OpenAI Presence: AI-агенты для бизнеса переходят в промышленный продакшн

OpenAI announced the launch of Presence in early August 2026 — a new enterprise product for the industrial deployment of AI agents. Presence is oriented toward external deployments: customer service, workflow automation, and scenarios where agents work directly with a business's end users. In complex cases, OpenAI's own engineers join the process.

How Presence Differs from Workspace Agents

Presence fills a niche that Workspace Agents do not cover. OpenAI's Workspace Agents are a solution for a company's internal perimeter: automating email, document management, employee tasks, and HR processes. Presence, on the other hand, targets the "front line": agents interact with a company's clients — handling inquiries, managing first-line support, and executing transactional operations on behalf of external users.

The fundamental difference lies in the operating conditions. External agents work in a significantly less predictable environment: they encounter users without specialized AI training, process atypical requests, and carry direct reputational risk for the client with every significant error.

Key characteristics of Presence according to The Decoder:

• Target scenarios — customer service and internal business processes
• Deployment type — external (not only within the company)
• Difference from Workspace Agents — focus on interaction with external users
• Support — OpenAI engineers join in complex cases

Why OpenAI Involves Its Engineers in Production

The most atypical feature of Presence is the direct involvement of OpenAI engineers in support. When an agent encounters a non-standard scenario that goes beyond standard configurations, the company's team diagnoses the problem and adapts the agent's behavior. OpenAI takes on part of the operational responsibility — an approach atypical for platform providers.

For the enterprise client, this lowers the barrier to entry. At launch, an agent almost always encounters edge cases: atypical phrasings, conflicting requests, scenarios that were not anticipated during design. Previously, all this configuration work fell exclusively on the client's internal AI team.

For OpenAI, this is a shift in positioning: from a pure model provider, the company becomes a co-executor of deployment, with a potentially higher average deal value and deep client lock-in.

"Presence is designed to bring AI agents into production," reads the official product description from

OpenAI published by The Decoder in August 2026.

Enterprise AI

Agents: The Main Battle in the Enterprise Market

Moving AI agents from pilot to production is technically more complex than it appears. An agent in industrial operation must work reliably across thousands of different scenarios, correctly handle edge cases, and meet the security requirements of a specific industry. This is precisely the stage where most enterprise AI projects stall.

By mid-2026, the enterprise AI agent market had become the main competitive arena in the enterprise segment. Salesforce promotes Agentforce, Microsoft — Copilot Studio with Microsoft 365 integration, Anthropic — Claude for Enterprise with an emphasis on safety and regulatory compliance.

OpenAI with Presence is betting on a managed service in which the model provider remains an active participant in operations. This is an unusual position: platform companies traditionally leave implementation to partners and system integrators.

What This Means

Presence signals a strategic shift for OpenAI — from selling APIs and developer tools to managed, turnkey enterprise solutions. The involvement of the company's engineers in the support loop increases the value of the offering for enterprise clients who are not ready to handle AI agent production incidents on their own.

How well this managed approach scales as the client base grows is an open question. The more clients connect to Presence, the more complex cases will require the involvement of OpenAI engineers. Watching how the company handles this load will be telling.