DigitalOcean Launches Managed Agents in Public Preview, Targeting the Agent Infrastructure Gap

DigitalOcean has opened a public preview of Managed Agents, a cloud service that supplies the runtime, tooling, and inference plumbing companies need to put AI agents into production. The offering packages isolated microVM execution, controlled access to external tools, and serverless model inference into one managed layer. News of the launch was reported by InfoQ, with Sergio De Simone covering the release.

The pitch is straightforward: most teams building agents today spend their early engineering cycles stitching together sandboxes, permission checks, and model endpoints rather than improving the agent itself. DigitalOcean's answer is to rent that plumbing. Each agent runs inside its own microVM, which keeps one workload's code and data away from another's, a separation that matters once agents start executing generated code or calling internal systems. Tool access is governed through the platform rather than left to each developer to enforce, and inference is billed on a serverless basis so capacity tracks usage instead of sitting idle.

DigitalOcean is not the first vendor to spot this demand. AWS, Google Cloud, and Microsoft Azure have all pushed agent-building frameworks and runtime services over the past two years, and a crop of startups sells sandboxing and orchestration tools aimed at the same buyers. What DigitalOcean brings is its existing base of smaller engineering shops and independent developers who already host applications on its droplets and Kubernetes service. For those customers, adding agent workloads without opening a second cloud account removes a real source of friction and billing overhead.

The choice of a public preview rather than general availability signals the service is still being hardened. Public previews typically mean APIs can shift, limits are capped, and support is lighter than what paying customers eventually receive. That matters for anyone weighing Managed Agents against a self-built stack, since migration costs cut both ways once an agent's state and tool integrations are wired to a specific runtime.

The larger question is whether managed infrastructure becomes the default way agents get deployed, the way managed databases and managed Kubernetes did. Agents differ from those workloads in one important respect: they take actions in the world, which puts audit trails, credential scoping, and rollback squarely on the platform. How DigitalOcean handles those controls, and at what price, will decide whether Managed Agents stays a convenience for small teams or becomes a serious option for larger ones. The preview period should produce the first hard answers.

Developers can sign up for the preview now, and the coming months will show how quickly DigitalOcean ships the enterprise-facing pieces, particularly finer-grained permissions and logging. Watch for pricing details and a general availability date, both of which were absent from the initial announcement.