OpenAI Powers AWS Bedrock Managed Agents: CEOs Interview
OpenAI CEO Sam Altman and AWS CEO Matt Garman joined an interview to discuss Bedrock Managed Agents, powered by OpenAI. The conversation occurred under embargo, leading to a Tuesday release instead of the usual Thursday schedule, and at 1pm Eastern rather than 6am. The interviewer conducted the session last Friday, asking about OpenAI's Microsoft deal that previously gave Azure exclusive access to OpenAI models.
Late Sunday, word spread of a Microsoft announcement. On Monday, Microsoft and OpenAI revealed an amended agreement. This change lets OpenAI serve products on other cloud providers, including AWS. Microsoft stays as OpenAI's primary cloud partner. OpenAI products launch first on Azure unless Microsoft lacks support for needed features.
Key details from Microsoft's post include: Microsoft remains OpenAI's primary cloud partner, and OpenAI products will ship first on Azure, unless Microsoft cannot and chooses not to support the necessary capabilities. OpenAI can now serve all its products to customers across any cloud provider. Microsoft will continue to have a license to OpenAI IP for models and products through 2032. Microsoft's license will now be non-exclusive. Microsoft will no longer pay a revenue share to OpenAI. Revenue share payments from OpenAI to Microsoft continue through 2030, independent of OpenAI's technology progress, at the same percentage but subject to a total cap. Microsoft continues to participate directly in OpenAI's growth as a major shareholder.
AWS's Impact on Startups and AI Parallels
Azure once held an edge as the sole hyperscaler offering OpenAI models. Enterprises often prefer their existing clouds, benefiting competitors like Anthropic. The exclusivity hurt Microsoft's investment in OpenAI amid Anthropic's growth. OpenAI views AWS as a big chance, forgoing Azure revenue short-term. OpenAI also dropped the AGI clause, extending the deal to 2032 regardless.
Bedrock Managed Agents resembles Codex in AWS. Local setups simplify security and complexity. This offering targets organizations with data in AWS, easing agent workflows across enterprises.
Matt Garman, AWS CEO since his intern days, compared AI to original cloud building. Builders gain access to powerful tools without massive costs. Developers once needed millions for data centers; now a credit card suffices. AI lowers barriers further: small teams build fast without years of coding or large groups.
Early AWS faced no rivals, focusing on fungible compute. AI training demands integrated superclusters and networking. Adoption speed surprised all. Explaining cloud took time in 2006; AI inference shifts focus, but uptake is quicker.
Sam Altman listed platform shifts: Internet, cloud, mobile, AI. Cloud transformed startups at Y Combinator's start. Pre-cloud, startups rented colocation, assembled servers, raised big funds. AWS cut costs, enabling tiny investments. YC rode this wave from early days.
Startups thrive on platform shifts with faster cycles and less capital. AI scales revenue rapidly; YC batch expectations shift monthly. Incumbents adopt quicker than with cloud.
AWS retains startup favor: scale, availability, security, ecosystem. Most scaling startups use AWS. AWS offers credits, advice, go-to-market help. Garman meets startups quarterly.
Many API users pair AWS compute with OpenAI for AI.
Details on Bedrock Managed Agents
Bedrock Managed Agents packages OpenAI frontier models in AWS-native runtime with identity, permissions, state, logging, governance, deployment.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Altman described agents as virtual co-workers performing company tasks. The product aids stateful agents. Codex shows the direction.
Harness matters greatly: runtime, tools, state, memory, permissions, evals. Model and harness integrate, not separable. Users can't distinguish contributions.
Integration occurs post-training and prompting. Tool-calling baked deeper over time. Expect more model-harness, pre-post training fusion. Early days like Homebrew Computer Club.
Users want outcomes, not details. Smarter models need less system prompts. Past GPT-3 hacks now unnecessary.
Customers built integrations themselves: memory, tools, data. Bedrock handles identity in AWS VPC, easing enterprise use.
Pre-AWS, building required servers, engineers. AWS simplified basics. Current models work but setup pains persist, even for non-devs.
Collaboration unlocks reliable features impossible alone.
Local Execution Versus Cloud and Future Challenges
Codex started cloud-based, shifted local for ease: local data, setup. Cloud suits intensive tasks, closed laptops.
Local acts like castle-and-moat security. Production needs zero-trust, permissions.
Local limits scale, sharing, enterprise security. Bridge local-cloud needed, like iPhone apps.
Agents demand production-like builds from start: identities, permissions.
Unresolved: agent accounts versus user. Note agent logins? Primitives lacking. Mental models for software, access evolve with autonomous agents.
Garman noted cloud workloads enable security, policies for agents.
Related on Neura Market