Naïve Raises $28.5M, With Human KYC Still Required

TechCrunch’s August 6, 2026 report made public Naïve’s $28.5 million Series A, led by Nexus Venture Partners. The financing brought the San Francisco startup’s total funding to roughly $32 million and backs infrastructure intended to let software agents set up and operate businesses through a common interface.

The round itself has independent confirmation: Axios’s August 7 deal record lists the same $28.5 million Series A, with Nexus leading and Y Combinator, Zetta Venture Partners and Liquid 2 Ventures participating. The product automates service provisioning, but it does not make an AI agent the legally unaccountable owner of a company: verified people and controlled payment relationships remain part of the system.

The round backs an operating stack, not just incorporation

Naïve is building a shared infrastructure layer through which agents can obtain services that founders would otherwise arrange with separate providers. The Y Combinator profile for Naïve identifies the company as founded in 2025 and describes one configuration spanning compute, payments, identity, incorporation and outside tools, with governance applied to every action.

For beginners, that stack is easier to understand as seven connected layers:

Formation: business details, industry classification, filing requests, status tracking and formation documents.
Identity: verification records for the people connected to a company or managed card.
Payments: virtual cards, funding, spending limits and transaction records.
Communications: domains, email inboxes, phone numbers and SMS access.
Compute: hosted applications, databases, cloud instances, sandboxes and mobile devices.
Memory and routing: systems intended to preserve business context and select models for different tasks.
Governance: permissions, budgets, approvals, audit trails and usage metering.

Formation is therefore an entry point rather than the entire product. Agents can provision operational resources and act within configured permissions, while Naïve supplies the infrastructure around them rather than the underlying reasoning models. Customers still bear the cost of model calls and other metered services used by their agents.

Available automation still stops at external gates

Naïve’s public catalog of more than 40 primitives places company formation, virtual cards, email, phone, research, cloud infrastructure and mobile devices behind one bearer token, credit balance and SDK. The same page marks the agent-orchestration runtime as coming soon and puts freight, commercial leases and wholesale accounts in a forthcoming physical-operations category.

The catalog also shows why a common API is not the same as complete autonomy. Formation includes hosted identity checks for founders; managed cards require cardholder identity setup; and outbound SMS becomes available only after carrier approval of a registered campaign. Filing authorities, identity providers, payment partners and carriers can therefore delay or reject actions outside an agent’s control.

This is the central boundary around the “autonomous company” label. Naïve can automate the assembly and operation of services, but it cannot remove the legal identities and third-party approvals that make consequential actions valid.

Human KYC and payment remain mandatory for formation

The clearest boundary appears in Naïve’s US LLC workflow. An agent can retrieve NAICS classifications, provide proposed names and business details, create a formation request, monitor its status and collect completed documents. It cannot move the filing forward until the founders satisfy the identity requirement.

Naïve’s formation documentation requires all founders to pass KYC and the verification record to be marked ready for formation. Creating the request generates a hosted checkout for $349; a user completes payment, and only after the payment status changes to paid can the filing be submitted. The listed fee covers the managed formation service, state filing fees, registered-agent service, an EIN application and required legal documents.

The workflow also prevents a simple automation loop from repeatedly creating the same charge. A duplicate request attached to an existing non-failed formation returns the existing record. If a filing fails after payment, the fee remains associated with that record and the user may need support or a retry of the original submission.

Programmable payments do not eliminate ownership

Virtual cards give an agent a bounded way to buy domains, computing capacity or subscriptions without exposing a general bank account. Funding, spending limits and transaction tracking make the expenditure programmable, but a person or organization still decides which agent receives a card and how much it may spend.

That distinction turns governance into part of the product rather than an administrative afterthought. Identity remains connected to managed cards, while balances, permissions and approval rules constrain what agents can do. External providers may also suspend or revoke access regardless of the instructions given to an agent.

The same control structure applies to communications and compute. An agent may send email, deploy an application or run a research task, but only through services and credentials made available to it. Automation reduces manual setup; it does not erase accountability for the resulting actions.

Inference and governance costs could exceed formation

Incorporation is a bounded transaction, whereas operating an agent-run company is continuous. Model inference, stored context, research calls, compute, communications and failed-task retries can continue accumulating for as long as agents remain active.

The financing is expected to support work on model routing, memory, orchestration and a serverless agent runtime. Those systems are intended to reuse business context, send tasks to suitable models and avoid dedicating a full virtual machine to every agent. They remain product plans and company claims, not independently demonstrated cost reductions at scale.

This makes cost control inseparable from autonomy. The formation charge is visible before checkout, but a poorly constrained agent loop can repeatedly consume models and infrastructure. Spending caps, approval thresholds, revocation controls and audit records may consequently matter more to sustained operations than the initial filing.

What the funding establishes now

The Series A establishes investor backing for a broad infrastructure thesis: agents can automate much of the technical work involved in assembling and operating a company when formation, identity, payments, communications and compute share a control layer. It does not establish that a business can operate without accountable people or that Naïve’s planned routing and orchestration systems have already contained costs at scale.

As of August 9, the documented boundary is clear. Founders must pass KYC, formation waits for payment, managed cards retain identity requirements, outbound messaging depends on carrier approval, and parts of the orchestration and physical-operations stack remain forthcoming. What remains unknown is how reliably customers can govern many agents over time—and whether the systems financed by the round can keep recurring inference and infrastructure costs under control.

Naïve Raises $28.5M—but Autonomous Companies Still Need Human KYC