Naïve raised $28.5 million to help AI agents form companies, provision software and manage business operations. The startup will use the Series A to cut agent costs through model routing, shared memory, lightweight runtimes and governance tools.
Naïve has raised $28.5 million in a Series A round to build infrastructure that lets AI agents handle company formation, business software setup and recurring operations.
Nexus Venture Partners led the round. Y Combinator, Zetta, Liquid 2 and angel investors including Gokul Rajaram, Apollo.io co-founder Tim Zheng and former HubSpot COO JD Sherman joined the financing. The round brings Naïve’s total capital raised to about $32 million.
One API for company operations
Naïve packages payments, email, phone numbers, cloud infrastructure, storage and incorporation behind one API. Developers can connect the API to tools such as Cursor, Claude Code and OpenAI Codex, then give an agent instructions to provision the resources that a business needs.
An agent can prepare a U.S. LLC application with a state, industry code, business description and proposed names. A person must complete KYC and KYB checks and make required payments, but an agent can handle much of the surrounding work.
The platform can provision email inboxes, virtual cards, phone numbers, databases and computing resources. It can connect those systems to services such as Stripe and QuickBooks, giving an agent access to the tools that support billing, accounting and customer operations.
Naïve gives customers controls for budgets and agent permissions. Customers can require human approval before an agent takes a sensitive action, which adds a review point to activities such as spending money, changing account access or communicating with customers.
The company offers templates for AI search-engine-optimization services, full-stack software-as-a-service applications, recruiting, accounting and customer support. One template includes a mobile emulator that lets agents operate smartphone applications on virtual devices.
Developers are testing agent-run businesses
Naïve says more than 30,000 developers signed up within months of its launch. CEO and co-founder Sean Dorje said the company increased annual run-rate revenue by 10 times during the past six months, reaching the low double-digit millions.
Customers use the platform to run AI automation agencies, faceless video channels on TikTok and YouTube, and a rental-car agency. Dorje said one customer used Naïve’s infrastructure to support a TikTok channel that published AI-generated videos of dancing and boxing cats and dogs.
AI automation agencies represent the fastest-growing customer group, Dorje said. These businesses sell agent systems to small companies, creating a market in which one company uses AI agents to build and operate services for another.
That pattern gives Naïve a distribution advantage. A founder may start with one automated service, then add agents for sales, support, bookkeeping and fulfillment. Each new function can create demand for another account, data store, payment connection or computing environment.
Agent costs create a second market
The cost of running agents can rise with every model call, context transfer and idle process. A system that appears inexpensive at small scale can consume a large budget when dozens of agents repeat tasks across the day.
Naïve plans to use the new capital to address that expense. The company is building a model router that sends each request to a suitable model, a memory layer that stores business context, and an orchestrator that divides work among agents.
The company also plans a serverless runtime for agents. The runtime will place agents inside lightweight JavaScript environments instead of assigning each agent a full virtual machine. Customers can pay for active execution instead of keeping a separate machine available for each agent.
That architecture could lower the cost of deploying large agent fleets. It also gives Naïve a path into existing companies that want to run agents inside established workflows, where inference costs can outweigh the cost of initial setup.
Dorje said demand for inference optimization and serverless agents has grown faster than demand for company-formation tools. Enterprise customers have shown interest, although Naïve has not identified them.
From setup tool to operating layer
Naïve’s first product addresses the administrative work that slows a new company. Its next products target the recurring costs that shape an agent business after launch.
The shift matters for the company’s revenue model. A founder may use incorporation, phone and card tools once. An agent platform can generate usage across model calls, storage, computing and business integrations for as long as the customer operates.
Naïve has 10 full-time employees. Dorje plans to use the funding to hire researchers and develop four infrastructure areas: virtualized sandboxes, model routing and inference optimization, a memory layer, and governance and orchestration.
The company must make those systems reliable enough for financial tasks, customer communication and access to sensitive business data. Human approval controls can reduce risk, but customers will judge the platform by the accuracy of its agents and the cost of correcting their mistakes.
Naïve’s traction suggests that developers want a faster path from an idea to a functioning business. The larger opportunity depends on whether the company can help customers operate hundreds or thousands of agent tasks without letting model and infrastructure bills erase the savings from automation.
Comments
Please log in or register to join the discussion