Naive launches a cloud stack for apps, agents and autonomous companies

Why it matters

Naive is betting that coding agents will choose infrastructure through machine-readable policies, shifting cloud distribution from dashboards toward agent-controlled APIs.

Naive launched an AI-native cloud stack on August 6th that aims to give coding agents one control plane for building applications, operating autonomous agents and forming the infrastructure around AI-run companies.

The release is the broadest version yet of the product co-founder Sean Dorje (@seandorje) has been assembling around a simple thesis: agents need access to the operational systems that human founders and engineering teams normally configure across dozens of services. Naive is packaging those systems behind one API key, command-line interface and declarative configuration file.

Naive's pitch now reaches well beyond tool calling. The platform covers application hosting, serverless functions, GPU workloads, Postgres databases, authentication, object storage, model routing, virtual cards, invoicing, email, phone numbers, company formation and third-party software connections. Naive says developers can use those components from coding agents including Claude Code, Codex and Cursor, or from agent runtimes such as Hermes and OpenClaw.

The breadth puts Naive in competition with several categories at once. A developer can already assemble similar capabilities from cloud providers, model gateways, backend-as-a-service products, payments platforms and agent frameworks. Dorje is betting that agents will favor a single machine-readable control plane over the dashboard-heavy process built for human operators.

Infrastructure as code for agents

The center of the release is an infrastructure-as-code layer that lets an agent declare identity, spending limits, budgets, approval rules and permitted tools in one configuration file. Naive then provisions and governs the underlying resources.

That design matters once an agent can spend money or act through a customer's accounts. Naive says its governance layer checks policy before a payment moves or a tool runs. The platform can require approval for sensitive actions, impose spending caps, issue short-lived sessions and record activity in audit logs.

Those controls are also Naive's answer to the risk created by consolidating so much authority behind one credential. An agent with access to compute, payments, customer software and incorporation services carries a larger blast radius than an assistant limited to generating text. Naive's product will depend on whether its server-side controls remain reliable as developers connect more consequential workflows.

Naive says its model router supports more than 300 models, with task-based routing, fallback providers and usage metering. Naive also advertises more than 100 third-party connections and a directory of over 10,000 tools. Those figures are Naive's own counts, and they combine different kinds of resources: model endpoints, integrations and callable primitives do not represent the same level of product depth.

Developers can start through the Naive quickstart, which exposes the platform through a CLI, hosted MCP server, REST API and software development kit. The onboarding flow allows an agent to register, receive an API key and begin calling primitives without a developer manually configuring each service.

Dorje is moving Naive down the stack

Dorje previously positioned Naive around autonomous companies: AI workforces with a CEO agent, task board, specialized workers and access to business tools. The new cloud framing moves Naive deeper into the infrastructure underneath those companies and opens the product to developers building their own agent software.

That shift was visible by June 17th, when Dorje introduced the Naive Developer Platform. That release described 38 primitives across identity, money, communications, generation, data, orchestration, application building and agent infrastructure. It also introduced a multi-tenant structure in which developers can isolate resources and policies for each customer.

Naive later packaged those capabilities into an Agent SDK for products that need to assign separate cards, credentials, sessions and tool access to each user. The August 6th launch ties those pieces to a larger cloud proposition: an agent should be able to build the application, deploy it, connect its tools and operate the resulting business through the same system.

Naive's immediate challenge is execution across an unusually large surface area. Hosting, databases, payments, identity verification and model routing each carry distinct reliability, security and compliance demands. Naive is choosing to own the coordination layer rather than asking developers to stitch those systems together themselves.

That gives Dorje a clear distribution bet. If coding agents become a primary interface for creating software, infrastructure vendors will need products that agents can understand and provision directly. Naive is building for that interface now, before the dominant cloud platforms finish adapting their human-centered consoles and product catalogs to autonomous users.