A critical assessment of the $28.5M Series A building the layer that issues real LLCs, EINs, and payment cards to autonomous AI agents — led by Nexus Venture Partners. A public, code-backed investigation alleges the flagship runtime is an un-attributed fork of a 41,000-star open-source project.
A public investigation (not-so-naive.vercel.app) alleges Naïve’s autonomous-company runtime is Paperclip — an MIT-licensed project with 41,000+ GitHub stars — rebranded with zero attribution. Cited evidence: 71+ residual “Paperclip” strings in the production bundle. Naïve has not publicly responded. Allegation, not adjudicated.
“30,000+ developer customers” against a $0 / $49 / $149 price card with a free trial implies the overwhelming majority are free signups, not paying accounts. “Low double-digit millions” ARR is a company-only figure. The word “customers” almost certainly conflates signups with revenue.
Naïve issues real US LLCs, EINs, and virtual cards to autonomous agents, making itself the KYC/KYB and money-movement intermediary for potentially thousands of agent-run entities. Who is the “responsible person” on an agent-filed EIN? The AML/sanctions/tax exposure is unaddressed.
Key Finding: Naïve is chasing a genuinely large category — agent infrastructure — with credible investors and real momentum signals. But the moat narrative rests on a runtime it may not have built, the headline metrics conflate free signups with paying revenue, and the defining regulatory risk of provisioning legal and financial entities to non-human actors goes unmentioned in its own materials.
Naïve’s pitch is that an agent can go from prompt to operating company by wiring together everything a business legally and technically needs — behind one interface.
US LLC formation with EIN issuance — a real legal entity, provisioned via API.
Virtual cards with hard spend caps; KYC/KYB; Stripe/QuickBooks integration.
Storage, hosting (Supabase, Vercel), plus phone and email (Twilio) provisioning.
Fallback chains across “300+ models” from Anthropic, OpenAI, Google, Meta, Mistral.
“Vetta” — durable agent teams with a board that “survives a crash.”
Spend caps and orchestration — marketed as control, but not a regulatory answer.
Most of the stack is orchestration over third parties. Incorporation, payments, cloud, comms, and models are all sourced from Stripe, Supabase, Vercel, Twilio, and foundation-model APIs. The defensible piece is supposed to be the agent runtime — which is exactly the component under dispute.
An independent investigation alleges Naïve’s runtime is Paperclip — an MIT-licensed, 41,000-star project by developer “Dotta” — rebranded and shipped with no attribution, in violation of the MIT notice requirement. The evidence cited is concrete: 71+ residual “Paperclip” references in Naïve’s minified production bundle, including config paths (~/.paperclip/instances/), localStorage keys, and PAPERCLIP_* env-var help text, with reproducible verification steps. The timeline is uncomfortable: Paperclip went viral in early March 2026; Naïve’s first SSL certificate appeared roughly four days later. Naïve has not publicly addressed the claim. This is an unproven allegation — but it directly undercuts the “fund four research bets / hire researchers” moat narrative.
Two facts absent from the funding coverage change how the traction story should be read.
As recently as late 2025, this was Relixir — an AI-SEO tool (YC 2025, $2M seed, same two founders). The abrupt pivot to agent infrastructure coincided with Paperclip’s virality. There is no public explanation of the switch, and the “years of infrastructure work” framing sits awkwardly against a months-old brand.
“30,000 customers” and “low double-digit millions ARR” are presented together, but a $0/$49/$149 price card and a free trial mean most of those 30,000 are almost certainly free accounts. No paying-customer count, net retention, or gross margin is disclosed. Every growth figure traces to Naïve’s own press cycle.
The investor thesis is a category bet, not a metrics endorsement. Nexus’s Abhishek Sharma frames it as “autonomous companies” being the next decade’s trillion-dollar wave — conviction about where the world is going, not a verification of Naïve’s revenue, moat, or code provenance.
Six structural risks the $28.5M Series A does not resolve.
A public, code-backed allegation that the flagship runtime is an un-attributed MIT fork of Paperclip. Unresolved; precedent (PearAI, Pickle/Glass) shows YC startups usually apologize fast — Naïve has stayed silent.
Incorporating and funding real companies for autonomous agents raises unaddressed AML/KYB, money-transmission, sanctions, and fraud-liability questions that could invite a regulator or banking-partner shutdown.
The product is largely orchestration over Supabase, Vercel, Stripe, Twilio, third-party models, and (allegedly) Paperclip — little defensible proprietary technology behind a 10-person team.
“30,000 customers” and “low double-digit millions ARR” are company-only figures that likely conflate free signups with paying revenue; no retention or margin data disclosed.
The team pivoted from AI-SEO to agent infrastructure within months and has limited track record operating provisioning, payments, and compliance infra at scale.
The open-source developer audience is the exact funnel Naïve sells into. Attribution backlash from a 41K-star project’s supporters directly threatens acquisition.
Naïve is selling the picks and shovels for a gold rush that may not have arrived yet. The category is real and the backers are serious. But before treating this as a durable infrastructure bet, the diligence questions are unavoidable: did they build the runtime they’re raising on, how many of the 30,000 actually pay, and who is legally accountable when an autonomous agent incorporates and spends? Naïve has answered none of them publicly.
Based entirely on publicly available information, including the TechCrunch announcement of August 6, 2026. Company-reported figures are labeled as such; the attribution allegation is presented as an unproven, third-party claim Naïve has not answered.
