OpenAI's New Agent Platform Presence Is Already Resolving 75% of Support Calls On Its Own!

Hi, it's Shiichan! Today's story is about enterprise AI agents, and it kind of surprised me: an AI agent is resolving 75% of support calls without any human help.

OpenAI News openai.com

What was announced?

From OpenAI's News: OpenAI introduced Presence, an enterprise product for deploying trusted AI agents in production. Presence lets agents answer questions, resolve issues, use company systems, take approved actions, and escalate to a person when needed. It pairs model reasoning with policies, guardrails, and escalation rules that check for accuracy and performance.

Each deployment starts with one specific job — think resolving billing issues, supporting insurance claims, or handling employee IT requests. The agent only gets the knowledge and system access needed for that job, while the company decides the policy: what the agent can do, when it needs approval, and when a person should take over. After launch, production sessions and escalations surface gaps, and Codex proposes updates that teams can test and approve.

Why it matters

For enterprises, the challenge isn't proving an AI agent can work anymore — it's making it reliable enough for high-value production work. Agent behavior also has to keep adapting as products, policies, and user behavior change. That takes more than a model: it needs the systems, evaluations, and deployment know-how to keep improving the agent as conditions shift, and that's exactly what Presence is built to provide.

What changes

Presence is available today for real-time voice and chat experiences — customer support, outbound sales, and high-risk internal workflows. For a billing issue, for example, the agent understands the request, verifies the customer, looks up account information, applies company policy, and takes the approved action.

Companies choose what stays consistent across deployments (policies, evaluations, escalation rules) versus what changes per workflow or channel, so teams can build on what already works instead of starting from scratch for every new use case.

The results so far are striking: Presence powers OpenAI's own English-language phone support line (1-888-GPT-0090). Within weeks, it met or exceeded the benchmarks used to grade human frontline support quality, and now resolves 75% of inbound issues without human help. Its Codex-powered improvement loop cut human handoffs by 15 percentage points in just 10 days.

Other enterprises are piloting it too:

• BBVA is exploring AI-powered voice support for everyday banking in Mexico
• SoftBank is testing natural Japanese-language customer conversations
• IAG is exploring support during high-demand events like severe weather

Dive Deep

Presence is built from policies and standard operating procedures, guardrails, approved actions, simulations, evaluation tools, and a Codex-powered improvement process.

Before launch, teams test against common requests, edge cases, and higher-risk scenarios. Simulations and graders check whether the agent reached the right outcome, followed policy, used tools correctly, and escalated when it should have. Guardrails step in if an interaction drifts outside the company's boundaries.

That work doesn't stop at launch. Production sessions, escalations, and quality signals show teams where the agent is doing well and where it needs attention as policies, products, and user behavior evolve. A Codex plugin for Presence investigates those signals and suggests updates, and teams can test each proposed change against the live version before approving a controlled rollout. For use cases beyond what the product supports today, OpenAI's Forward Deployed Engineers and partners can help bring it into production.

Right now, Presence is available through a limited general availability program for qualified enterprise customers — it's not something you can just sign up and self-serve yet.

Wrap-up

• OpenAI introduced Presence, an enterprise AI agent platform for answering questions, resolving issues, using company systems, taking approved actions, and escalating to people
• Real-time voice and chat support is available today, for customer support, sales, and internal workflows
• OpenAI's own phone support line matched human-quality benchmarks within weeks and now resolves 75% of issues without help; a Codex-driven loop cut handoffs by 15 points in 10 days
• BBVA, SoftBank, and IAG are piloting it
• The design leans on trust: pre-launch simulations plus a Codex-driven post-launch improvement loop
• For now it's limited to qualified enterprise customers through a GA program, so individual developers can't try it directly yet