Naive Labs is to make each token do more, so autonomous companies can become a cost-efficient reality.
Assembly kits versus a governed surface
Stripe Atlas + a business bank with cards + Twilio + an email API already exist. Naïve's reported answer: those assume a human finishes the last mile, dashboards, verification, noticing when something looks wrong, and an agent wants one governed surface. Product thesis, not a moat. If autonomous companies stick, AWS / Google Cloud / Microsoft will notice without a press release. Neat competitor grids for a category this young are mostly fiction until overlapping traction shows up the way Dorje's numbers did. Adjacent startups are chasing sandboxes, orchestration, pre-built specialist crews. The build-versus-buy split hasn't settled.
There's a version of this story where Naïve becomes the default "company object" agents talk to, the way Stripe became the default payments object. There's another version where a few operators hit real revenue, hire humans, and rebuild the stack in-house because they don't want their corporate nervous system rented. I don't know which one wins. Coverage on retention and churn is basically nonexistent beyond Dorje's ARR growth claim.
I keep coming back to Dorje's line about making each token do more. Paperwork is once. Token burn is monthly. Skew toward developers who just wanted an LLC without a lawyer, and Naïve is a workflow product. Skew toward production agent fleets on the runtime and router, and $28.5M starts to look sized for the problem. The Series A funds researchers against those four infrastructure bets. Shipping them tightly enough that customers notice is the next chapter, and a 10-person team does not get infinite runway to miss.
I'd watch whether customer stories graduate from content channels to businesses with payroll-like recurring costs, and whether the governance gateway survives its first public incident without becoming a punchline, not the next viral TikTok of animals throwing hands. For now TechCrunch has a small team, real ARR growth claims, and a strangest-proof-point video of AI cats boxing. I'm weirdly glad they led with the cats. At least nobody pretended this was only enterprise workflow software.
What $28.5M has to buy
Series A money disappears into hiring and cloud bills faster than pitch decks admit. Dorje's public shopping list, sandboxes, routing, memory, governance, is four hard systems problems. A 10-person company can ship thin slices of each. It cannot out-research every lab chasing the same ideas unless the incorporation-and-cards wedge keeps paying the lights. That's the bet Nexus is making. Whether the cats-and-TikTok cohort converts into something sturdier is the bet I'm less sure about, and Dorje, to his credit, didn't pretend otherwise in the anecdotes he chose to share.
- Funding
