As more options emerge to manage complex multi-agent systems, the era of enterprises connecting operational chains and shadow agents is coming to an end. As organizations move AI agents into production, the question remains: "how will we manage them?"
Google and Amazon Web Services offer fundamentally different answers, indicating a split in the AI stack. Google’s approach is to handle agent management at the system layer, while AWS’s harness method is built at the execution layer.
As competing companies release or update their agent-building platforms – Anthropic – the debate over how to manage and manage them has gained new energy in the past month. agents controlled by the new Claude and improved OpenAI Agents SDK— giving development teams options for managing agents.
AWS with new capabilities added Bedrock AgentCore optimizes speed—relying on trailers to bring agents to product faster—while offering identity and tool management.
Meanwhile, Google’s Gemini Enterprise Kubernetes adopts a management-based approach using a management plane. Each method looks at how agents transition from short-term task helpers to longer-running objects within a workflow.
Upgrades and umbrellas
Here’s what’s actually new to understand where each company stands.
Google has released a new version of Gemini Enterprise, bringing together its enterprise AI agent offerings – Gemini Enterprise Platform and Gemini Enterprise Application – under one umbrella.
The company rebranded As Vertex AI Gemini Enterprise Platformapart from the name change and new features, it’s still basically the same interface, he insists.
“We want to provide a platform and front door for companies to access all the AI systems and tools that Google has to offer,” Maryam Gulami, senior director of product management at Gemini Enterprise, told VentureBeat. “The way you can think about it is that the Gemini Enterprise App is built on top of the Gemini Enterprise Agent Platform, and the security and management tools are provided for free as part of the Gemini Enterprise App subscription.”
On the other hand, AWS has added a new managed agent harness to Bedrock Agentcore. The harness “replaces the initial build with a configuration-based starting point powered by Strands Agents, AWS’ open source agent framework,” the company said in a press release shared with VentureBeat.
Users define what the agent does, the model it uses, and the tools it calls, and AgentCore works to bring it all together to run the agent.
Agents now become systems
The shift toward state-of-the-art, long-running autonomous agents has forced a rethinking of how AI systems behave. As agents move from short-term tasks to long-term workflows, a new class of failure occurs: state slippage.
As agents continue their activities, they collect state-memory, responses, and evolving context. Over time, this state becomes obsolete. Data sources vary or tools may return conflicting answers. But the agent becomes more sensitive to inconsistencies and less accurate.
Agent reliability becomes a systemic issue, and managing this drift may require more than faster execution; may require vision and control.
It’s this point of failure that platforms like Gemini Enterprise and AgentCore try to avoid.
While that change is already happening, Gholami acknowledged that customers will dictate how they want to operate and manage any agent that has been around for a long time.
“We’re going to learn a lot from customers where they’re going to use long-standing agents, where they’re giving these autonomous agents a task to just go ahead and do it,” Gholami said. “Of course there are tricks and balances to iron out, and the agent may come back and ask for more information.”
New AI stack
What is becoming increasingly clear is that the AI stack is divided into different layers, solving different problems.
AWS, and to some extent Anthropic and OpenAI, optimize for faster deployment. Claude Managed Agents abstracts much of the background work to get an agent up and running, while the Agents SDK now includes support for sandboxes and out-of-the-box trailering. These approaches aim to lower the barrier to deploying agents.
Google offers a centralized dashboard to manage identity, enforce policies, and monitor persistent behavior.
Businesses need both.
As some practitioners see it, their businesses need to have a serious conversation about how much risk they are willing to take.
“The bottom line for enterprise technology leaders currently considering these technologies can be stated as follows: while the question of agent harness and runtime is often seen as build and buy, it’s primarily a risk management issue. If you can manage your agents through a third-party runtime because they don’t impact your revenue streams, there’s only one thing to consider from a business perspective,” Rafael Sarim Oezdemir, head of development at EZContacts, told VentureBeat in an email.
Rapid iteration allows teams to experiment and discover what agents can do, while centralized control increases trust levels. What enterprises need is to ensure that they are not locked into systems that are only designed to execute agents in a single way.
