OpenAI has introduced Presence, a platform that lets enterprises deploy dependable AI agents. These agents can answer questions, resolve issues, work inside corporate systems, take approved actions and hand off to a human when needed. They are not chatbots but full assistants built to sit deep inside a workflow.
What happened
OpenAI Presence is a finished product that has already been battle-tested in production. It pairs model reasoning with policies, guardrails and escalation rules that keep quality and throughput in line. Presence already runs OpenAI's English-language phone support channel at 1-888-GPT-0090, where it handles open-ended requests, verifies callers, uses account context and takes approved actions. Within a few weeks the system matched or beat the quality metrics used to assess front-line human agents, and it now resolves 75% of incoming issues without a person involved. Thanks to a Codex-driven improvement loop, handoffs to humans dropped by 15 percentage points in just 10 days.
Presence bundles everything you need to put agents into production: policies and standard operating procedures, guardrails, approved actions, simulations, evaluation tooling and a Codex-based improvement process. Once live, production sessions and escalations expose gaps, and Codex proposes updates that teams can test and approve, helping the agent keep up with shifting customer behaviour. For now OpenAI Presence is available to select enterprise customers under a limited general availability program, with deployment handled by OpenAI engineers and systems integrators. It is not a self-serve product yet.
Why it matters
Presence marks a meaningful step in the evolution of AI agents for business. It is not a capability demo but a complete solution that has already proven itself in the field. AI agents that can process complex requests, integrate with corporate systems and make decisions on their own take real load off human staff and lift the efficiency of business processes. It reshapes how companies think about using AI for customer interaction and internal automation. In effect OpenAI is moving from shipping base models to shipping ready-made, integrated solutions, which lowers the bar for companies without deep in-house agent-building expertise.
What it means in practice
For founders and developers building AI products, Presence opens new ground and sets new tasks:
- Focus on integration and orchestration: where the emphasis used to be on building individual models, what matters now is integrating off-the-shelf agent solutions into existing infrastructure and orchestrating how they run.
- Writing policies and rules: the success of Presence agents rests heavily on clearly defined policies, guardrails and escalation rules. Prompt engineers and system architects will need to spend more attention not just on the model but on its "behaviour" inside business logic.
- Monitoring and improvement: the Codex-based improvement loop underlines how much continuous performance monitoring and iterative refinement matter. That calls for tooling to collect feedback, analyse data and ship changes quickly.
- More room for startups: Presence is not self-serve yet, but its arrival sets a precedent and shows the direction of travel. Startups can focus on niche agent solutions built on experience and architecture like Presence, or build tooling to deploy and manage them.
- Higher security and ethics bar: deploying AI agents that make decisions and take actions demands sharper attention to data security, privacy and ethical behaviour. Developers will have to account for that at every stage of the product lifecycle.