Artificial intelligence news.
Straight from the original announcements.

AWS upgrades Bedrock AgentCore with Strands defaults and lifecycle hooks

November 19, 2026, 21:01By the AI News Bites Editorial TeamSource: AWS
AWS upgrades Bedrock AgentCore with Strands defaults and lifecycle hooks
Image: AWS

Amazon Bedrock AgentCore harness now uses AWS’s open-source Strands harness as its foundation, adding managed defaults, built-in tools and lifecycle hooks for controlling agent actions. AWS says the update is available automatically in the AgentCore harness customers already use, with no new resource type or integration required.

Key points
  • AgentCore harness now uses the open-source Strands harness foundation.
  • Managed defaults include a recommended model, task-oriented instructions and prompt caching.
  • Built-in tools support planning, web retrieval, filesystem access and task delegation.
  • Lifecycle hooks can allow, deny or report agent requests and tool calls.
  • The updates arrive automatically in the AgentCore harness already in use.

A managed foundation for longer agent tasks

Strands provides an assembled agent foundation with recommended instructions, tools and context management. AgentCore operates that foundation as a managed AWS service, so customers do not need to provision compute, bundle and host the agent loop, manage session scaling or operate tool-execution infrastructure themselves.

Each session runs in a secure, kernel-isolated microVM in AgentCore runtime, with managed compute, scaling and high availability. AWS connects the harness to production services including session persistence, streaming responses, usage metadata, observability, AgentCore memory, Identity, Gateway, Browser, Code Interpreter and customer-provided containers.

Customers retain control over the model, instructions, tools, permissions and lifecycle policies, while AgentCore manages the underlying execution environment.

Defaults add tools and context controls

When customers omit the model, Harness uses a recommended default model that evolves as new models are released. A managed system prompt directs the agent to gather context, use suitable tools, continue until a task is resolved, adapt to failures or denied actions, and verify its work before reporting completion.

For supported models using Converse on Amazon Bedrock, automatic prompt caching covers the system prompt, tool definitions and conversation history. AWS says repeated calls with the same prompt and tool surface can use lower-priced cache-read tokens, reducing input processing latency and cost.

Harness also summarizes older conversation turns after the context window reaches a configurable threshold. AWS says this allows Harness to run at about 34% lower token cost than other leading harnesses. Additional capabilities include context offloading, structured planning through todo_write, task delegation through subagent, webpage retrieval through web_fetch, and separate read, write and edit filesystem tools.

Lifecycle hooks put checks around consequential actions

Lifecycle hooks let customers run deterministic code at defined points in the agent loop. AWS Lambda hooks can allow or deny requests or tool calls, while Amazon Simple Notification Service topics and Amazon EventBridge event buses can record activity or start asynchronous workflows without blocking the agent.

The four supported events are before_invocation, before_tool_call, after_tool_call and after_invocation. A Lambda denial before invocation stops the invocation, while a denial before a tool call skips that call and lets the agent loop continue. After-tool-call hooks retain the result and stop further inference, while after-invocation hooks report the decision. Output that has already streamed cannot be retracted.

AWS gives refund approval as an example. A before-tool-call Lambda can deny an issue_refund request above the $500 auto-approval limit, allowing the harness to skip the refund while the agent tells the customer that the request was escalated. Hooks also apply to subagents, so delegated tool calls run through the same before-tool-call and after-tool-call checks.

Existing configurations remain in control

Managed defaults can be overridden. Customers can specify a model ID, replace the managed system prompt, configure a truncation policy or list the allowed tools. If allowedTools is omitted or set to ["*"], Harness supplies the current recommended built-in tool set. AWS advises applications with a fixed security or capability boundary to list only the tools they need, such as read and web_fetch without shell access or file modification.

Customers configure hooks with CreateHarness or UpdateHarness. If a Lambda hook times out or returns an error, its failure mode applies and defaults to deny. The update therefore combines a broader default toolset with explicit controls for organizations that need predictable handling of high-impact agent actions.

Related news

AWS lets AI agents buy x402 products through Marketplace billing
Products and toolsNovember 20, 2026, 00:02

AWS lets AI agents buy x402 products through Marketplace billing

AWS has added AWS Billing to Amazon Bedrock AgentCore payments, allowing AI agents to discover and purchase x402 enabled software, APIs, data and content through AWS…

BMW Group sets governance model for hundreds of AI agents
For developersNovember 19, 2026, 23:20

BMW Group sets governance model for hundreds of AI agents

BMW Group is developing an enterprise governance model for hundreds of AI agents used across its R&D organization, where more than 12,500 developers work on…

AWS SageMaker AI Adds Monitoring for Inference Endpoint Drift
Products and toolsNovember 19, 2026, 22:28

AWS SageMaker AI Adds Monitoring for Inference Endpoint Drift

Amazon SageMaker AI can now identify latency drift, underused GPUs and missed serving optimizations on eligible generative AI inference endpoints. The monitoring is…