Run AI Agents at Scale | Introducing Okteto Agent Fleets
Building the Future of Development Environments for AI Agents
AI agents are moving fast. What started as single-agent experiments in terminal windows is quickly turning into something more ambitious like multi-agent systems that collaborate, take initiative, and run complex workflows in parallel. The future isn't one agent. It's fleets of them.
But here’s the challenge: we need real infrastructure to support this, especially at enterprise scale.
Many teams are beginning to scale agent usage to broader adoption across engineering workflows. But the tools and infrastructure are lagging behind that ambition. Isolation becomes harder to maintain, policies are difficult to enforce, and managing agent deployment at scale introduces new overhead for platform teams.
What teams need is a way to give every agent its own remote environment that runs on their company infrastructure. These environments must be ephemeral, secure, and aligned with how your applications run in production. That unlocks real use cases for parallelism and scale. It enables teams to verify code rather than have agents simply suggest it. Most importantly, it supports agentic development in a way that meets enterprise standards.
Why Agentic Development Requires Infrastructure
Scaling AI agents from a single instance to a functioning fleet is a shift that requires more than just the right logic or workflows. It requires an infrastructure orchestration layer.
We’re seeing a wave of multi-agent frameworks and building solid proof-of-concepts for local dev. You’ve probably even seen the workflow where developers are running five terminals with five different agents. It works, to a point. But when those agents need to run concurrently, access internal systems, or follow organizational policies, that approach begins to fall apart. We know that local setups and shared clusters are simply not built to support repeatability, security, and scale.
You start hitting familiar friction: configuration drift, lack of auditability, resource contention, and security concerns. These aren't signs of bad practice. They're symptoms of missing infrastructure.
So, the moment has arrived. We’re pushing the boundaries at a faster pace than we’ve seen in a very long time and we need to quickly solve for these limitations in order to unlock real agent adoption.
Jack Clark, cofounder of Anthropic, described the opportunity clearly:
"I think it's actually going to be the era of the manager nerds now, where being able to manage fleets of AI agents and orchestrate them is going to make people incredibly powerful."
But to unlock that potential, you need ephemeral, isolated and reproducible environments for every agent. You need observability into what’s running and where. And you need automation that lowers the barrier for developers while keeping platform teams in control.
That’s the foundation we built Okteto Agent Fleets on.
Introducing Okteto AI Agent Fleets in Beta
We’re excited to launch the early beta for Okteto AI Agent Fleets, built to help platform and engineering teams orchestrate, run and test agents at scale.
With Agent Fleets, every AI agent gets its own fully provisioned environment, ready with the services, configs, and dependencies it needs. These environments are designed to be ephemeral: they spin up in seconds, run the task, and shut down when done.
Beyond just generating suggestions, agents can now build, deploy, and test code end-to-end inside these environments. That gives developers confidence in the output, and they no longer need to manually verify every change or review every line. When the agent completes its workflow, you know the code works because it’s already been validated in a production-like setting.
Why Okteto Environments Fit Agent Workflows
As agents become an extension of our teams, they’ll need the same access and resources that developers need today such as code, infra, tests, network policies, etc. And they’ll need isolated spaces to experiment, fail, try again, and improve without the risk. Okteto environments provide exactly that with strong security baked in:
- Ephemeral by design: every agent gets a clean slate. No stale state or config bleed across runs. It also ensures every agent operates in a repeatable context, making tests more reliable and environments easier to debug and audit.
- The power of your infra: maintain full control by running environments inside your own cloud or Kubernetes infrastructure. Ensures agents can interact safely with internal APIs and services, and allows platform teams to enforce organizational security policies while maintaining production parity.
- Realistic and consistent: environments that give agents a place to actually run and verify their code, not just suggest it. Each action happens in real-time, in a space that reflects how your application behaves in production. You get tested, working code, not just output that looks right.
- Instant provisioning: automate the provisioning of environments so platform teams can support hundreds of concurrent agent runs without bottlenecks.
- Fully governed and observable: control access, set resource limits, monitor infrastructure usage, and track how many agents are running at any time. Platform teams get full visibility into activity and costs, making it easy to manage scale while maintaining oversight.
These are the same capabilities already used by teams at Monday.com, LaunchDarkly, and Ruggable to power developer environments. Now, they’re available for AI agents too.
Real Use Cases from Early Testing
Early users are already putting Agent Fleets to work on real engineering projects. These aren’t demos or toy apps. They’re production-grade workflows powering real development.
Parallel feature development: A backend engineer launched two agents in isolated dev environments to work on separate features. One agent refactored the billing service to support regional tax logic. The other scaffolded a new FastAPI service for fraud detection. Both agents ran independently, passed all tests, and opened pull requests with production-ready code. This enabled faster delivery across multiple initiatives without added complexity.
Automated onboarding of a legacy Go service: A platform engineer used an agent to onboard a live billing API into Okteto. The agent generated an Okteto Manifest and Docker Compose file, spun up the full environment, connected a seeded Postgres database, ran tests, and opened a pull request with a preview environment. What normally took days was completed in minutes.
Greenfield scaffolding for internal tools: A developer used an agent to scaffold a new FastAPI application for an employee directory. The agent generated the full project structure, wrote tests and configuration files, ran validation with Okteto Test, and launched a preview environment for internal review. This gave the team a production-like foundation to build on quickly.
These stories show the real potential of pairing agents with cloud-native infrastructure. Okteto makes it possible for agents to build, test, and ship working code inside environments that reflect your real application, not just generate suggestions.
As Sourcegraph’s “Revenge of the Junior Developer” puts it, the rise of agents levels the playing field. The right tools make anyone more productive, but only if the environment supports that progress.
That’s the opportunity in front of us. Okteto gives these agents the environment they need to go from limited experiments to productive contributors. Without infrastructure that supports experimentation and scale, agents won’t deliver results.
100x Productivity for the Agent Era
Oketo Agent Fleets represents a fundamental leap forward in how agents are used in modern software delivery, and in how developers interact and manage them.
If AI agents are already delivering a 10x improvement over traditional development for repetitive and routine tasks, teams have the opportunity to multiply that impact by another 10x with parallel agents and Okteto can help get them their faster. With one-click access to instant agent environments, developers can confidently delegate tasks to agents and immediately verify the output without switching contexts or troubleshooting broken setups.
This is not just a tooling upgrade. It is a reimagining of the developer experience. Okteto brings together infrastructure automation, agent orchestration, and production realism into a seamless, developer-first workflow
And while developers will love managing fleets of agents, others prefer deep focus solving a problem with a single agent. The future of development should make room for both. It should amplify strengths, and remove friction.
At Okteto, we’re building with that in mind. The Okteto Platform gives you infrastructure to scale dev, test, preview and agent environments without adding overhead so developers can do what they do best - build great products.
This is what 100x productivity can look like.
Try Okteto AI Agent Fleets
We’re looking for teams experimenting with multi-agent orchestration who want a safe place to run workflows and are ready to push AI in real infrastructure.
Once in beta, it takes minutes to spin up your first agent fleet. No cluster setup or manual provisioning required.
Early users will leverage Anthropic’s Claude Sonnet 4, a top model for coding tasks. Users can start instantly with included LLM credits or Bring Your Own Key (BYOK).
👉 Join the Okteto AI Agent Fleets beta to start scaling parallel AI agent workflows today.
Final Thoughts: The Era of the Manager Nerds
Agentic AI is not a trend. It’s a paradigm shift in how software gets built.
Some developers will become managers of fleets, designing workflows, tuning behavior, and scaling outcomes. Others will stay heads-down, solving big problems. Both need agent environments that run on their infrastructure and don’t get in the way.
At Okteto, we’re building for both. And we’re ready to help you start today.
