Naïve Raises $28.5M Series A for AI Agent Infrastructure

Naïve, a San Francisco-based AI agent infrastructure company, raised a $28.5M Series A led by Nexus Venture Partners, with participation from Y Combinator, Zetta Venture Partners, Liquid 2 Ventures, and DEEPCORE. Combined with Naïve's previously disclosed $2M seed round, the financing brings total funding to $30.5M. The company did not disclose a valuation.

Founded in 2025 by Sean Dorje, CEO, and Dennis Zax, CTO, Naïve is building infrastructure designed to let AI agents do considerably more than generate text or complete isolated tasks. Its stack brings together compute, identity, payments, memory, model routing, governance, orchestration, and access to 10,000+ tools.

The timing matters because the AI-agent market is beginning to encounter a less glamorous problem than intelligence: operations. An agent can be brilliant inside a demo and still be useless as a business operator if it cannot securely spend money, maintain identity, remember context, access software, execute tasks, and operate within defined permissions. Naïve is betting that this operational layer becomes valuable infrastructure as agents move from assistants toward economic actors. That is a much harder proposition than giving a chatbot a nicer interface, which is precisely why the $28.5M deserves attention.

What Happened

Naïve announced the $28.5M Series A on August 5, 2026. Nexus Venture Partners and Abhishek Sharma led the financing, joined by Y Combinator, Zetta Venture Partners, Liquid 2 Ventures, DEEPCORE, and other operators backing the company. Naïve's previously disclosed $2M seed brings total funding to $30.5M.

Sean Dorje, CEO, and Dennis Zax, CTO, dropped out of UC Berkeley to build Naïve. The company is tackling a deceptively simple question: what does an AI agent need when its assignment moves from writing an email to actually operating the business? Quite a bit, as it turns out.

Naïve combines KYC/KYB capabilities, LLC formation, email, phone, virtual cards, cloud infrastructure, model routing, memory, orchestration, governance, and human approval for sensitive actions. The platform supports 300+ AI models and provides access to 10,000+ tools, with integrations including Stripe, GitHub, and QuickBooks. Naïve isn't simply trying to produce another intelligent agent. It is building the machinery an agent needs to function inside the messy, permission-heavy, financially consequential environment called a company. Software demos tend to skip that part. Businesses don't get to.

Why Naïve's $28.5M Series A Matters

The AI market has spent enormous amounts of capital making models smarter. Intelligence, however, is only 1 component of useful autonomy. Businesses operate through permissions, accounts, payments, databases, memory, compliance controls, software systems, schedules, and humans who occasionally need the ability to say absolutely not. Naïve is packaging those requirements into infrastructure for agents.

The company says 30,000+ developer customers have signed up since its public launch. Sean Dorje says annualized run-rate revenue increased roughly 10x over 6 months into the low double-digit millions. Reported use cases include AI automation agencies, faceless TikTok and YouTube channels, and an autonomously operated rental-car business. The rental-car example is particularly revealing because it pushes the AI-agent conversation away from clever digital assistants and toward software participating in actual economic activity.

That transition changes what matters. Reliability matters. Cost matters. Identity matters. Permissions matter. Memory matters. Governance really matters once software gets access to the company credit card. AI autonomy gets considerably less philosophical when somebody has to reconcile the statement.

The Infrastructure Economics Behind Naïve

Naïve's business thesis becomes more interesting when viewed through AI economics rather than AI novelty. The company reports agent startup times of 2.3 milliseconds using 1.2MB per instance, along with memory recall requiring 11x fewer tokens per query. Naïve is also working on serverless agents and inference optimization.

Those numbers point toward a basic economic problem facing agent developers: autonomous systems can become expensive remarkably quickly. An agent that repeatedly invokes models, retrieves context, maintains state, interacts with external tools, and waits for tasks can consume resources even when the underlying business activity isn't generating equivalent value. Naïve is attacking that cost structure at the infrastructure layer.

If AI agents proliferate, developers may not want to separately assemble identity, payments, compute, model access, memory, orchestration, integrations, and governance every time they build one. The opportunity isn't merely helping an agent think. It is reducing the operational tax required to let the agent work.

From AI Assistants to Autonomous Companies

Naïve describes its broader ambition around infrastructure for autonomous companies. Strip away the science-fiction aroma and the business proposition becomes fairly concrete. A useful autonomous agent needs authority to act, infrastructure through which to act, memory of what already happened, tools for completing the action, and controls defining what it absolutely cannot do. Naïve is trying to put those pieces behind unified infrastructure.

That approach reflects a broader maturation of the AI-agent category. The first wave of generative AI centered heavily on output: text, images, code, answers. Agent systems push toward execution, and execution is where software meets consequences. A generated paragraph can be wrong and annoy somebody. An autonomous system touching payments, customer records, infrastructure, or company operations can be wrong and create a meeting involving lawyers.

That makes governance part of the product rather than paperwork added later. Naïve's capability policies, audit logs, and human approval mechanisms for sensitive actions address a basic requirement for moving agents into higher-stakes environments. Autonomy without control isn't enterprise software. It's an incident report waiting for a timestamp.

What Nexus Venture Partners Is Backing

Nexus Venture Partners is effectively backing an infrastructure thesis about where value could accumulate as AI agents become more capable. The Series A also includes Y Combinator, Zetta Venture Partners, Liquid 2 Ventures, and DEEPCORE. Naïve's earlier $2M seed round brings total funding to $30.5M.

For investors, infrastructure offers different exposure to the agent market than betting on individual applications. Agent applications will compete over specific workflows and customer categories. Infrastructure can potentially sit beneath multiple applications, business models, and agent architectures. If developers increasingly need common services for identity, payments, memory, compute, inference, integrations, and governance, the underlying layer becomes strategically important.

Naïve still has to prove that developers want these capabilities consolidated rather than assembled from specialized providers. That's the commercial fight hiding underneath the technical one. The company's 30,000+ developer customers and reported 10x annualized run-rate revenue growth give that question considerably more weight than a presentation deck.

What Naïve Plans to Build Next

Naïve plans to use the Series A to fund research across serverless agents, inference optimization, persistent memory, and agent-workforce orchestration, alongside engineering hiring. Each priority addresses a practical constraint on agent deployment. Serverless execution targets the economics of keeping agents available without dedicating expensive resources to idle workloads. Inference optimization attacks model costs. Persistent memory gives agents usable business context across interactions. Orchestration addresses the increasingly complicated problem of coordinating multiple agents and tasks.

Taken together, the roadmap suggests Naïve sees the next phase of agent infrastructure as an efficiency and coordination problem as much as an intelligence problem. Model capability will continue improving, but better reasoning doesn't automatically create better unit economics, stronger controls, or reliable execution. Those problems require infrastructure.

The companies that solve them won't necessarily produce the flashiest demos. They may produce something considerably more valuable: systems businesses can actually afford and trust to run.

What This Signals for the AI-Agent Market

Naïve's $28.5M Series A illustrates a shift in where sophisticated capital is looking within AI. The question is moving beyond whether agents can perform useful work. Increasingly, the question is what infrastructure makes that work repeatable, economical, governable, and commercially viable. Sean Dorje and Dennis Zax are positioning Naïve directly inside that transition.

If autonomous agents become meaningful economic actors, an infrastructure market should form around them just as previous computing shifts created valuable layers around cloud infrastructure, payments, identity, security, and developer tooling. That doesn't mean autonomous companies arrive tomorrow, or that humans should begin clearing their desks for an army of JavaScript executives. It means the underlying technical requirements are becoming concrete enough for developers to build against, customers to test, and investors to finance.

Naïve now has $28.5M more to determine how much of that infrastructure it can own. The name makes the closing argument almost unfair. Building operating infrastructure for autonomous companies sounds naïve until 30,000+ developers start testing the premise. Then the more interesting question becomes who owns the rails if they turn out to be right.