What Are Managed Agents? Examples, Costs and How to Start

Managed agents are what most owners actually want when they say they want an AI agent. Not another repo to maintain, not a weekend evaluating LangGraph against CrewAI, not a conversation with a cloud architect about which region to pin a Postgres instance to. They want the agent to be there, to do the job, and to send the invoice to approvals instead of to the void.

That is the promise of a managed agent: the vendor runs the agent for you, you give it the task. The question is which service to pick, what it really costs, and how to start without a six-month project. This explainer answers that for teams in Germany and the EU, with Superkind as the European choice next to AWS, Microsoft, Google and OpenAI.

What Are Managed Agents?

A managed agent is an AI agent that the vendor hosts and operates for you: it runs on the vendor’s infrastructure, uses models the vendor routes to, keeps the sessions and memory, applies the approval and audit rules, and connects to your systems through integrations the vendor maintains. You use it; you do not run it. The opposite is a self-built agent, where your team writes the code, hosts the runtime and keeps the agent loop alive in your own cloud with tools like n8n, LangGraph or custom code.

IBM frames an AI agent as a system that can “reason, plan, and act to achieve goals” using tools and memory⁷. McKinsey adds that agentic AI is built on generative foundation models that can act in the real world and execute multi-step processes⁸. “Managed” is the operational prefix: all the plumbing that makes an agent production-grade - compute, model access, session state, tool calls, logs, policy - is the vendor’s problem, not yours.

How Superkind Does It: The European Managed Agent for Teams

Superkind is an AI employee for your team, hosted and operated by Superkind in Europe. You write to it in Outlook, email, Microsoft Teams, Slack or the Superkind web app like a colleague, and it does the task in your existing programs through 1,500+ integrations - DATEV, SAP Business One, HubSpot, SharePoint, Excel - then asks before anything goes out. Hosted in Europe, GDPR compliant. There is no agent loop for your team to run and no model to self-host.

1. 1

   You write the task

   Forward a supplier invoice in Outlook or write in Teams or Slack, in plain language, the way you would brief a colleague. No new tool to open, no SDK to pick.

2. 2

   Superkind does it in your systems

   It works across the programs you already use - DATEV, SAP Business One, HubSpot, lexoffice, Personio and 1,500+ others - with your company knowledge and your rules, so nobody copies data between systems by hand.

3. 3

   It asks before anything goes out

   Before Superkind sends, books or changes anything that leaves your team, it shows you the result and waits for Approve or Reject. Approvals happen in the channel you are already in.

4. 4

   It becomes a routine

   What you explain once keeps running (“every Monday, send the standing quote”). Every task makes Superkind smarter, and that knowledge - your customers, prices and processes - stays in the company even when people leave.

From the first task to your own app

A managed agent should grow with the work. When Superkind outgrows single tasks, you do not start over - you build on top of what it already knows about your company.

• Start - sign up for free, give the first task in Outlook, Teams or Slack, and watch it get done.
• Relief - what you explain once keeps running as a routine. Before anything goes out, Superkind asks for approval.
• Knowledge - every task makes Superkind smarter. The Company Brain (your customers, prices, processes) is never lost.
• Growth - when the work gets more complex, build custom solutions, apps with their own interfaces and integrations on the platform. Your own Super App, built yourself or with the Superkind team. It all belongs to you.

How Managed Agents Work

Under the hood, every managed agent service is a short list of the same parts. Knowing them helps you compare offerings without getting lost in product-page prose.

1. A hosted runtime - the compute that runs the agent loop. On AWS this is AgentCore Runtime; on Microsoft it is Foundry hosted agents in VM-isolated sandboxes³; on Google it is Agent Runtime in the renamed Gemini Enterprise Agent Platform⁵. With Superkind, Europe-hosted compute is simply part of the service.
2. Model routing - the service picks the model (or lets you pick) and handles the API call. Bedrock gives you Anthropic, OpenAI, Meta, Amazon and others; Foundry gives you the Microsoft model shelf plus Azure OpenAI; Google uses Gemini. Superkind uses OpenAI, Anthropic, Mistral and Google, processed in Europe.
3. Session state and memory - the agent remembers past turns, past tasks and (for most services) your company data. Vendors offer this as a managed store so you do not run a database for it.
4. Tools and integrations - how the agent actually touches your systems. The big-cloud platforms expect you to wire up tools yourself through APIs or MCP servers. Superkind brings 1,500+ integrations out of the box.
5. Approvals and audit - the human-in-the-loop step and the log of what the agent did. Managed agents built for business work ship this; developer frameworks leave it to you.
6. Policy and identity - who can run which agent, what data it can see, how long it keeps context. Microsoft now assigns each hosted agent its own Entra Agent ID³; Superkind maps this to your existing SSO and Trust Center DPA.

Managed vs. Self-Built in Comparison

The real split is not technical, it is where the work ends up. The table below lays the two models side by side, with a developer framework shown for contrast.

A self-built agent gives you the most control; a managed agent gives you the most delivered work for the least staff cost. Most Mittelstand teams do not need their own agent runtime any more than they need their own Postgres fork.

Examples: The Managed-Agent Services on the Market

“Managed agent” covers a wide range in practice, from an infrastructure platform you still build on, to a finished AI employee you message. Superkind is the European example built for teams; the others are listed below with what each one really is, from vendor pages.

• Superkind - the AI employee for teams in Europe. You write it a task in Outlook, Teams, Slack or the web app; it works in DATEV, SAP Business One, HubSpot, SharePoint and 1,500+ other systems with your rules and knowledge, asks before anything goes out, and runs on EU-hosted infrastructure, GDPR compliant. Free start up to 150 euros in credits, Business from 50 euros per month for unlimited users (as of October 2026).
• Amazon Bedrock AgentCore - AWS’s managed runtime for agents, with memory, policy and a managed harness in preview that lets you define an agent from a model, system prompt and tools¹. In October 2026, AWS added Bedrock Managed Agents powered by OpenAI as a public preview on top². You own the agent design and the integrations; AWS runs the infrastructure.
• Microsoft Foundry Agent Service - Microsoft’s hosted platform for building and running agents on Azure. Prompt agents are defined in the Foundry portal; hosted agents are your own code running in VM-isolated sandboxes with persistent file systems and an Entra Agent ID³. Reached general availability in 2026⁴.
• Google Gemini Enterprise Agent Platform - the rebranded Vertex AI Agent Builder. Agent Runtime (formerly Agent Engine) is the managed runtime, billed per vCPU-hour and per GiB-hour of memory, with model tokens priced separately⁵.
• OpenAI Agents SDK / Assistants API - OpenAI’s managed route for agents. The Agents SDK is the successor to the Assistants API and is tied to OpenAI’s Responses API; it is OpenAI-only, so you cannot use Anthropic, Google or open-source models in the same agent without going outside it.
• Microsoft Copilot Studio - Microsoft’s low-code builder for agents inside the Microsoft 365 stack. Billed in Copilot Credits via a 25,000-credit pack for 200 US dollars a month or pay-as-you-go at 0.01 US dollars per credit⁶. Internal agents used by staff with Microsoft 365 Copilot licences draw no additional credits.

What Does a Managed Agent Cost?

Pricing follows the model. Infrastructure platforms price infrastructure; team-level agents price the work.

• Big-cloud managed agents - usage-based. Google Agent Runtime lists on-demand rates around 0.0864 US dollars per vCPU-hour and 0.0090 US dollars per GiB-hour of memory, plus model tokens and session storage⁵. Final cost depends on how long each task runs and how many run per month.
• Copilot Studio - 25,000 Copilot Credits for 200 US dollars a month, or 0.01 US dollars per credit pay-as-you-go. A generative answer costs 2 credits, an agent action costs 5⁶. One industry example put a customer-facing support agent at about 1,700 US dollars a month at 10,000 conversations.
• OpenAI Agents SDK / Assistants - priced on tokens plus hosted tool use. There is no per-agent flat fee; you pay for what the agent reads, writes and runs.
• Superkind - free start with a one-off up to 150 euros in credits, no credit card and no sales call. Business from 50 euros per month (25,000 credits), with the popular step at 100 euros (50,000 credits), 200 and 500 euro steps, and unlimited users - no price per seat. A question costs 10 to 50 credits, a task 50 to 500 credits, and “100 euros covers up to 1,000 tasks or 5,000 questions”. Enterprise is custom. Prices as of October 2026, excl. VAT.
• The hidden cost either way - staff time on the self-built path. A platform team to run your own agent loop costs far more than any managed service bill.

Data protection and EU hosting

A managed agent acts on your data, so where it runs is not a footnote. Many big-cloud agent services default to US regions and expect you to configure data residency and sign the DPA yourself. Superkind handles this by design.

For back-office work this matters most where personal data lives: supplier invoices with contact names, HR files in Personio, customer records in HubSpot. A managed agent you can point at those systems without a US transfer is a different conversation at the data-protection review.

How to Start With a Managed Agent

You do not need a strategy deck, an in-house AI team or a six-month architecture review. Pick one task and let the managed agent prove itself.

If you want to see what this looks like in a specific role, the AI back-office clerk page walks through the end-to-end tasks a managed agent takes over in administration. The integrations page lists the systems Superkind connects to today.

Conclusion

Managed agents move the hard part - running an agent in production - off your team and onto the vendor. The big-cloud services are the infrastructure layer: powerful, usage-priced, still a build. Superkind is the European managed agent for teams: an AI employee you write to in Outlook, Teams or Slack, with 1,500+ integrations, EU hosting and approval before anything goes out.

Pick the task on your Monday that eats the most hours of back-office time. Give it to a managed agent this week and approve the first result. That is the whole shift from “we should look into agents” to “the agent handled it”.

Frequently Asked Questions

Related Articles

• Enterprise AI Agent Platforms: A Buyer’s Guide
• AI Governance Platforms: What to Look For
• Multi-Agent Orchestration: How It Works
• How to Build an AI Agent Without Code

Henri Jung

Co-founder of Superkind, the AI employee teams write to in Outlook, Teams and Slack. Henri helps companies, from small teams to the Mittelstand, hand their routine work to AI - without new tools, consultants or an in-house AI team.