Managed Agents - Don't Get Locked In

Author: Sam Witteveen · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Managed Agents - Don’t Get Locked In · Channel: Sam Witteveen · 10:39 · Watch on YouTube

Topics: AI agents, AI infrastructure and hardware, Local and open-source models, Open source and software ecosystems

Summary

The video explains how managed agents combine a model, an agent harness, and a cloud runtime or sandbox, and distinguishes services that provide the loop from those that host a loop you bring yourself. It compares offerings from Anthropic, Google, AWS, Microsoft, and LangChain, using Claude Managed Agents to illustrate concepts such as agents, environments, sessions, and events. The video argues that these products can be economically attractive to providers because long-running agents use substantial tokens and may add session charges, while also creating different forms of vendor lock-in. It closes with practical questions for choosing between a managed agent and self-hosting, including persistence, server-side code execution, data residency, and zero-data-retention requirements.

Key points

A managed agent typically combines a model, a harness that handles the model-and-tools loop, and a runtime or sandbox where the agent runs.
Some offerings provide the harness and loop; others let users bring their own loop and host it in an isolated runtime.
The video describes Claude Managed Agents in terms of agents, environments, sessions, and events.
Managed agents can manage sessions, events, memory, tools, skills, credentials, tracing, and scheduling.
Long-running agent interactions can trigger multiple reasoning and tool loops, increasing token use; some services may also charge for session time.
Model providers can lock users into their model suites, cloud providers can tie agents to their serving and memory infrastructure, and LangChain’s operational lock-in centers on LangSmith.
Managed agents may suit tasks that need persistent state across disconnects or a server-side sandbox for code execution.
Self-hosting may be preferable when the agent loop is the product, or when data cannot leave an organization’s infrastructure or zero-data retention is required.
For locally hosted models, the video suggests considering local sandboxing options such as Docker sandboxes.

Chapters

Start	Chapter	Summary
00:00	What managed agents are	The video introduces the growth of managed-agent offerings and explains the model, harness, and runtime layers, including two ways providers host agent loops.
03:44	Inside Claude Managed Agents	Using Claude as an example, the video explains agents, environments, sessions, and events, and notes how managed services handle related infrastructure.
05:21	Economics and lock-in	The video discusses token use, possible session-hour charges, and how model providers, cloud providers, and LangChain create different forms of lock-in.
08:34	Choosing managed or self-hosted	The video offers questions about persistence, code sandboxes, control of the agent loop, data residency, and local deployment to guide the choice.

Quotes

these managed agents are the providers running the harness and a sandbox, not just the model. — 01:04

The managed agents is for all the long-running async stuff. — 02:40

So you’re now paying for time, not just tokens. — 06:24

Their lock-in is gonna be operational. — 07:28

First off, do you need state that survives a disconnect and can resume days later? — 08:34

Concepts and entities

Managed Agents, Harness, Claude Managed Agents, Gemini Managed Agents, AgentCore, Foundry hosted agents, Deep Agents, LangSmith, Docker sandboxes

Full transcript (original language)

00:00 Okay, have you noticed that over the past five months or so, almost every major AI company has shipped a product called managed agents? alongside people shipping new models, alongside people shipping new frameworks, we’ve got this whole new area of managed agents. So Anthropic was the first one to kick it off. They started it in April or even perhaps a little before with some of the ideas there And I’ve gotta say, most of the other companies have just copied their ideas from the Anthropic offering. So Google came out in May with their Antigravity agent, but really,

00:33 followed around the same time with their Gemini managed agents. and then June brought AWS to the game with Agent Core. Microsoft, also June , Foundry hosted agents. And then LangChain came along in mid-August with a managed deep agents public beta. And supposedly on September 29th, OpenAI’s dev day, they’re gonna roll out their version of managed agents as well. So this is often on top of other SDKs and other offerings that the companies are actually offering.

01:04 So these managed agents are the providers running the harness and a sandbox, not just the model. And what I want to do in this video is open up one of these and show you actually what’s in the box. and then I’ll look at why these six companies, all with different businesses, all decided that this was the product they needed to get into. And it kind of tells you a lot about how they’re thinking about this entire business strategy. Okay, so what actually is a managed agent? You can think of this as three layers. The bottom layer is the model, and of course, that started out

01:36 with Opus and Claude at the time. now has moved to Fable on the Anthropic side and the latest GPT models on AWS and Azure. Oh yeah, and then, don’t forget Gemini. But on top of the model, the middle part is the harness, and this basically is the loop. this is what calls the models, what picks the tools, what runs them And it feeds all the results back in and manages that context window. and then finally on top of these is the actual runtime. So where does the loop actually run?

02:08 The sandbox that holds your credentials that often does observability or tracing The thing that survives and lives in the cloud by itself. So the whole managed element in managed agents here means the provider owns level two and level three of this structure. And of course, if you’re Anthropic or OpenAI, you actually own that model as well. so the pitch has been quite simple. People have talked about the different APIs for messages, for chats, for completions. They’re simply what allows you to communicate with the model.

02:40 The managed agents is for all the long-running async stuff. So you can kind of think of this as two flavors. Flavor one, they’ve got the loop. You send in your tasks and a config. You never have to write that loop. You never have to actually call that model. and that’s basically Claude managed agents, Gemini agents, Agent Core Harness, and even the LangChain managed deep agents. Flavor two, on the other hand, you actually bring the loop. So perhaps this is a loop that you’ve written yourself with LangGraph or with ADK or it’s just your own code from scratch.

03:11 You put it in a container, they host it for you with some kind of per session isolation. But they also provide the runtime for things like tool execution, memory and other server side tools as well. So that’s basically things like the Foundry hosted agents, Agent Runtime used to be called Agent Engine a few weeks ago and may change its name again in a few weeks time And even LangChain has LangSmith deployment for your own LangGraph code. It’s kind of worth noting that both Google and LangChain are kind of having an each-way bet on both of these which certainly for Google makes a lot of

03:44 sense because you’ve got some things being built by Google DeepMind and other things being built by Google Cloud. Another axis that you should be looking at for some of these things too is for the harness itself, is it closed like Claude, Antigravity, AgentCore, et cetera, or is it something that’s actually open source like Deep Agents? All right, so since Claude managed agent seems to be the template that started this all off, Let’s look at the actual concepts in there. So you’ve got four things here, agent, environment, session, and events.

04:16 So the agent obviously is the model, the system prompt, tools, the ability to call MCP servers, skills and these are things that you create once and then reference by ID. The second thing is the actual environment. So this is where it all runs. The cloud sandbox of where it’s run could either be an Anthropic cloud sandbox or they are supporting a variety of different other sandboxes, including some self-hosted sandboxes on your own infra. Next up, you have the session. This is one running instance doing one task, and then you have the

04:49 actual events, the messages that go back and forth, the user turns, the tool results, the status updates. Now pretty much all of this maps over to the Gemini managed agents where they’re giving you a fresh Linux sandbox as a remote environment and one of the cool things that they’ve got there is that you can actually customize that file system and mount your own skills or your own agents MD file. So the managed agents are literally taking care of all these sessions and events and memory and tools and skills and credentials and

05:21 tracing and scheduling all for you. So why are they all doing this? Well, basically, it’s the economics. And if you think about it, the economics of one run of a chat call is just literally a round trip. You call the model, you get something back, maybe your agent uses it, et cetera. And a good example of this is even Google’s own docs say that a single Antigravity agent interaction typically burns somewhere between 100K and 3 million tokens.

05:51 and that basically all happens because a single interaction can trigger multiple reasoning and tool loops. You don’t really have control over that. That’s all running server side. Now currently during preview, Google’s not actually charging you any extra for the Session server, but that certainly hints that that’s gonna come just like Anthropic with their Claude managed agents, that’s charging you a session hour fee on top of the tokens. So you’re now paying for time, not just tokens. And you’ve gotta think these long-running agents like this are the highest token

06:24 product a model company can sell. It runs while you’re asleep. It manages all the different turns itself. You just have to tell it what to do, hope that it actually does that because most of these agents don’t have any sort of built-in verification system. And then you just have to pay the bill. Of course the other thing is with the companies that actually are using their own model, this all becomes a lock-in game to lock you into their suite of models. you wanna use Claude managed agents? You’re gonna have to use Claude Fable, Opus, Sonnet, Haiku.

06:56 You want to use Gemini managed agents? you’re going to have to use one of the Gemini models. now if you look at the cloud companies, whether that’s AWS, Azure, or even Google Cloud, they’re taking the opposite bet. Their managed agents, whether that’s AgentCore, whether that’s ADK, et cetera. They’re making the model interchangeable when the runtime isn’t. So of course, you’re running it on their serving architecture. you’re gonna have to probably use it on their memory system or a memory system that’s running in their cloud. and then you’ve got LangChain with the deep agents.

07:28 So, they’re really a framework company. So they’re neither a model nor a cloud. They’re basically trying to be the control plane. and the cool thing there is that they allow you to use any model, any sandbox. This can be self-hostable if you want it to be. But their lock-in is gonna be operational. Their goal is to get you to use LangSmith, and once your traces, evals, deployments, et cetera, are all in LangSmith. Moving from that is gonna be a really big challenge. So let me make the point. None of these services is inherently bad. They all have things going for them.

07:59 Running Claude Managed Agents is an easy way for you to basically get something up, get it hosted, get it running, give it a sandbox, give it a memory, give it access to all the different things that you get with the Claude Agents SDK. And in many ways, if you’re already using a lot of the Claude models, this can often just end up making your life a whole lot easier. You might want to be a little bit more careful with the cloud providers though, because they’re often taking a number of different products that they had going before and are now bolting them together to make their version of managed agents.

08:34 So how should you think about whether you want to go with a managed agent system or whether you want to take a harness and run it yourself? Well, there are five key questions that you should be asking yourself. First off, do you need state that survives a disconnect and can resume days later? If the answer is yes there, you probably want to go with the managed agents. It just makes things a lot easier. Second, do you need a sandbox for code executions that’s gonna live server side? Again, if the answer’s yes, you probably wanna give managed agents a good look.

09:04 But what if you’re designing your own agent? Or what if your loop is actually your product? In cases like that, you definitely want to look at self-hosting and perhaps just using parts of these managed solutions like a sandbox in the cloud or a memory tool or an observability tool. Lastly, one of the key things you wanna make sure that you ask yourself too is can your data actually leave your infra, and do you actually need zero-day retention? If your data is not allowed to leave your own infra or you do need zero data retention, that rules out things like Claude managed agents.

09:36 it’s just not eligible for zero data retention Because session store history and sandbox state are both server side, and that rules out a lot of enterprises that are actually looking at doing this kind of thing. So each of these managed agents both has their pluses and their minuses. They each have their own amount of control that you actually get, to run your agent in their cloud. And of course, I haven’t even touched on local AI. If you’re actually using a locally hosted model, you want to be able to put all of this on your own machine anyway.

10:07 And then perhaps you wanna look at things like Docker sandboxes, which allow you to have a safe and secure sandbox running fully locally. So over the next few weeks, I’m planning to have a look at a number of these different managed agent solutions, different harness solutions both the ones that I’ve mentioned here, and perhaps The OpenAI one, if they ship at the end of September. And also some of the open agent systems that we’re now seeing start to evolve out there. So let me know in the comments if you’ve tried these out, what you actually found to be the pluses or minuses of the managed agents.

10:39 And which ones seem most interesting to you? Anyway, as always, if you found the video interesting, please click like and subscribe, and I will talk to you in the next video. Bye for now.