OpenAI just admitted the model wars are over, and they lost the moat.
The Summary
- OpenAI launched Presence, corporate software that connects AI agents to internal company data, workflows, and existing systems for customer support, sales, billing, insurance claims, and IT service requests
- This marks OpenAI's clearest pivot from selling model access to competing directly in enterprise software, putting them against both traditional SaaS players and their own customers
- The product includes testing simulations, guardrails, human review systems, and an investigative tool called Codex that debugs agent failures and suggests fixes
- Access is gated: companies must work directly with OpenAI engineers or approved partners, signaling this is a high-touch, enterprise-grade play
The Signal
The model business is commoditizing faster than anyone at OpenAI will say out loud. When you're two years past "we have AGI" whispers and your main revenue strategy is moving up the stack into middleware, the message is clear: margins on raw inference are collapsing. Anthropic, Google, Meta, and a dozen open-source forks can all generate coherent text. The question is no longer "whose model is smarter" but "whose model is integrated."
Presence is OpenAI's answer to that question. It's infrastructure for the agent economy. The product doesn't just run AI agents, it makes them legible to enterprises terrified of hallucinations and compliance nightmares. You get testing sandboxes, policy guardrails, human-in-the-loop gates, and automated graders that use AI to check AI outputs. That last part is either recursive genius or a red flag depending on how much you trust AI to grade its own homework.
"The agent receives only the knowledge and system access required for that job."
The Codex debugging tool is the sleeper feature here. When an agent screws up, which they will, Codex investigates the failure and suggests fixes. This is OpenAI building the DevOps layer for a world where your "developers" are language models. If that works at scale, it's a wedge into every company trying to deploy agents without hiring a team of prompt engineers.
But here's the conflict: Presence competes directly with OpenAI's API customers. Anyone building agent orchestration tools, workflow automation, or customer service AI is now wondering if their infrastructure partner just became their competitor. Salesforce, ServiceNow, Intercom, all the companies integrating GPT-4 into their products, they're looking at Presence and doing the math. OpenAI wants to sell you the model and the platform and the workflow and the voice layer. That's the Amazon playbook: be the infrastructure, then build on top of it, then compete with everyone using it.
The case-by-case rollout with mandatory engineer involvement isn't a feature, it's a signal that this product isn't ready for self-service. OpenAI is doing services work. High-touch, white-glove onboarding for enterprises who'll pay seven figures to automate their call centers. That's a different business than selling API credits. It scales differently. It requires account executives, solutions architects, customer success teams. OpenAI is becoming an enterprise software company whether they like it or not.
The Implication
Watch who OpenAI partners with for Presence access. Those "designated partners" are the companies that decided integrating was smarter than competing. Everyone else is now building in OpenAI's shadow, wondering when the platform becomes the product.
If you're building agent infrastructure, your wedge is vertical specificity. Go deep in healthcare, legal, logistics, anywhere Presence is too horizontal to follow. If you're an enterprise considering Presence, the real question isn't "does this work" but "how much lock-in am I signing up for." OpenAI is betting you'll trade flexibility for speed. That bet might be right.