The era of prompt chains and shadow agents linking together enterprises is nearing its end as more options emerge for orchestrating complex multi-agent systems. As organizations move AI agents into production, the question remains: "How will we manage them?"
Google and Amazon Web Services provide fundamentally different answers, reflecting divisions in the AI stack. Google’s approach is to run agentic management at the system layer, while AWS’s harness method is set in the execution layer.
The debate over how to manage and control gained new energy last month as competing companies released or updated their agent builder platforms – Anthropic with new cloud managed agents and OpenAI with enhancements to the Agent SDK – giving developer teams options to manage agents.
Bedrock is optimizing for AWS velocity with new capabilities added to AgentCore – relying on harnesses to get agents to product faster – while still offering identity and device management.
Meanwhile, Google’s Gemini Enterprise takes a governance-centric approach using a Kubernetes-style control plane. Each method offers a glimpse of how agents move from short-burst task assistants to longer-lasting entities within the workflow.
upgrades and umbrellas
To understand where each company stands, here’s what’s really new.
Google released a new version of Gemini Enterprise, bringing its enterprise AI agent offerings—Gemini Enterprise Platform and Gemini Enterprise Applications—under one umbrella.
The company has rebranded Vertex AI as the Gemini Enterprise Platform, although it stresses that apart from the name change and new features, it is still basically the same interface.
“We want to provide a platform and a front door for companies to access all the AI systems and tools that Google provides,” Maryam Gholami, senior director of product management for Gemini Enterprise, told VentureBeat in an interview. “The way you can think about it is that Gemini Enterprise Applications is built on top of the Gemini Enterprise Agent Platform, and the security and governance tools are all provided for free as part of the Gemini Enterprise Applications subscription.”
On the other hand, AWS added a new managed agent harness to Bedrock AgentCore. The company said in a press release shared with VentureBeat that Harness “replaces upfront builds with a configuration-based starting point powered by Strands Agents, AWS’s open source agent framework.”
Users define what the agent does, what models it uses and what tools it calls, and AgentCore works to tie them all together to run the agent.
Agents are now becoming systems
The shift toward statewide, long-lived autonomous agents has forced a rethinking of how AI systems behave. As agents move from short-lived tasks to longer-running workflows, a new class of failure is emerging: state drift.
As agents continue to work, they accumulate situation-memory, reactions, and emerging context. With time that condition becomes chronic. Data sources change, or tools may give conflicting responses. But the agent becomes more sensitive to inconsistencies and less truthful.
Agent reliability becomes a system issue, and managing that drift may require more than fast execution; This may require visibility and control.
This is the failure point that platforms like Gemini Enterprise and AgentCore try to prevent.
Although this change is already happening, Gholami acknowledged that customers will decide how they want to run and control any long-running agent.
“We’re going to learn a lot from customers where they may be using long-running agents, where they assign these autonomous agents a task to just go ahead and do,” Gholami said. “Of course, there are tricks and balances to get right and the agent can come back and ask for more input.”
New AI Stack
What is becoming increasingly clear is that the AI stack is separating into different layers that solve different problems.
AWS and, to a lesser extent, Anthropic and OpenAI, optimize for fast deployment. Cloud Managed Agents eliminate much of the backend work to stand up an agent, while the Agents SDK now includes support for sandboxes and ready-made harnesses. These approaches aim to reduce the barriers to getting agents up and running.
Google provides a centralized control panel to manage identities, enforce policies, and monitor long-running behaviors.
Enterprises probably need both.
As some businessmen see it, their businesses have to have a serious conversation about how much risk they are willing to take.
“The main conclusion for enterprise technology leaders currently considering these technologies can be formulated this way: While the agent harness vs. runtime question is often thought of as build vs. buy, it is primarily a matter of risk management. If you can afford to run your agents through third-party runtimes as long as they do not impact your revenue streams, that’s OK. Conversely, in the context of more critical processes, the latter option is the only option to consider from a business perspective. “Rafael Sarim Ozdemir, head of growth at EZContacts, told VentureBeat in an email.
Rapid iteration allows teams to experiment and explore what agents can do, while centralized control adds a layer of trust. Enterprises need to ensure that they are not locked into systems designed for a single way for agents to execute.
<a href