Anthropic launched Claude Managed Agents today. A lot of people are already talking about how revolutionary it is, and I'm sure more will be in the following days and weeks.
I find this a little amusing, because they're treating it like a totally new category, but it's not. Google has had a managed agent runtime called Agent Engine on Vertex AI for months. You hand over your agent, it runs on managed infrastructure, and you get sessions, scaling, and observability. This is pretty much the same benefit that Claude Managed Agents purports to provide.
Thisdoesn't mean the two platforms are interchangeable, but Claude Managed Agents are also not revolutionary and it's worth understanding how it fits into what's available.
I built the same agent on both platforms today, and the developer experience is genuinely different. But the "Anthropic just changed everything" framing is kind of overblown.
Managed Agents is a hosted service where you define an agent (model, system prompt, tools, MCP servers), define a container environment, start a session, and stream events back and forth. The Console has a guided quickstart that walks you through it.
The built-in tools that they give you agent include bash, file operations, web search, and a text editor. You can also provide your agent with custom tools connected through MCP servers. The agent runs on Anthropic's infrastructure, and sessions persist across disconnections. You never write an orchestration loop, they abstract that away. You describe what you want, and Anthropic runs the harness for you.
Their engineering blog was explicit about why they designed it this way. They said harness assumptions go stale as models improve. They want to own the harness so they can swap it out as Claude gets better, without requiring you to redeploy. That's a reasonable argument if you trust Anthropic to make the right decisions for your use case. But of course the tradeoff is that you have less control over certain things.
Agent Engine takes a code-first approach. You write your agent as Python (typically with the ADK), test it locally, and deploy it to a managed runtime on Agent Engine in Vertex AI. The deployment takes 3-5 minutes while Agent Engine provisions a container and installs dependencies. Once deployed, you get managed sessions, Cloud Trace, Cloud Logging, and IAM integration.
The same code that runs on your machine runs in production. You can iterate locally quickly without going through anyone's cloud infrastructure. Especially for complex agents with multiple tools, that local development loop can be really helpful.
Additionally, the tools are literally written in Python. Anything you can do in Python you can explicitly and easily pass into your agent as a tool. In Claude this would typically be done through MCP.
For simple agents, Managed Agents wins on setup time. Under a minute versus a Python project and a deployment wait. But Agent Engine gives you things configuration can't. With Agent Engine can customize the orchestration loop, add retry logic between tool calls, chain agents together, or swap models by changing one string. With Managed Agents, you can steer via events but you can't really rewrite the loop.
There's also the issue of model lock-in. Claude Managed Agents runs Claude only, but Vertex AI Agent Engine is model-flexible and framework-flexible. It runs the ADK, LangChain, LangGraph, CrewAI, and others, and has more than just Google'sGemini models available.
Another thing is tool definition. Claude's built-in tools require no configuration, which is more convenient if they give you what you want but also they are less flexible. On Agent Engine, tools are Python functions. More flexible (your tool can be anything you can write in Python), but you're making it yourself.
Additionally, obviously if your agents need to talk to Cloud Storage, BigQuery, or respect your org's IAM policies, Agent Engine is already inside that ecosystem. The Managed Agents equivalent is connecting MCP servers for each service.
The bigger story is that Anthropic just ventured into being a platform company. Until today, they basically sold access to their models. Now they're selling a platform that hosts your agents too.
It kind of makes sense strategically that they would try to close this gap with Google and other companies.
So Claude Managed Agents and Vertex AI Agent Engine are different, and Claude Managed Agents is useful, but it's not like Anthropic revolutionized the game with this launch.
