# Presence: Governed Production Agents—Limited GA, Not Self-Serve

Times of AI Desk · 2026-07-22 · Work

[https://timesof.ai/2026/07/openai-presence-enterprise-agents](https://timesof.ai/2026/07/openai-presence-enterprise-agents)

> OpenAI launched Presence for trusted enterprise voice and chat agents with policies, guardrails, simulations, and a Codex improvement loop—limited GA via Forward Deployed Engineers, not self-serve. OpenAI’s own English phone support runs on it and reportedly resolves ~75% of inbound issues without humans.

Enterprise buyers are past demos; they buy **governed production agents**. Presence is OpenAI’s explicit productization of years of forward-deployed agent work—with a public reliability claim on its own support line.

**OpenAI** introduced **OpenAI Presence**, a **deployed enterprise product** for **trusted voice and chat agents** that answer questions, resolve issues, use company systems, take **approved actions**, and **escalate to people**. Presence is available to **eligible enterprise customers** through a **limited general availability** program led by **OpenAI Forward Deployed Engineers** and select systems integrators—it is **not** self-serve. OpenAI’s own English phone support (**1-888-GPT-0090**) runs on Presence and reportedly **resolves ~75% of inbound issues** without human assistance.

## What Presence is

OpenAI frames the enterprise problem as no longer “can agents work?” but “can they work **reliably in production** as policies and behavior change?” Presence packages:

| Component | Role |
|-----------|------|
| **Job-scoped agents** | One high-value workflow (billing, claims, IT service, etc.) with only required knowledge and system access |
| **Policies & SOPs** | What the agent may do, when approval is required, when a human takes over |
| **Guardrails** | Intervention when interactions leave company boundaries |
| **Simulations & graders** | Pre-launch tests on common requests, edge cases, and higher-risk scenarios |
| **Codex improvement loop** | Production sessions and escalations feed proposed updates that teams test and approve |
| **Channels day one** | **Voice and chat** for support, outbound sales, and high-risk internal workflows |

OpenAI works with each customer to identify workflows, connect systems, set permissions, test, and bring agents into production. Insights from deployments feed research and product for other customers.

## Design partners and internal proof

Named launch-context partners exploring Presence-style trusted agents:

- **BBVA** (Mexico) — design partner for financial voice service experiences  
- **SoftBank** — Japanese-language customer conversations  
- **IAG** — support during high-demand events such as severe weather  

Internally, OpenAI says Presence powers its English-language phone support, met or exceeded frontline human-support quality benchmarks within weeks, and that a Codex-powered improvement loop **cut human handoffs by 15 percentage points in 10 days**.

## Availability and competitive context

| Item | Status |
|------|--------|
| **Access** | Limited GA for eligible enterprises; FDE + SI led |
| **Self-serve** | **Not yet** |
| **API voice** | Continues for customers using frontier models via API |

Presence sits in a crowded enterprise agent market (contact-center platforms, cloud CS agents, and lab-native stacks). OpenAI’s differentiation claim is **full-stack production control**—policy, evals, escalation, and post-deploy improvement—not only model access.

## Limits

- ~75% resolution and 15-point handoff cut are OpenAI’s internal claims.
- Limited GA only; self-serve packaging is unannounced.
- Design-partner outcomes are exploratory, not audited case studies here.

## Sources

- [OpenAI: “Introducing OpenAI Presence” (July 22, 2026)](https://openai.com/index/introducing-openai-presence/)
- [@OpenAI primary announcement (July 22, 2026)](https://x.com/OpenAI/status/2079916436232036614)
- [MLQ News: OpenAI Launches Presence (July 22, 2026)](https://mlq.ai/news/openai-launches-presence-an-enterprise-ai-agent-platform-for-voice-and-chat-workflows/)