OpenAI Presence: The Enterprise AI Agent Platform That Resolves 75% of Calls Without Humans

Core Features

Presence bundles everything enterprises need to run production agents under strict governance :

• Policies and SOPs: Companies define what the agent can do, when it needs approval, and when a person must take over.
• Guardrails: Block risky actions — for example, limiting which databases an agent can query or regulating how it interacts with customers .
• Approved actions: Only pre-scoped actions (retrieve info, update systems, complete transactions) are permitted.
• Simulations and evaluation tools: Before launch, teams can test agents against common requests, edge cases, and high-risk scenarios. Automated graders check whether the agent reached the right outcome, followed policy, used tools correctly, and escalated appropriately .
• Codex-powered improvement loop: After launch, production sessions and escalations reveal gaps. Codex proposes updates (e.g., harness changes) that teams can test and approve, creating a continuous "measure → improve → ship" cycle . In OpenAI's own deployment, this loop reduced human handoffs by 15 percentage points in just 10 days .
• Human escalation paths: Structured context handoff when a policy, risk, or workflow requires human judgment .
• Controlled rollout, monitoring, and rollback: New versions are staged and can be rolled back based on production evidence .

Presence currently supports voice and chat channels, including real-time conversational experiences for customer support, outbound sales, and internal workflows .

Results on OpenAI's Own Phone Support

OpenAI uses Presence to power its English-language phone support line at 1-888-GPT-0090. The agent handles open-ended requests, verifies callers, uses account context, and takes approved actions. Within weeks, it met or exceeded the benchmarks used to grade frontline human-support quality. The agent now resolves 75% of inbound issues without human assistance .

Early Enterprise Customers

All three are design partners or early evaluators, not yet in broad production .

Availability and Pricing

Presence is in limited general availability (limited GA) as a managed deployment . It is not a self-serve product. Key details:

• Access depends on workflow fit, implementation readiness, and available delivery capacity .
• OpenAI's Forward Deployed Engineers and select systems integrators work directly with each customer through a managed process: scoping, integration, security review, simulations, staging, and monitored rollout .
• Pricing is per-customer and per-deployment, not published. The Register described it as "boots-on-the-ground prices" reflecting the consulting-like delivery model .
• It is distinct from ChatGPT Workspace Agents — Presence agents are scoped and deployed with OpenAI's team, not created through a workspace interface .

How It Fits OpenAI's Broader Enterprise Strategy

Presence marks a clear strategic shift from selling models and API tokens to selling deployed, managed AI infrastructure . Several signals confirm this:

• From model provider to service integrator. Presence bundles policies, guardrails, evals, and a Codex improvement loop — replicating the full stack that contact-center middleware vendors charge for . OpenAI is effectively competing with systems integrators and CCaaS platforms, not just other model providers.
• Hands-on delivery model. OpenAI assigns Forward Deployed Engineers to each Presence customer, and the product is delivered via a managed engagement rather than a self-serve dashboard . The Register characterized this as OpenAI "trying the consulting path" .
• Tied to the OpenAI Deployment Company. In May 2026, OpenAI launched (with 18 founding partners including BBVA) the OpenAI Deployment Company ("DeployCo"), a $4B+ vehicle to provide enterprise deployment services . Presence is the product expression of that strategy.
• Compounding cross-customer learning. OpenAI states that "generalized insights from every deployment inform ongoing research and product development, improving the product for all customers over time" . This turns each deployment into a data and improvement moat.
• Escalating enterprise commitments. The same week, BBVA expanded ChatGPT Enterprise to all 120,000 employees , and the broader enterprise wing now includes Frontier (cross-system agents) and AgentKit (developer tools) — with Presence as the highest-touch, most governance-focused tier.

In short, Presence is OpenAI's bet that enterprises will pay a premium not for a smarter model, but for a reliable, governable, continuously improving agent system that comes with OpenAI's engineers and institutional learning built in .