Managed Agents show where AI deployment is heading

What Managed Agents actually does

Until now, building a production agent meant months of infrastructure work. Sandboxed code execution, credential management, state checkpointing, scoped permissions, error recovery. All of that before anyone used the thing.

Managed Agents handles that layer. Developers define the agent's tasks, tools, and guardrails. Anthropic runs the infrastructure.

The key capabilities are:

• Long-running sessions that operate autonomously for hours
• Multi-agent coordination, where agents spin up and direct other agents (currently in research preview)
• Scoped permissions and identity management built in
• Execution tracing for every tool call and decision

In internal testing on structured file generation, Managed Agents improved task success by up to 10 points over a standard prompting loop. The biggest gains came on the hardest problems.

Pricing sits at standard Claude API token rates plus $0.08 per session-hour of active runtime.

What the early adopters tell us

The launch quotes are worth reading carefully. They reveal the pattern.

Notion built agents that let teams delegate coding, slides, and spreadsheets without leaving their workspace. Rakuten deployed specialist agents across engineering, product, sales, marketing, and finance within a week per agent. Sentry paired its debugging tool with an agent that writes the fix and opens the pull request.

Every case follows the same shape. A specific, repeatable task. A clear handoff point. A defined scope of what the agent can and cannot touch.

None of these companies asked an agent to run the business. They gave it a bounded job with guardrails.

The governance question matters more than the technology

This is where most SMEs should focus their attention.

Managed Agents includes scoped permissions, identity management, and execution tracing. These are not features. They are requirements. An agent that processes documents, opens pull requests, or sends messages on behalf of your team needs clear rules about what it can access and what it cannot.

Most UK SMEs have not answered that question yet. Many have not written an AI policy. Fewer still have defined which decisions a human must always make.

The technology to deploy agents is arriving faster than the governance to manage them. That gap is where risk lives.

AI adoption is 80% people and culture, 20% technology. That ratio does not change because the technology got more capable. If anything, it sharpens. A more powerful tool in the hands of a team that has not agreed on how to use it creates more problems, not fewer.

What to do this week

You do not need to build an agent today. You do need to start preparing for a world where agents are normal.

Three things worth doing now:

• Write an AI policy if you do not have one. Define what AI can and cannot decide without a human. Managed Agents makes that question urgent, not theoretical.
• Identify your most repeatable, bounded tasks. The early adopters all started with work that has a clear input, a clear output, and a defined scope. Invoice processing, meeting preparation, code review, document extraction. Map yours.
• Train your team on what agents are and how they differ from assistants. An assistant waits for a prompt. An agent acts on a goal. Your people need to understand the difference before agents arrive in their tools.

Where gecco fits

gecco works with UK SMEs through every stage of AI adoption, from first training sessions through to agent readiness. Our SCALE framework moves teams from awareness to independence. Most businesses are still building confidence with Assistants and Automations. That is the right place to be. Agents are the destination, not the starting point.

If you want a clear read on where your team stands, take the AI readiness assessment.