Amazon Web Services Inc. said today it’s taking steps toward simplifying the creation of the underlying infrastructure for new autonomous artificial intelligence agents.
With an update to its Amazon Bedrock AgentCore platform today, developers can access a new “managed agent harness” feature and command-line tools that eliminate the manually intensive backend work that slows down new projects. The update comes on the eve of Google Cloud Next, where Google LLC is expected to announce a raft of agent-related services and partnerships.
The company said it’s trying to speed up the time it takes to get new AI agents up and running in production. By giving AgentCore a prebuilt orchestration layer and tighter integration with the most popular AI models, AWS is replacing a days-long setup process with a task that only takes a couple of minutes to get through.
In a blog post, AWS explained that one of the main challenges developers face in building AI agents isn’t the underlying logic. Rather, it’s setting up the “agent harness” that includes the compute resources, authentication protocols, persistent storage and sandboxes for code execution – all essential resources agents need to run in production.
Another major task involves building the “orchestration loop” logic, which is needed to call the underlying AI model, decide which tools to use and manage context windows. This underlying infrastructure is a major pain point that forces developers to spend days thinking about infrastructure rather than agent behavior.
To get around this, AWS has launched the managed agent harness in AgentCore, which is a managed, framework-agnostic platform designed to build, deploy and operate AI agents at scale without managing infrastructure. Powered by the open-source Strands Agents framework, it enables developers to define a new agent’s model, tools and instructions quickly using a simple configuration file instead of writing custom code each time.
With this approach, swapping out a model such as Claude Opus 4.6 for Google Gemini 3 no longer requires teams to rewrite masses of new code. Instead, they can simply adjust their application programming interface parameters and make it happen almost instantly.
AWS is also streamlining the process of deploying new agents in production. Typically, this means stepping outside the code editor it was built in to set up a deployment pipeline and configure environments – another mundane, hours-long manual process. With the new AgentCore CLI, developers can retain the same workflow across the full development lifecycle, from prototyping to deployment. AgentCore CLI handles all of the logistics using infrastructure-as-code, ensuring that the agent’s configuration is reproducible and version-controlled.
Finally, in a nod to the reality that most developers are using coding assistants to create the code for new AI agents, AWS has announced a suite of prebuilt skills for Kiro, Claude Code, Codex and Cursor. This is all about enhancing context, giving coding agents curated, up-to-date knowledge of AgentCore’s best practices. The company said this is invaluable, because it means AI coding agents will make fewer wrong turns on an AgentCore platform that evolves at breakneck speed.
These updates push Amazon Bedrock AgentCore to the forefront of AI agent development. Its tight integration with agentic frameworks like CrewAI, LangGraph and LlamaIndex allows it to serve as a solid foundation for running AI agents at scale. Early adapters of the new features say they can now validate new ideas in minutes rather than days or weeks.
“Parrot Analytics helps media, entertainment and investment firms understand what content audiences care about, using AI agents to turn unstructured signals into intelligence to reduce risk and increase content investment returns,” said Parrot Analytics Inc. Engineering Director Sanjeev Sharma. “With AI agents now running across the product suite and internal teams, the team is scaling fast. Amazon Bedrock AgentCore’s new developer experience capabilities give the company’s agent developers a faster path from idea to deployment. Switching models or adjusting agent behavior is a configuration change, not a rewrite, so they can experiment more and ship improvements faster.”
AWS said the managed agent harness tool is launching in preview in four regions: US West (Oregon), US East (N. Virginia), Asia Pacific (Sydney), Europe (Frankfurt), while the new AgentCore CLI and coding assistant skills can be found in every AWS region that currently offers AgentCore.
Visit Silicon ANGLE to read this article.
