OpenAI (@OpenAI) launched OpenAI Presence on July 22nd, giving large companies a managed product for deploying voice and chat agents across customer support, sales and internal workflows.
Presence packages OpenAI models with the controls required to let an agent access corporate systems and take actions. Enterprises define the agent's permitted knowledge, system access, operating policies and escalation rules. OpenAI's pitch centers on the work surrounding the model: testing behavior, enforcing permissions, monitoring production sessions and revising the agent as policies or customer behavior change.
The delivery model is the important part. Presence is available only to eligible enterprise customers through a limited general availability program. OpenAI Forward Deployed Engineers and selected systems integrators lead each deployment, and Presence cannot be purchased or configured as a self-serve product. OpenAI did not publish pricing or the eligibility requirements for the program.
That structure moves OpenAI further into the implementation work required to put agents in production. OpenAI engineers help customers select a workflow, connect internal systems, establish permissions, test the agent and manage its launch. Selected systems integrators can continue supporting deployments as customers add workflows or channels.
Codex proposes changes, while customers retain approval
Presence starts with a defined job, such as handling billing problems, supporting insurance claims or resolving employee IT requests. An agent receives only the information and system permissions that OpenAI and the customer determine it needs for that assignment.
Before release, simulations and automated graders test whether the agent reaches the expected outcome, follows policy, uses tools correctly and hands work to a person when required. Guardrails can intervene when a conversation moves beyond the boundaries established for the deployment.
After launch, Presence analyzes production sessions, escalations and other quality signals. Codex, using a Presence plugin, investigates failures and proposes changes. Customers can test those proposals against the production version before approving a controlled rollout.
The approach gives OpenAI a recurring role in the operation of the agent, rather than ending its involvement when a customer gains model access. It also creates a feedback channel between enterprise deployments and OpenAI's research and product groups. OpenAI says generalized findings from customer work will inform future versions of Presence and its underlying technology.
OpenAI is its own first reference customer
OpenAI says Presence runs its English-language AI phone support channel, which answers calls at 1-888-GPT-0090 in the United States and Canada. The channel handles general product and account questions without requiring a ChatGPT account.
According to OpenAI, the agent resolves 75% of inbound issues without human assistance. OpenAI also says a Codex-powered improvement process reduced handoffs by 15 percentage points over 10 days. Those figures are self-reported. OpenAI did not disclose call volume, the number of sessions included in the measurement or the benchmarks used to compare the agent with frontline support staff.
OpenAI's support documentation also defines narrower boundaries than the broader Presence product supports. The phone agent cannot connect callers to a live agent, initiate an escalation or account review, submit a report or guarantee a follow-up. Callers must complete separate support workflows for those requests. That distinction shows how Presence is meant to be configured around each customer's policies rather than deployed with one universal set of actions.
OpenAI says calls to its support line may be retained, including audio recordings and transcripts, for service improvement, model training, safety and abuse prevention. The help page tells users to verify important information because models can make mistakes.
Design partners are still testing specific workflows
OpenAI named BBVA, SoftBank and IAG as early enterprise partners, though its descriptions indicate that the deployments remain exploratory or in testing.
BBVA is working with OpenAI on voice support for routine banking needs in Mexico. Daniel Ordaz, BBVA Mexico's head of AI transformation, described the bank as a Presence design partner focused on refining financial-service voice interactions.
SoftBank is testing Japanese-language customer conversations. Tadahisa Murakami, a vice president leading data and digital transformation at SoftBank, said frontline employees had rated the agent's Japanese conversations highly for naturalness and accuracy. OpenAI did not publish the evaluation method or results.
Australian insurer IAG is exploring agents that could help customers during severe weather and natural disasters, when claims and support demand can rise quickly. Julie Batch, CEO of IAG's retail insurance business in Australia, said the work is focused on claims navigation and service availability during high-demand periods.
Presence gives OpenAI a product for enterprises that want operational agents without assembling the full evaluation, policy and deployment stack themselves. The limited release also keeps OpenAI closely involved in the first deployments, where failures in permissions, escalation or tool use carry considerably greater consequences than a poor chatbot answer.