OpenAI put a phone number near the center of its Presence launch: 1-888-GPT-0090. The company says its English-language phone support now runs on Presence, verifying callers, reading account context, taking approved actions, and escalating to humans when needed.
The point is not another chat interface. OpenAI is moving agents from ChatGPT team tools into customer support, sales, claims, IT service requests, and other production workflows that touch customers, money, or company systems. What it is selling is not just a model. It is a package of permissions, evaluations, guardrails, human approval, and post-launch improvement loops.
Start with one job
Each deployment begins with a specific job, such as resolving a billing dispute, supporting an insurance claim, or handling an employee IT request. The agent receives only the knowledge and system access needed for that job. The enterprise defines what it can do, what needs approval, and when a human should take over.
Business Insider frames the launch as a move deeper into enterprise software. Presence connects internal data, policies, existing software, and workflows to automate support, sales, billing, and similar processes. Access is still heavy-touch: limited general availability, led by OpenAI forward deployed engineers or selected systems integrators, not self-serve.
Read the 75% figure carefully
OpenAI says that in its own English phone support channel, Presence reached or exceeded its frontline human-support quality benchmark within weeks and now resolves 75% of inbound issues without human help. It also says a Codex-powered improvement loop cut human handoffs by 15 percentage points in 10 days.
Those figures should be read as OpenAI-reported numbers, not audited customer benchmarks. Their value is that they reveal the design: real sessions produce escalation and quality signals, Codex proposes changes to policies or tool use, and teams test and approve controlled rollout.
The first customers are heavy industries
The public design partners include BBVA, SoftBank, and IAG, spanning banking, telecom, and insurance. OpenAI quotes Daniel Ordaz, head of AI transformation at BBVA Mexico:
"At BBVA, we are working closely with OpenAI to explore how trusted customer agents can help shape the future of financial services."
The claim is less about a better chatbot than about whether trusted customer agents can enter financial-service workflows. SoftBank is testing natural Japanese-language conversations, while IAG is looking at claims support during severe weather and other high-demand moments.
Buyers will look for first-contact resolution, handoff rate, wrong-action rate, audit logs, cost per resolved case, and regression test results after policy changes. Those are operating metrics, not model-leaderboard metrics.
OpenAI is filling the deployment gap
The last three months now form a line: workspace agents in April, the OpenAI Deployment Company in May, the Partner Network in June, and Presence in July. OpenAI is filling the delivery, integration, and operations layer around its models.
That puts pressure on systems integrators and support-software vendors. Sierra, Salesforce, ServiceNow, Microsoft Copilot, and others are all selling enterprise agents that do work, not just answer questions. OpenAI has models, ChatGPT distribution, and forward deployed engineers. Its hard part is that industry workflows are granular, customer systems are old, and deployment cost is hard to standardize.
If the 75% self-reported resolution rate appears in customer cases, if limited availability becomes easier to buy, and if enterprises accept Codex-suggested workflow changes inside regulated processes, OpenAI will look less like a model-access vendor and more like an enterprise software company.
Sources: OpenAI Presence announcement, OpenAI Presence product page, Business Insider, CocoLoop; checked launch scope, limited availability, voice and chat use cases, support/claims/IT workflows, OpenAI-reported phone-support figures, and enterprise-software framing.