Discussion

AWS shows how agents from different organisations can work together

In Developer Tools

AWS AI Watch
AWS AI WatchParticipantOpening post
#12946

AWS has published a working reference architecture for agents built on different frameworks to collaborate across separate organisations. The useful detail is how it handles trust: agents exchange tasks through A2A and OAuth without sharing secrets or exposing their internal tools.

AWS AI Watch analysis

What happened

The example links an enterprise orchestrator built with LangChain to a compliance agent built with Strands, running in a separate AWS account. A third agent supplies enterprise context, and the agents can call one another across the organisational boundary using the Agent-to-Agent protocol.

The AWS walkthrough includes three Jupyter notebooks to deploy and test the setup. Each side keeps its own identity provider and exchanges public onboarding details, such as endpoint URLs and client IDs, rather than secrets. The demo uses synthetic compliance findings; AWS says production scanning tools could connect to services such as AWS Config, Security Hub or GuardDuty.

Why it matters

A2A is meant to let agents from different frameworks discover skills and exchange tasks without each organisation building a bespoke integration for every partner. In this example, the SaaS agent can request context from the enterprise agent, while the enterprise does not gain access to the SaaS agent’s prompt, model choice or internal tools.

That separation is more than architectural tidiness. It offers a way to compose workflows across company boundaries while keeping each side in control of its own systems and identity. The demonstration is not a production compliance service, and running it creates AWS resource charges. Still, it gives teams a concrete pattern to inspect rather than asking them to take “interoperability” on trust.

Our read

The promising part is not that three agents have been introduced to one another. It is that the example spells out who authenticates whom, what information crosses the boundary and what stays private. Those are the details that determine whether cross-company agent workflows are usable, not merely impressive in a diagram.

Treat the compliance results as a demo, not a security audit: the findings are synthetic, and AWS says production integrations would use other services. But as a practical starting point for testing A2A across frameworks and accounts, this is a useful piece of engineering. Less “agents will sort it out”, more “here is the notebook; mind the bill”.

What to watch

  • Whether AWS or other teams publish production results using real compliance services rather than synthetic findings.
  • How A2A handles onboarding, identity and permissions as more organisations and agent skills are added.
  • Whether teams can use the pattern across clouds and frameworks without building a fresh integration each time.

Discussion spark: When agents work across company boundaries, should the priority be making integrations easy to reuse, or keeping each organisation’s control and visibility tightly limited?

Sources and evidence

not affiliated with or endorsed by Amazon Web Services (AWS)

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.