Naïve has raised $28.5 million in Series A funding to develop infrastructure that enables AI agents to operate businesses through a unified application programming interface. The round was led by Nexus Venture Partners, with participation from Y Combinator, Zetta Venture Partners and Liquid 2 Ventures.
Angel investors included Gokul Rajaram, Apollo co-founder Tim Zheng, former HubSpot executive JD Sherman, Amazon executive Gert Lanckriet, DocuSign executive Robert Chatwani and Codecademy co-founder Zachary Sims.
Palo Alto-based Naïve was founded by Sean Dorje and Dennis Zax, two 20-year-old University of California, Berkeley dropouts who have worked together since they were 14.
The founders previously built and sold machine-learning company ezML while they were teenagers before participating in Y Combinator.
Naïve is developing an operating stack designed to give AI agents the technical, financial and organizational capabilities required to run real-world businesses.
The company argues that coding agents can now build functional software quickly, but transforming that software into an operating company still requires developers to connect numerous services covering incorporation, payments, email, communications, cloud infrastructure and accounting.
Many of these systems were designed for human users rather than autonomous software agents, creating integration and governance challenges.
Naïve seeks to consolidate those functions behind a single configuration file and unified API.
A developer can have a coding agent create the configuration file, after which Naïve provisions the required operating infrastructure. This can include business incorporation, virtual payment cards, email inboxes, mobile phone numbers, computing resources, AI models and memory.
The platform also includes a governance gateway that evaluates actions before execution. Developers can establish budgets, approval requirements and capability restrictions governing what an agent can spend or do.
Naïve’s real-world identity capabilities include know-your-customer and know-your-business verification, limited liability company formation, email accounts, phone numbers and virtual cards. These tools are designed to allow agents to operate through identifiable legal and economic entities.
Its cloud services include relational databases, hosting, computing, object storage, authentication and other infrastructure required to run applications.
Naïve also offers serverless environments for agents including Hermes, Claude Code, Codex and Vetta. The company represents agents as lightweight, serializable states rather than continuously rented virtual machines.
The platform’s multi-agent orchestration capabilities support task management, scheduling and communication among specialized agents.
Naïve also routes inference requests among different hosted AI models based on the complexity of each task. The company aims to direct calls to the lowest-cost model capable of producing an adequate answer while reducing the amount of context supplied to each model.
Its shared memory layer is designed to accumulate information across the company and provide each agent with only the context required for a particular task.
The platform can connect with more than 10,000 third-party tools used by businesses, including Stripe, GitHub, Supabase and QuickBooks.
Governance functions include capability policies, immutable audit logs and human approval requirements for sensitive actions.
Naïve plans to direct the Series A capital toward four research areas: serverless runtimes, inference optimization, shared memory and multi-agent orchestration.
The company said conventional agent infrastructure can be inefficient because agents frequently remain idle while waiting for responses or external actions.
Naïve’s serverless runtime uses V8 isolates and charges for the periods in which an agent is actively executing rather than maintaining continuously available virtual machines.
According to the company, the architecture can achieve a 2.3-millisecond cold start while requiring approximately 1.2 megabytes per agent.
Its inference research will focus on reducing model expenses by selecting smaller language models for routine steps and using batched inference for longer-running processes.
Naïve’s memory system is intended to distill conversations and activity into structured facts. The company claims the system can outperform existing approaches on recall while using approximately 11 times fewer tokens per query.
Its orchestration technology delegates tasks to different subagents and model tiers rather than using the most expensive frontier model for every interaction. Naïve says its serverless approach can operate at approximately one-hundredth of the cost of certain conventional configurations.
The company’s broader objective is to maximize the amount of useful work completed for every token consumed by autonomous agent systems.
Naïve believes that as businesses delegate more work to AI agents, spending on inference and agent execution could become a significant operating expense.
The platform is intended to make autonomous companies more economically practical while ensuring that human developers and operators retain control over budgets, permissions and consequential decisions.
KEY QUOTES:
“The last two years proved autonomous software. The next decade is about autonomous companies. Naïve gives millions of entrepreneurs and small businesses worldwide the turnkey infrastructure to build and run autonomous companies, without needing to become AI experts.”
Abhishek Sharma, Partner at Nexus Venture Partners
“In this next decade, agent spend is set to scale into the trillions as companies completely run on agents that finish work autonomously.”
“Our vision at Naïve Labs is to make each token do more, so autonomous companies can become a cost-efficient reality.”
Sean Dorje, Co-Founder and CEO of Naïve
