Your Agent’s Next Thousand Tools – Mike Chambers

“Agentic AI: Towards the Next 1,000 Years” Mike Chambers – AWS

This document serves as a reference for Mike Chambers – AWS keynote on October 1, 2026. It summarizes the key concepts and announcements so that you can immediately apply the knowledge gained.

1. Major Theme: Unleashed The Agentic AI Ecosystem

The overarching theme of the keynote was “Agentic AI: Towards the Next 1,000 Years” . The central message is that we have reached a point where building complex, agentic AI systems no longer needs to be complicated “string manipulation”. The focus is shifting from manually coding the “while loop” that drives an agent to specifying and configuring a standardized Harness . The core formula introduced by the speaker is: Model + Harness = Agent . By working with modular, scalable components such as the Amazon Bedrock Agent Core and the open-source Strand Harness SDK , developers can focus on what really matters: creating intelligent, autonomous, and valuable AI agents for the future.

2. Subthemes and Chapters

The speaker structured the keynote around the evolution of building, managing, and innovating agentic AI.

Step 1: The Core of an Agent – The Harness

The traditional while loophas evolved into the concept of a Harness . This is the fundamental structure that connects all components of an agent: the prompts , the skills (via progressive disclosure), the tools to interact with the outside world, the agency (freedom of action), and the memory .

The speaker emphasized that for most use cases, a simple data structure (a harness.jsonfile) suffices to define a complete agent. This principle forms the basis of the Amazon Bedrock Agent Core , a service that converts this specification into a scalable, managed agent.

Step 2: Build and Evaluate to Scale

Next, the speaker introduced the Amazon Bedrock Agent Core , a “massive toy box” with managed components that can be deployed modularly:

• Managed Runtimes: For directly deploying your agent based on the harness specification.
• Memory: A serverless solution for managing short- and long-term memory.
• Gateways: An API gateway that makes your data sources (APIs, databases) accessible to your agent.

A crucial part of this phase is evaluation and observability . The speaker warned that an agent that gives no errors does not necessarily work well . Agent-based systems are non-deterministic and complex. The new Amazon CloudWatch Armory is designed to address this problem by giving developers insight into their agent’s traces, logs, and interactions directly in their IDE (via the Armory plugin). This makes it possible not only to see that the agent is responding, but also whether the response is correct, and to iterate quickly.

Step 3: The Future – Open Source Innovation with Strand Harness SDK

The technology behind the Bedrock Agent Core is the Strand Harness SDK , a fully open-source framework for building and deploying agents. This framework is provider-agnostic, hostable anywhere, and gives developers the freedom to fully customize the harness for advanced applications.

The most future-oriented section was Strand Harness Labs , the R&D wing where experimentation with the next generation of AI takes place. The speaker gave an example of a robot controlled by the SDK, which demonstrates the connection between agentic AI, robotics, and world models.

The Big Announcement: Strand Decisive-B – An Open Source Decision Model

As a highlight, the speaker announced the direct release of Strand Decisive-B , a small, fast, and fully open-source decision model (also known as the “System 1” model). Unlike large language models, this model is optimized to make quick decisions based on data. Because it is small enough to run locally on a laptop, the speaker encouraged the audience to experiment and innovate with it. The repository and the model weights went live during the presentation.

3. Inspirational Quotes

1. The ‘while loop’ is now central to some of the most advanced programs we have, and that is, of course, the agent.
2. “Model plus harness. That is what an officer does.”
3. “Is it green on the dashboard? Often it is. But does it really work? The problems don’t always lie with the officer themselves, but in the upstream components.”
4. We are watching Strand Decisive-B. This is a fully open source, yes, truly open source, no open-weights, machine decision model.
5. I encourage you all: get started with this, clone this, take this model and build something really cool, because it is small enough to run on your own machine.

4. Key Action Points

1. Experiment with the Agent Core CLI: Install agent-core-cliand use it agent-core create harnessto generate the data structure yourself harness.json. Start by defining your own tools and skills to understand the basic principles of an agent.
2. Integrate Observability From Day One: Use the Amazon CloudWatch Armory plugin in your IDE. Start monitoring and debugging your agent interactions immediately. Create evaluation datasets from your traces to continuously measure and improve your agent quality.
3. Download and Innovate with Strand Decisive-B: Go to the Strand Harness repository, download the new Strand Decisive-B model, and start building applications that require fast, local decision-making. Share your creations with the community.