But the firm just released Presence, which is not yet another chatbot prototype. This is a product developed by OpenAI based on years of successful enterprise deployments, and designed to create agents for companies to solve all kinds of problems from billing issues to insurance claims, to IT tickets.
Also read: Meta Ray-Ban has a new rival, and it’s backed by Google and Samsung
What it actually does is a well-proven product to help enterprises deploy AI agents, which can answer questions, solve problems, use enterprise systems, perform allowed actions, and escalate to humans when necessary. Each deployment of the product covers one single task. The agent is provided with knowledge and access to systems needed for the completion of the task, and the company decides on the policies regarding what actions the agent can perform, what actions need approval, and when escalation to humans is needed.
What makes this interesting is not the agent itself but what comes next. Issues discovered during production runs and escalations become known, and Codex produces suggestions for updating the agent, which teams can test and implement to make the agent more effective at adapting to changing customer behavior. Put another way, OpenAI’s coding model looks at where the support agent fails and codes a fix for that, which is then signed off by a human. This is a truly novel approach, in which a coding agent does continuous operational improvement on a distinct customer-facing agent instead of a static chatbot that requires monthly interventions from a human engineer to train it again.
Also read: Samsung Galaxy Z Fold 8 vs Galaxy Z Fold 7: India price, specs and features compared
OpenAI is currently implementing this on itself. Presence runs the company’s English-language phone support channel and handles open requests, verifies identity, considers account context, and takes actions based on customer approval, and according to the company’s information, OpenAI now solves 75 percent of incoming problems without human intervention, having reduced human handoffs by 15 percentage points in only 10 days through this Codex-driven improvement loop. Note that these are OpenAI’s figures for their own support service, not any independently validated numbers; take them for what they are.
BBVA in Mexico, who is experimenting with voice support through artificial intelligence in its regular operations, SoftBank, who is running pilot projects for Japanese language-based natural conversation with their clients, and IAG, who is trying out ways to provide customer support during high traffic periods, such as bad weather, were some early customers listed.
What sets the Presence program apart from other artificial intelligence product launches is that you cannot subscribe to it. Presence is currently offered to eligible enterprise customers through limited general availability programs, in which deployments will be done by Forward Deployed Engineers from OpenAI and selected systems integrators and is clearly not available in a self-serve manner. This is an important development. OpenAI does not offer subscriptions to its APIs anymore, and it takes the same approach as consulting firms such as Accenture and Deloitte, at least for the class of applications in question.
It all comes down to trust and control: pre-launch guardrails, simulations for edge cases, and post-launch feedback loops that involve humans in approvals. Whether this approach will work under heavy production load, particularly in heavily regulated areas like finance and insurance, is what to watch out for.
Also read: RBI’s draft AI model ignores agentic AI risk, experts warn