A leaked internal memo from OpenAI lays out a clear message for employees: the company sees the enterprise AI race moving beyond raw model performance and into agent platforms, integrated product stacks, and deployment at scale. The four-page note was sent by chief revenue officer Denise Dresser on April 13 and later published in full by The Verge.
Dresser said enterprise customers are no longer judging AI tools only by model capability. They want systems that fit into workflows, knowledge bases, permission structures, and day-to-day operations. In the memo, she wrote that OpenAI’s biggest constraint right now is not demand but capacity, with hiring remaining a top second-quarter priority.
Spud positioned as the next foundation for work intelligence
The first priority in the memo is winning the enterprise model layer. Dresser wrote that customers are not paying for models in isolation. They are paying for business outcomes: faster writing, better analysis, more efficient coding, stronger customer support, and higher-quality decision making. OpenAI places its upcoming model Spud at the center of that push.
According to the memo, Spud has received positive early customer feedback and is described as OpenAI’s smartest model so far, with stronger reasoning, better understanding of intent and dependencies, stronger execution, and more reliable output in production settings. Dresser added that improved model performance will lift the rest of the stack through higher token limits, lower latency, and more dependable execution across complex workflows.
Frontier and Amazon seen as key pieces of enterprise expansion
The second priority is the agent platform layer. OpenAI argues that the market has shifted from prompts to agents, and enterprise buyers now need more than model output. The memo highlights orchestration, control, observability, security, integration, and governance as core requirements. It positions Frontier as the default platform for enterprise agents, handling how those systems are built, deployed, managed, and scaled.
A third priority is expansion through Amazon. Dresser wrote that OpenAI’s partnership with Microsoft has been a major foundation for success, but it also limits access to some enterprise environments, especially where workloads already live on Amazon Bedrock. Since announcing that relationship in late February, the company says customer demand and inquiries have been extremely strong.
The memo also points to Amazon Stateful Runtime Environment as an important upgrade because it adds memory, context, and continuity across interactions. That would move OpenAI from stateless model access toward systems that can run over time across complex business processes. In OpenAI’s view, this should reduce adoption friction for AWS-native customers and improve its standing with regulated and security-sensitive buyers.
From separate products to a full-stack sales and deployment strategy
The fourth priority is selling a complete AI-native stack rather than isolated products. The memo lists ChatGPT for Work, Codex, API, Frontier, and Amazon Runtime as different entry points into the same broader platform. Dresser said customers begin from different places: some start with employees, some with developers, some with internal systems, and some with external products. OpenAI’s task is to meet them at each entry point and expand usage across the stack.
The fifth priority is deployment. The memo says the biggest obstacle in enterprise AI is no longer whether the technology works, but whether companies can roll it out successfully and at scale. OpenAI’s answer is DeployCo, described as a deployment engine designed to help customers prove value faster, reduce risk, and extend adoption across the organization. Dresser also said DeployCo would work with Frontier Alliance partners as part of a repeatable route to market execution.
The memo also takes direct aim at Anthropic
Beyond strategy, the memo includes an unusually detailed assessment of competition, with Anthropic named directly. Dresser argued that Anthropic’s public narrative is built on fear, restriction, and the idea that AI should be controlled by a small elite. She also wrote that mistakes in compute strategy are now visible in the rival’s products through rate limits, weaker availability, and less stable performance.
The note says Anthropic gained an early foothold through coding use cases, but warns that a single-product approach becomes a burden if AI competition turns into a platform battle. On finances, the memo claims Anthropic’s public $30 billion revenue run rate is overstated because revenue-sharing arrangements with Amazon and Google are being presented on a gross basis. OpenAI’s analysis, as described in the memo, puts that overstatement at roughly $8 billion.