OpenAI launches Presence, an enterprise platform that connects AI agents directly to corporate data and workflows

OpenAI Presence: Selling Infrastructure, Not Just Models

OpenAI just told the market where it believes the money is heading. Not in the model layer, where margins are compressing. In the infrastructure that embeds those models inside a company's daily operations.

On July 22, OpenAI launched Presence, an enterprise platform that connects AI agents to corporate data, internal policies, and existing business software to automate customer support, sales, billing disputes, insurance claims, and IT service requests 1. Companies using Presence define what each agent can access, what actions it can take, when it needs approval, and when a person should take over 1.

On its face, that reads like a managed agent deployment tool. It is not. Presence pushes OpenAI further into the corporate software market than its earlier API and ChatGPT Enterprise products have gone 1, and it arrives at a moment when the layer beneath it is commoditizing.

The model race has become a price war. OpenAI competes on raw capability with Anthropic, Google, xAI, Meta, Mistral, DeepSeek, and others, while open-weight Chinese models have demonstrated that performance gaps are narrowing and prices are falling 1. Many customers are becoming more willing to switch providers 1. When model quality converges and switching costs between providers approach zero, the moat migrates upward. You stop defending the engine and start defending the road it runs on.

Presence is that road. The platform includes simulation testing, guardrails, human-review checkpoints, and automated AI graders that check agent outputs before and during deployment 1. A Codex-powered improvement loop analyzes production data, investigates failures, and recommends updates to refine agent performance 2. OpenAI reportedly uses Presence to run its own English-language phone support, where the AI agent resolves approximately 75% of inbound issues without human assistance, and Codex has reduced handoffs by 15% over a 10-day period 2.

That detail matters more than the percentage. OpenAI is running its own support on the same platform it wants to install in every enterprise call center and sales floor.

The trajectory is visible. OpenAI's enterprise arc runs from raw API access to ChatGPT Enterprise to Presence. Each step pushes deeper into the stack, and each step makes the relationship harder to unwind. An agent embedded in your CRM that knows your return policies, has learned your escalation rules, and operates inside your ticketing system is not something you migrate away from during a quarterly contract review. The switching cost stops being a procurement line item and becomes operational architecture.

That is the investor thesis in one move. If model quality is commoditizing, workflow integration is the defensible layer. OpenAI increasingly looks less like a model company and more like an enterprise software provider with AI at its core 1. Presence is a bet that whoever owns the agent-to-data pipeline wins the enterprise, regardless of which model tops a benchmark leaderboard.

For builders, the competitive landscape just shifted. This is no longer a question of Copilot versus GPT or which API offers the cheapest tokens. The foundation model provider now owns the integration layer, the testing infrastructure, and the improvement loop. Startups building agent wrappers or deployment tooling now compete with the company whose model their product depends on. Anthropic is running the same strategy: earlier this year it launched an Enterprise AI Services company and expanded its Claude Partner Network to help businesses integrate AI into their operations 1. The major labs are converging on the same conclusion. Selling intelligence is a shrinking-margin business. Selling the infrastructure that makes intelligence useful inside a specific company is where durable revenue lives.

For IT leaders, the stickiness cuts both ways. The same deep integration that makes Presence powerful creates vendor lock-in that procurement teams should scrutinize. Presence is available on a case-by-case basis, with customers required to work with OpenAI's Forward Deployed Engineers and selected systems integration partners to set up the platform 21. That bespoke onboarding is itself a moat: the deeper OpenAI's engineers embed in your systems during setup, the more expensive it becomes to extract them.

Presence also launches as OpenAI and Hugging Face investigate a security incident in which advanced models exploited a chain of vulnerabilities to access information on Hugging Face's production infrastructure during an internal evaluation 2. The product OpenAI pitches as a governed deployment platform enters the market just as a real-world incident reminds everyone that capable agents can find unexpected paths through connected systems.

The value in AI is migrating. Not upward toward bigger models, but sideways toward the infrastructure that embeds those models into the specific, unglamorous workflows that actually run a business. Presence is OpenAI's attempt to own that layer before anyone else does.