AWS cuts agent setup to 3 calls

An AI agent is software that keeps looping: it reads a prompt, decides whether to call a tool, uses the result, and tries again. Amazon Web Services says Bedrock AgentCore can now run that loop for developers instead of making them wire it up first. (aws.amazon.com) AWS announced the update on April 22, adding a managed harness in preview, an AgentCore command-line tool, and new AgentCore skills for coding assistants. The company said the goal is to move from an idea to a working prototype faster. (aws.amazon.com) The headline claim is that a developer can define an agent with a model, system prompt, and tools, then run it in three API calls. AWS said the harness handles reasoning, tool choice, action execution, and response streaming without orchestration code. (aws.amazon.com 1) (aws.amazon.com 2) In plain terms, the harness is the scaffolding around the model: compute, storage, security, authentication, and failure handling. AWS said most teams had been spending days building that scaffolding before they could test whether an agent was useful. (aws.amazon.com) AWS is also pushing a new AgentCore CLI that scaffolds projects, runs local development with hot reload, deploys with AWS Cloud Development Kit, and invokes deployed agents. The documentation lists support for Strands Agents, LangChain and LangGraph, Google Agent Development Kit, and OpenAI Agents. (docs.aws.amazon.com 1) (docs.aws.amazon.com 2) Another piece of the update is persistent filesystem state, which keeps an agent’s files, installed packages, and build artifacts across stop-and-resume cycles. AWS paired that with a command API that can run shell commands inside the session’s micro virtual machine. (aws.amazon.com) That matters most for coding agents, where the working directory acts like short-term memory. AWS said earlier sessions lost generated code and dependencies when the micro virtual machine shut down, forcing teams to rebuild state or push files into Amazon Simple Storage Service by hand. (aws.amazon.com) AWS is not replacing the code-based path. Its documentation says code-based agents remain the default and general-availability option, while the managed harness is still in preview for teams that want the fastest route to a running agent. (docs.aws.amazon.com) The preview also has clear limits. AWS said the managed harness is available in four Regions — US West (Oregon), US East (N. Virginia), Europe (Frankfurt), and Asia Pacific (Sydney) — while the CLI is available in 14 Regions where AgentCore runs. (aws.amazon.com) AWS says developers who outgrow the config-based harness can export the orchestration into Strands-based code and keep the same platform underneath. The pitch is simple: start with configuration, then drop into code only when the agent needs custom routing or multi-agent logic. (aws.amazon.com)