OpenAI Presence: Enterprise AI with Expert Engineers

OpenAI Presence: A New Approach to Enterprise AI

On July 22, OpenAI announced the launch of Presence, an innovative offering for businesses looking to integrate AI agents. Unlike traditional products available online, Presence is a managed service, accessible through a limited general availability program. OpenAI has opted for a model where deployments are supervised by carefully selected Forward Deployed Engineers and system integrators around the world.

This launch marks a turning point for OpenAI, which had previously focused on selling API keys and usage licenses. With Presence, the approach is different: each project is treated as a unique mission, whether it involves resolving a billing dispute or managing an IT service request. The agents are equipped only with the knowledge and access necessary to complete their tasks, and clients define the rules for intervention, validation, and human handoff. After the initial deployment, Codex analyzes the sessions to suggest improvements, which are then tested and approved by the client's team before being implemented.

OpenAI's documentation is particularly candid about the work involved. Its help center outlines a six-step process ranging from defining business outcomes to reviewing security, privacy, and legal aspects, through simulation and acceptance testing, phased deployment, and post-launch iteration. A Presence agent, it states, does not become production-ready simply by ingesting documents.

A Model to Address AI Project Failures

OpenAI's decision to offer a managed model addresses real market needs. According to Gartner, over 40% of agent-based AI projects are at risk of being abandoned by 2027, primarily due to insufficient governance and poor operational management. Presence aims to remedy these issues by integrating simulations and rigorous assessments to ensure that agents achieve the desired outcomes, adhere to policies, and use their tools correctly before any external interaction occurs.

Practically everything that Presence encompasses directly targets this diagnosis. Simulations and evaluators test whether an agent has achieved the right outcome, followed policy, used its tools correctly, and escalated when necessary, before anyone outside the company interacts with it. Safeguards come into play when interactions exceed defined limits. Session recordings and action histories provide auditors with elements to review. Escalation paths give a person structured context rather than a mere cold transcript, and new versions are deployed through controlled rollouts with the option to revert.

Companies have spent two years discovering that the difficult part of a production agent lies in integration, permissions, and change management. A provider that sends engineers to perform this work addresses what buyers have actually failed to accomplish, rather than delivering another dashboard and qualifying the lack of a solution as a customer problem.

Access Constraints to Presence

Access to Presence is not open to everyone. OpenAI imposes strict eligibility criteria based on workflow fit, readiness for implementation, and delivery capability. The latter is particularly crucial, as it depends on the engineers available to integrate clients' core systems.

Delivery capability is a consulting constraint. The software is evolving; engineers who have access to a bank's core systems may not be available. The title of Forward Deployed Engineer is borrowed from Palantir, where it describes personnel embedded in client operations for months, and the associated economy is nothing like measured inference. By placing its own FDEs and named partners at the forefront of each deployment, OpenAI has penetrated the market segment occupied by integrators it will also rely on for growth, which is a viable arrangement as long as volumes remain low, but becomes more complicated thereafter.

This also raises a question for anyone considering a contract. When the model provider is also the implementation partner, the lines of responsibility for poorly applied policy in production must be written rather than assumed.

AI Agents Still in Testing Phase

Although OpenAI presents Presence as a proven product, it is important to note that this term refers to accumulated experience rather than a long-standing market presence. OpenAI's helpline, 1-888-GPT-0090, serves as proof of concept, with impressive results: 75% of issues resolved without human intervention and a 15 percentage point reduction in human handoffs within ten days.

Companies like BBVA, SoftBank, and IAG are currently testing Presence. BBVA is exploring voice support for banking in Mexico, SoftBank is testing interactions in Japanese, and IAG is focusing on support during high-demand events. Daniel Ordaz, head of AI transformation at BBVA Mexico, describes the bank as a design partner helping to shape and refine voice experiences for financial customer service. Design partners are normal and helpful during a limited general availability. However, none of the three are presented as using Presence at scale, which is worth noting alongside the term proven.

Still Unclear Details

OpenAI has not yet disclosed the pricing for Presence, leaving companies without a benchmark to evaluate costs against other contact center solutions. Furthermore, the model used by Presence is not specifically named, although the documentation indicates that it involves OpenAI's models, adjusted according to workflow needs.

Channel support during this limited availability phase includes voice and chat, with specific integrations confirmed for each deployment. Data management follows a similar pattern, with contracts serving as reference documents.

Presence stands apart from ChatGPT Workspace Agents, which remain self-service options for internal teams using ChatGPT and Slack. OpenAI thus offers three similar solutions, differentiated primarily by who performs the work, leaving companies to choose based on delivery capability as much as on the technology itself.