OpenAI Presence – Enterprise AI Agent Platform for Trusted Voice and Chat Workflows

1. What Is OpenAI Presence

OpenAI Presence is OpenAI’s enterprise platform for deploying and operating governed AI agents in production. Announced on July 22, 2026, it targets high-volume, high-stakes workflows such as customer support, sales outreach, billing resolution, insurance claims, and internal IT service desks. Instead of handing teams only model access, Presence packages policies, permissions, simulations, guardrails, evaluations, and a continuous-improvement loop into a managed deployment. OpenAI says the same stack already powers its English-language phone support line and resolves about 75% of inbound issues without a human handoff.

Image source: Official article

Technical positioning: Presence sits above raw model APIs as a production operating layer for agents. It is designed for task-scoped workflows with clear boundaries—what the agent may know, which systems it may touch, which actions require approval, and when a person must take over. The confirmed launch focus is realtime voice and chat; broader channel ambitions such as email appear in marketing materials but should be treated as deployment-specific until confirmed in the delivery scope.

Development background: The product reflects years of enterprise deployment work and OpenAI’s own large-scale support operations. As foundation-model competition intensifies and switching costs fall, OpenAI is expanding from “sell model access” toward “deliver managed agent operations.” Deployments are led by OpenAI Forward Deployed Engineers (FDEs) and select global systems integrators rather than a self-serve console.

Core value: Enterprises keep control while agents answer questions, resolve issues, use company systems, take approved actions, and escalate when needed. After launch, production sessions, escalations, and quality signals feed a Codex-powered improvement process (via a Presence plugin). Proposed updates can be tested against the live version and then rolled out in a controlled way, so agents adapt as policies and customer behavior change.

Technical characteristics: Presence centers on least-privilege, job-scoped access; pre-production simulations with automated graders; runtime Guardrails; and human-approved iteration. When a use case exceeds current product support, FDEs and partners can still help bring it into production. OpenAI also continues to support customers who prefer to build voice agents directly on the OpenAI API.

2. Key Features

• Job-scoped policies and permissions: Each deployment starts with a specific job—billing issues, insurance claims, employee IT requests, and similar. The agent receives only the knowledge and system access required for that job. Companies define what the agent can do, when approval is required, and when a person should take over.

• Simulations and automated graders: Before users are exposed to an agent, teams can stress-test common requests, edge cases, and higher-risk scenarios. Graders evaluate whether the agent reached the intended outcome, followed policy, used tools correctly, and escalated when appropriate—shifting discovery left of production.

• Runtime Guardrails: Guardrails can intervene when an interaction moves outside defined boundaries, such as unauthorized actions or sensitive-data handling. Combined with approvals and human handoff rules, this creates layered protection for regulated industries.

• Codex-powered continuous improvement: After launch, Presence monitors production sessions, escalations, and quality signals. Codex, using the Presence plugin, investigates gaps and proposes updates. Teams can compare each proposed change with the production version, approve it, and execute a controlled rollout. OpenAI reports that on its own English phone line, this loop reduced human handoffs by about 15 percentage points within roughly 10 days (company-reported figure).

• Realtime voice and chat experiences: Presence natively supports realtime voice and chat agents for customer and employee workflows. Contact-center integration, authentication, routing, and human-handoff design are confirmed per deployment. Marketing pages may mention wider channels such as email; treat those as optional ambitions unless included in your statement of work.

• Approved actions across enterprise systems: Within policy limits, agents can look up accounts, update tickets, advance claim steps, or perform other approved operations by connecting to internal systems. Access is intentionally narrow, avoiding blanket credentials.

• Managed delivery with FDEs and integrators: Presence is offered through a limited general availability program. It is not self-serve. OpenAI FDEs and selected systems integrators lead workflow selection, system connection, policy setup, testing, and go-live. Early public design/partner mentions include BBVA (Mexico voice banking support), SoftBank (Japanese conversational quality testing), and IAG (high-demand event support exploration).

3. How to Use

1. Engage your OpenAI account team: Request access through your OpenAI account representative or an approved systems integrator. Eligibility depends on workflow fit, implementation readiness, and delivery capacity. There is no public self-serve signup.

2. Pick one high-value starter job and define boundaries: Work with FDEs to choose a first workflow with clear SOPs—billing resolution, claims intake, password resets, and similar. Document allowed knowledge sources, systems, actions, approval points, and mandatory human takeover conditions.

3. Configure policies, SOPs, and Guardrails: Encode operating procedures and least-privilege permissions. Start with high-frequency, rule-clear paths and require human review for irreversible or sensitive actions.

4. Run simulations and graders before launch: Validate common, edge, and high-risk cases. Check outcome quality, policy adherence, tool use, and escalation behavior. Fix gaps in knowledge or policy before production traffic.

5. Operate with monitoring and controlled rollouts: After go-live, track resolution rates, escalation rates, and guardrail triggers. Review Codex proposals, test them against the live agent, and approve controlled updates. Re-run permission and evaluation reviews when adding systems or channels.

Notes: Delivery speed depends on OpenAI/partner capacity. Prepare API inventories, identity/permission maps, and compliance requirements early. Pricing and implementation scope are customer-specific; OpenAI has not published a public price list.

4. Pros and Cons

5. Competitive Comparison

Selection guidance: Choose Presence if you want a high-touch, governed path into production and already standardize on OpenAI. Choose Copilot Studio for low-code builds tightly bound to Microsoft 365/Teams. Choose Vertex AI Agent Builder or Bedrock Agents if your platform team prefers cloud-native self-serve orchestration and can invest in its own evaluation and operations layer.

6. Editor's Take

Presence is less about a flashier model and more about productizing the hard enterprise work: permissions, evaluations, guardrails, human takeover, and post-launch improvement—delivered with FDEs in the loop. That is the right problem for companies stuck between impressive demos and brittle production agents. OpenAI’s self-reported metrics (about 75% automatic resolution on its English phone line, and roughly a 15-point handoff reduction in about 10 days via the Codex loop) are directionally compelling, but buyers should insist on reproducing those measurements on their own workflows. Recommendation: ★★★★☆ (4/5). Points off for limited GA access, opaque pricing, and channel/integration ambiguity until a statement of work is locked. If OpenAI later opens more self-serve paths and publishes broader independent benchmarks, the score can rise.

7. Use Cases

• Customer support over voice and chat: Handle billing questions, refund status, and product how-tos within policy. Escalate only when judgment or exceptions appear. Best for centers with mature SOPs and measurable SLAs.

• Insurance claims and surge events: Collect documents, check policy rules, and keep support available during weather or outage spikes. Public reporting has cited insurers such as IAG exploring this pattern.

• Banking and telecom routine service: Use cases like BBVA Mexico’s voice banking design partnership and SoftBank’s Japanese conversational quality testing show the fit for regulated, identity-sensitive interactions.

• Internal IT and HR service desks: Password resets, access requests, software provisioning, and common troubleshooting with least-privilege connections to directory and ticketing systems.

• Outbound sales follow-ups: Qualify leads, book meetings, and write CRM updates while tightly limiting what the agent may promise or change. Lets human sellers focus on high-intent conversations.

8. FAQ

Q: How do we get access to OpenAI Presence?

A: Contact your OpenAI account team or an approved systems integrator and apply for the limited general availability program. Access depends on workflow fit, readiness, and delivery capacity. Prepare a starter job definition, system inventory, compliance constraints, and success metrics such as auto-resolution rate, escalation rate, and handle time.

Q: How is Presence different from building agents on the OpenAI API?

A: The API path is flexible but leaves policy, evaluation, guardrails, monitoring, and improvement mostly to you. Presence is a managed production platform that packages SOPs/permissions, simulations/graders, Guardrails, human escalation, and a Codex improvement loop, with FDE-led delivery. OpenAI states it will continue supporting customers who prefer direct API-based voice builds; the two options are complementary rather than exclusive.

Q: Which channels are supported today—does it include email?

A: In limited GA, the confirmed emphasis is realtime voice and chat. Contact-center integration, authentication, and handoff design are scoped per deployment. Product marketing may mention broader channels such as email, but you should not assume those capabilities unless they appear in your delivery agreement.

Q: How is Presence priced?

A: OpenAI has not published a public price list. Commercial terms are tailored to deployment scope, integrations, channels, and managed services. Budget for implementation and ongoing operations in addition to any platform fees.

Q: How are security and permissions handled?

A: Agents are scoped to the knowledge and system access required for a specific job. Enterprises define which actions need approval and when humans must take over; Guardrails can intervene when interactions leave bounds. Request formal documentation on encryption, data handling, and compliance certifications during security review.

9. Project Links

• Product page: https://openai.com/business/openai-presence/
• Official announcement: https://openai.com/index/introducing-openai-presence/
• Help Center: https://help.openai.com/en/articles/20001405-openai-presence
• OpenAI Business: https://openai.com/business/