AWS takes an open source turn in its agent-building strategy

We keep hearing that agents are going to be a big deal. Even if that hasn’t happened yet at scale, it's a fair bet that agents are going to become a significant part of every company's enterprise software strategy in the years ahead. The hyperscalers are trying to create platforms where companies can build and manage agents in an effort to anchor workloads on their infrastructure.

This week, AWS announced Strands Labs, an open source agent development platform, available on GitHub. "We’re introducing Strands Labs, a new Strands GitHub organization designed to give developers the ability to get hands-on with experimental, state-of-the-art approaches to agentic AI development," the company wrote in a blog post.

This builds on the company's Strands Agents SDK released last year, which Amazon claims has been downloaded over 14 million times, suggesting there is a desire for an agentic framework. But instead of simply layering the experimentation piece onto the SDK, the company created the lab part as a separate project.

The idea is to encourage agent research and testing in a safe way, hence the sandbox approach. There's a twist here though. The company isn't just providing this tool as part of the Amazon ecosystem. Instead they have decided to make it open source through an Apache 2.0 license. 

But the separation of Labs from the Strands Agents SDK isn’t merely cosmetic. It reflects a deliberate architectural and strategic choice, says David Linthicum, a longtime Deloitte cloud consulting executive who now runs his own firm, Linthicum Consulting.

"The real signal is AWS separating production-ish Strands Agents from Strands Labs. This matters because developers can ship fast, break things and iterate on agent experiments without destabilizing the core. That’s a very enterprise-aware move," Linthicum told FastForward.

What's in the box?

For starters, the company is offering three projects out-of-the-box as a way to prime the pump for developers. These include robotics, which connects agents to physical systems; a robotic simulation environment for testing in a simulated 3D environment; and AI Functions, which lets developers define tools in natural language that agents can invoke.

Jay Anderson, an analyst at Moor Insights & Strategy, likes the project approach as a way to attract outside companies and consulting firms to use the tools. "AWS has been making solid inroads with partners and these new projects like the robotics project could be very useful to attract new builders and partners," he said.

Linthicum believes Strands Labs is a step in the right direction because it gives developers a way to experiment as agentic frameworks mature, but it won't solve every problem that's holding agents back from wider deployment. "It offers less orchestration glue, more built-in patterns, and crucially more attention to the unsexy stuff like instrumentation/observability (OpenTelemetry) and a deployment toolkit," he said. But that won't solve big problems around governance, security, and "who’s accountable when the agent does something dumb."

The monetization challenge

Anderson points out that Amazon offers a level of stability that other frameworks could be missing. "If you are worried about things like scale, security and model choice from a framework, it’s a good option versus the more closed agentic frameworks out there," he said.

Amazon is walking a fine line with this approach. By open sourcing Strands Labs, it makes things easier for developers who don’t want to be aligned too closely to one provider. Keeping it separate from the main SDK helps preserve its reliability in real‑world use, even if that means pushing monetization further down the road.

"The big challenge for AWS is going to be how they measure the success of this strategy. Open sourcing is often a circuitous path to monetization and is a less direct GTM motion," Anderson said. The real test will be whether open experimentation ultimately leads to production workloads running on AWS.