Last updated: 22 September 2026
AI models can answer questions. AI agents are supposed to go further — using tools, working through multiple steps and completing tasks on a user's behalf.
But building reliable agents requires much more than simply connecting an AI model to a few tools.
Amazon Web Services has now released Strands Harness, an open-source framework designed to handle some of that complexity.
WHAT IS STRANDS HARNESS?
Strands Harness sits between an AI model and the environment in which an agent operates.
It helps developers manage things such as tools, context, memory and the repeated interactions required for an agent to complete longer tasks.
Importantly, AWS has designed it to work with different AI models rather than requiring developers to build around a single provider.
Developers can build an agent locally and later deploy it in different environments without completely rebuilding its underlying architecture.
WHY CONTEXT MATTERS
One of the biggest problems with AI agents is context.
An agent working on a long task can accumulate huge amounts of conversation history, tool results and other information. Sending all of that back to the AI model repeatedly increases token usage and cost.
Strands Harness introduces context-management techniques designed to decide what information the model actually needs.
AWS says this approach reduced token usage by 28% across six internal benchmarks while maintaining comparable accuracy.
That figure comes from AWS's own testing, so independent benchmarks will be useful for understanding how well those savings translate to other applications.
WHY THIS IS INTERESTING
The AI industry is moving beyond chatbots toward systems that can actually perform work.
A coding agent might inspect files, modify code and run tests. A research agent might search multiple sources, compare information and produce a report. Other agents could interact with business systems and automate repetitive workflows.
The underlying AI model is only one part of making those systems work.
Frameworks such as Strands Harness are attempting to build the infrastructure around the model that manages tools, memory and context.
AWS has released the project under the Apache 2.0 licence, allowing developers to inspect, modify and build on the technology.
The bigger AI competition may therefore not only be about who builds the smartest model.
It may also be about who builds the best systems around those models.
MicLinks Radar — Discover. Understand. Build.
