Managed Agents

Your agents in production, watched during business hours or 24/7. We manage every version from release to retirement, and run the AgentCore or Foundry platform they live on.

Problem

Built, approved, and then nobody looked

Somebody builds the agent. Your rules say it may run. Then nobody watches whether it still does its job, which version is live, or what happens when its model is retired. Managed Agents does.

  • • Bad answers nobody can trace to the step that caused them
  • • Quality drifting quietly after a model update
  • • Nobody sure which version of an agent is running, or how to go back
  • • Agents left without an owner when the pilot team moves on
What we run

Approach

Your agents, looked after in production

Managed Agents runs on the same platform project as Managed AI: an AI landing zone with AWS AgentCore or Microsoft Foundry on top. Managed AI keeps the platform safe and in budget. Managed Agents keeps the agents on it working. Built on the platform's own tracing (AgentCore Observability on AWS, Foundry tracing on Azure).

Managed service

Managed Agents

Three things: we watch your agents, we manage their versions from release to retirement, and we keep the agent platform running.

Observability. Every run, prompt and tool call traced. Answer quality, speed, errors and cost per run watched during business hours, or 24/7 for business-critical agents, with alerts when they slip.
Version management. Versions of the agent, its prompts, tools and model. Every change tested before it goes live, and rolled back if it fails.
Agent lifecycle. From first release to retirement, including model switches when a model is retired.
Platform management. AgentCore or Foundry kept updated, patched and running, with capacity and quotas watched.
Talk to us

What it costs

Each agent is priced by size: a one-off price to build it or move it onto the platform, and a monthly price to run it. The highest row where your agent reaches a bigger size sets its size.

SML
Task One focused taskA process with several stepsMany steps or several agents
Integrations 1–2, read-only3–5, writes to one systemMany, writes to several systems
Autonomy Answers and suggestsActs with a person's approvalActs on its own within agreed limits
Users Your own peopleYour people, inside a business processYour customers, or a critical process
Data and risk No personal data, low riskLimited personal dataPersonal data or higher EU AI Act risk
Examples An internal guidance and document assistantDraft replies for customer service in the ticketing systemOrder or claims handling from start to finish
Build €1,000–5,000€5,000–10,000€10,000–20,000
Run €50/mo€100/mo€150/mo
Run, 24/7 €150/mo€300/mo€450/mo
Outcomes

Included in every size: S · M · L

  • • Watched while it matters. Quality, speed, cost and errors followed during business hours, or 24/7 for business-critical agents, with incidents handled through the Cloud² service desk.
  • • Every version known. You always know which version of each agent, prompt, tool and model is live.
  • • Bad releases rolled back. Every change runs through your test set first, and a bad release goes back to the previous version.
  • • Model retirements handled. We follow the notices, test the new model and switch in a controlled way.
  • • Cost per agent. Spend per agent run, so you know what each agent costs to keep.
  • • Agents we didn't build. OpenTelemetry export means it doesn't matter who built the agent.

The Run price covers business hours. 24/7 monitoring for business-critical agents is three times the Run price. Anything bigger than L is scoped in your proposal. The platform itself is built in the AI Landing Zone project, €3,000 per platform, shared with Managed AI. Model usage and cloud costs are billed at cost.

Who does what

We watch the agents, release and roll back versions, switch models and keep the agent platform running. You own each agent and approve what goes live.

Rollback has limits

We roll back the agent to an earlier version, not the actions it already took in your systems. Which actions an agent may take on its own is set in your governance model.

Open by design

Traces are exported over OpenTelemetry, so the service also covers agents we didn't build.

How It Goes

From first agent to many

Agents move onto the platform one by one, each with its own test set. Then we look after them for as long as they run.

Bring your agents in

As they go live

We build new agents by size, or move the ones you already have onto the platform.

A test set per agent

The questions and tasks the agent must get right, agreed with its owner.

Versioned from day one

The agent, its prompts, tools and model under version control.

A named owner

Someone on your side who approves what goes live.

Talk to us

Run them

Business hours or 24/7

Every agent watched, every change tested, every version managed until the agent is retired.

Observability

Quality, speed, errors and cost per run, in business hours or 24/7.

Versions and lifecycle

Releases, rollbacks, model switches and retirement.

The agent platform

AgentCore or Foundry updated, patched and running.

In the monthly report

Quality, incidents and cost per agent.

FAQ

Common questions

Does this include building agents?
Agent builds are priced by size, from S to L, shown under What it costs. Managed Agents takes over once they are in production.
How does this relate to Managed AI?
Both run on the platform built in the same project. Managed AI looks after the platform’s costs, policies, security, guardrails and identity. Managed Agents looks after the agents and the agent platform they run on.
Can you undo what an agent did?
No. We roll back the agent itself to an earlier version, not the actions it already took in your systems. Which actions an agent may take on its own, and which need a person’s approval, is set in your governance model.
Do you cover agents we built ourselves?
Yes. Traces are exported over OpenTelemetry, so it doesn’t matter who built the agent. It moves under management with its own test set.

Field Notes

Agents in practice

What we learn running AI agents in production.

Data

6 min read

How We Compressed 3 Months of Data Platform Work Into 2 Weeks

Building a data platform on Microsoft Fabric should be straightforward. So why does it still take months? Because the hard part was never the technology.

AI

2 min read

Cloud2 and 2NS: Security testing of the twoAI-platform

2NS conducted a security assessment of the twoAI platform developed by Cloud2. The goal of the testing was to verify the software’s level of security and to obtain external expert validation for Cl...

Platform Engineering

6 min read

Agent-Driven Development: Empowering the Next Phase With Platform Engineering

This blog is part of our Platform Engineering series. For a deeper understanding of our perspective on platform engineering, check out our other posts:
View all articles

Ready to Get Started?

Bring your first agent into production, and know someone is watching it.

Contact Us View All Services
Cloud Infrastructure