AWS has announced Amazon Bedrock AgentCore, a new set of services that assist developers in bringing AI agents into production.
Swami Sivasubramanian, VP of Agentic AI at AWS, made the announcement during the cloud giant’s 2025 New York Summit.
In his keynote, Sivasubramanian discussed how “undifferentiated heavy lifting” slows down the development of AI agents.
For instance, a developer may build something exciting on their laptop, demo it, get the CEO excited, but then they hit a brick wall trying to scale it and get it production-ready.
To break down that wall and make AI agents enterprise-ready, IT teams must go through several cumbersome processes. These include:
- Choosing the right models
- Doing extensive prompt engineering
- Creating evaluation frameworks
- Building memory and personalization features from scratch
- Setting up tool stacks manually
All these processes create what AWS termed a “chasm of production readiness”, with prototypes on one side and production on the other.
That’s where Amazon Bedrock AgentCore comes in, serving as the bridge between the two.
To provide that bridge, AgentCore comprises seven tools to assist developers. Examples include AgentCore Observability for a step-by-step visualization of agent execution, AgentCore Runtime for ensuring AI agents function in low-latency serverless environments, and AgentCore Identity for securing AI agent access to various business systems.
Nevertheless, all tools aim to help developers bring AI agents into production faster, reliably, securely, and at scale across the enterprise.
What’s the Relevance for Customer Service?
Customer service is ground zero for AI agents. Recognizing this, AWS released a preview of how businesses can deploy a “production-ready customer support assistant” with Amazon Bedrock AgentCore.
The customer support assistant connects with internal systems, retrieves context, and drafts responses for the service team.
Per the preview, developers can utilize various AgentCore tools to deploy such an assistant. One of those tools is AgentCore Memory, which automatically adds long-term memory to the assistant.
As a result, the assistant can act like a real agent, understanding what accounts the customer has, what they called about last time, etc.
So, instead of re-establishing context, like reviewing case notes or past bills, the assistant is already up to speed. That allows for more personalized, continuous customer conversations.
While that’s an exciting possibility, AWS customers can seemingly only build such an assistant via Amazon Bedrock, as there was no mention of whether this memory capability will become available to Amazon Connect customers.
Amazon Connect is AWS’s CCaaS solution. Over the years, AWS has become proficient at leveraging tools within its broader portfolio, like Bedrock, to improve Connect and differentiate the offering. Indeed, it’s one of the reasons why the tech juggernaut won the Best Enterprise Contact Center Platform prize at the recent CX Awards 2025.
So, while there is no mention of Connect, AWS likely has something up its sleeve for its CCaaS customers.
P.S. One key AgentCore feature not mentioned above is its compatibility with open-source systems and support for any foundational model, even if it’s not hosted on Bedrock. That provides more flexibility and extensibility for developers.
