OpenAI introduced a new platform for businesses called Presence on July 22, 2026, which integrates AI agents for phone and chat support firmly into internal systems. The service monitors interactions through policies and approval rules and escalates to humans when necessary. Presence is currently only available to selected large customers, not as a regular subscription.
Presence combines voice agents with approval rules
Presence relies on three elements, as OpenAI describes in the official announcement: Policies and permissions define what an agent is allowed to do. Automated simulations and evaluations test the agent before going live. Escalation rules hand over a conversation to a human if needed. Before launch, teams specifically test the agent against common inquiries, edge cases, and risky scenarios. In ongoing operations, a Codex-supported system evaluates interactions and escalations and suggests targeted adjustments that employees must review and approve before implementation.
According to OpenAI’s own statements, this feedback mechanism reduced the number of conversations escalated to humans by 15 percentage points within ten days. The technology is first being tested in-house: OpenAI’s English-language phone support is already running on Presence and resolves about 75 percent of incoming calls without human intervention – a figure that is not independently verified. The channel can be reached at the number 1-888-GPT-0090.
Banks and insurers are testing the service in everyday use
OpenAI mentions initial customer examples in collaboration with the trade media VentureBeat: The Spanish bank BBVA is testing voice support for everyday banking transactions in Mexico, the Japanese telecommunications company SoftBank is testing natural language customer conversations in Japanese, and the insurer IAG is evaluating the system for rapid assistance in major disaster situations such as severe weather. All three deployments are still in the testing phase, and none of the companies provided specific contract details or revenue figures.
The trend towards AI agents in the workplace extends beyond customer service. OpenAI itself had only recently introduced ChatGPT Work, an agent for internal office tasks such as documents, spreadsheets, and presentations. Other providers are also following suit – for example, the fintech company Block, which introduced AI agents as full team members with the chat app Buzz. Presence differs from this by having direct contact with external customers and stricter approval rules for risky actions.
Access remains limited to selected corporations
For companies, Presence is not a product for self-booking. Access, according to OpenAI, is exclusively through its own account team and is implemented by in-house Forward Deployed Engineers and selected system integrators. OpenAI does not disclose public pricing. Costs are determined individually for each customer based on the use case, which trade media such as The Register interpret as a shift to a more profitable consulting model. Whether and when Presence will also be available to European or German companies remains unclear – the company did not mention a region in the announcement. Since its launch on July 22, the service has been in a limited general availability. OpenAI did not provide a date for a broader opening.
The market researcher Gartner had already predicted before the launch that about half of the companies that wanted to downsize their customer service due to AI systems would likely reverse these plans by 2027. Human judgment remains irreplaceable in many situations, according to this view. Presence responds to this criticism with built-in escalation rules that are intended to specifically hand conversations over to humans rather than relying entirely on automation.
It will be crucial whether OpenAI can sustain the more profitable consulting model alongside its mass product ChatGPT in the long term or whether the high support effort per customer will hinder scaling. Should Gartner’s prediction prove correct, this model would affect companies that, after initial AI experiences, would return to human personnel anyway. The next test will likely be the expansion into additional languages and industries, which OpenAI has not yet scheduled.
