AIToday
Large Language ModelsOpen-Source AIAI Business & IndustryAmazon AI BlogPublished: Nov 20, 2026, 06:00 JST

Amazon Bedrock AgentCore harness moves to Strands harness

Send on LINE
Amazon Bedrock AgentCore harness moves to Strands harness

3 Key Points

  1. What happened

    Amazon upgraded the AgentCore harness to run on its open-source Strands harness, available under an Apache 2.0 license, and added lifecycle hooks so deterministic code can allow or deny agent actions at defined checkpoints.

  2. Why it matters

    With hooks, a Lambda function can check a refund against an approval limit before the refund tool runs, turning non-deterministic AI behavior into predictable, governed workflows, and the upgrade requires no new resource type or integration.

WHO IT HITSEnterprise teams building AI agents on Amazon Bedrock, especially those handling consequential actions like refunds, gain built-in policy checks and managed defaults without operating the underlying infrastructure themselves.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Amazon positions the upgrade as a way to remove assembly work from agent building. Strands harness supplies recommended instructions, tools, and context management, while AgentCore harness operates that foundation as a managed AWS service. Each session runs in a secure, kernel-isolated microVM in AgentCore runtime, so customers do not provision compute, host the agent loop, or operate tool-execution infrastructure.

The managed defaults are meant to evolve. When the model is omitted, the harness uses a recommended default model that changes as new models are released; proactive context management summarizes older conversation turns once a configurable threshold is reached. Explicit configuration, such as a model ID, systemPrompt, truncation policy, or allowedTools list, always takes precedence, which lets applications that need a fixed security or capability boundary opt out of the shifting defaults.

Lifecycle hooks are the other half of the pitch. Because they apply to subagents too, a delegated task cannot bypass a refund policy set at the parent level, and every decision appears in the InvokeHarness response stream as a hookEvent. Amazon also points customers to a Refund Desk sample in its AgentCore samples repository that walks through all four lifecycle events and the Lambda, SNS, and EventBridge target types.

FAQ
Do I need to migrate or adopt a new resource type to get these updates?
No. There is no new resource type or integration to adopt; both updates arrive automatically in the AgentCore harness you already use.
Which lifecycle events can hooks run on?
Harness supports four: before_invocation, before_tool_call, after_tool_call, and after_invocation. Only Lambda hooks can change the agent loop, while SNS and EventBridge hooks send notifications and do not block it.
What happens if a Lambda hook times out or errors?
The hook applies its failureMode, which defaults to deny.
Amazon AI BlogRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleAnthropic opens at $111 after record IPO