OpenAI
2026-07-24
OpenAI Launches Presence Agent Platform, Shifts Control Plane to Lock Enterprise AI Deployment
Summary
Key Takeaways
On July 23, 2026, OpenAI launched Presence, an enterprise agent deployment platform for conversational and voice agents. Core components include model inference, Permissions, Policies, Evaluation, Escalation Rules, and post-deployment improvement tools. OpenAI emphasizes that agent products are not just models but the 'control plane around the model,' defining who the agent acts for, which systems it accesses, when to escalate to human, how failures are reviewed, and how to improve based on real conversations. Business model references OpenAI's own production phone support line.
Same day, ChatGPT Health became fully available to logged-in US users 18+, integrating Apple Health and other services to connect medical records. It competes with Microsoft Copilot Health and Google Health Connect, facing HIPAA compliance and medical device classification challenges.
Strategically, OpenAI shifts from 'model API' to 'agent platform,' moving from model commoditization to platform lock-in and subscription revenue. The agent control plane becomes new infrastructure, echoing AI agent security moves by Cisco Antares, Palo Alto CyberArk, CrowdStrike SGNL. Consumer market accelerates, with health AI as a killer app. Meanwhile, Microsoft replaces OpenAI with MAI image models (85% cheaper), signaling a subtle shift in their relationship.
Why It Matters
OpenAI's Presence platform, while ostensibly helping enterprises deploy agents, is actually defending against Salesforce Agentforce, Microsoft Copilot Studio, and Google Vertex AI Agents by locking users into its control plane definitions. Once permissions, policies, and escalation rules are configured, migration costs become prohibitive.
OpenAI downplays the latency risk of the control plane: real-time agent interactions must pass through policy evaluation, model inference, and escalation logic, potentially increasing tail latency and hurting user experience. The platform's scalability under complex policies and high concurrency remains unproven.
ChatGPT Health may underestimate HIPAA compliance costs and privacy risks from integrating Apple Health data. Potential FDA classification could add regulatory burdens.
PRO Decision
【Vendors】Competitors should counter OpenAI's lock-in by offering open standard agent control planes supporting multiple model backends, emphasizing portability and interoperability. Microsoft Copilot Studio can strengthen Azure AI integration while ensuring compatibility with OpenAI models to lower switching costs. Salesforce Agentforce can leverage CRM data for deeply integrated workflows and highlight data sovereignty and compliance. Google Vertex AI Agents should promote its open model ecosystem and lock-in-free deployment options.
【Enterprises】CIOs and architects should conduct zero-trust audits of Presence: demand permission policy export APIs for data portability; evaluate latency and throughput under peak loads, especially policy evaluation impact on real-time conversations; require third-party security audit reports for HIPAA compliance; avoid fully binding core business logic to Presence, retain human escalation and fallback capabilities; adopt multi-cloud agent deployment strategies.
【Investors】Recognize that OpenAI's shift from model API to agent platform aims to boost valuation and subscription revenue, but competition is fierce and Microsoft's defection (MAI model replacement) signals risk. Long-term profitability depends on enterprise adoption and retention, but lock-in tactics may invite regulatory scrutiny. Monitor real adoption rates, churn, and compliance costs; be wary of OpenAI losing model API flexibility due to platform over-reliance.
Get 3-5 key AI infrastructure signals weekly →
💬 Comments (0)