Ben Thompson interviews Sam Altman and Matt Garman about Bedrock Managed Agents

Ben Thompson interviews Sam Altman and Matt Garman about Bedrock Managed Agents

On April 28, 2026, Ben Thompson interviews OpenAI CEO Sam Altman and AWS CEO Matt Garman about Bedrock Managed Agents, a joint product for running agents in an organization’s AWS environment. Thompson argues that Microsoft’s agreement change lets OpenAI serve customers on AWS, addressing a conflict: Azure’s exclusive access to OpenAI models had made it harder for enterprises to use them on their existing cloud. Altman and Garman describe the product as more than access to OpenAI models: it brings models together with AWS capabilities such as identity, permissions and data access. They say this could make organizational agents easier to build and manage, while Altman says their collaboration may also enable products that are difficult or impossible to build otherwise. The interview’s central tension is that local agents can be easier to use, while sharing them across an organization raises questions about identity, permissions and security. Garman says AWS will provide frontline support; Altman says the right account and access model for agents remains unresolved. Their broader conclusion is that developers may build a new generation of products as model capabilities progress, but the architecture and uses of these systems remain unsettled.

Publication: 74db1907-c165-88ed-804f-df5d28b81932

attribution — pass: This is a single Stratechery interview by Ben Thompson with Sam Altman and Matt Garman. The reading correctly attributes the framing about the Microsoft deal to Thompson and the product statements to Altman and Garman.

Last Friday I conducted the following interview with OpenAI CEO Sam Altman and AWS CEO Matt Garman about Bedrock Managed Agents, powered by OpenAI

summary — uncertain: The summary is faithful, but its central term "agents" (and "stateful agents" in the claims) is never explained, and the definitions list is empty. A non-specialist may not grasp what the product actually runs. Smallest repair: one plain-language gloss drawn from the source, such as Altman's description of AI agents doing work inside a company as "virtual co-workers." A suggested term-library addition is "stateful agent." Otherwise the actors, the deal context, the local-versus-cloud tension and the unresolved identity question are rendered clearly.

Virtual co-workers is kind of my least bad of the ways I’ve heard this described

date — pass: The reading's April 28, 2026 matches the originalPublished instant's UTC day and the page dateline. Kind is publisher-page, so no archive disclosure is needed. The retrieval date is noted. Forecasts, such as new products and future architecture, are framed as expectations. There is no hindsight framing.

Tuesday, April 28, 2026

support — pass: Each claim matches its quoted source text. Further statements are also supported: AWS frontline support, local agents being easier while sharing raises permission and security issues, and the Azure exclusivity tension. The anticipated benefits are labeled as expectations, not outcomes.

AWS will be that frontline support that you kind of interact with

media — pass: No media was selected and none was supplied, so there is no identity or relevance issue.