OpenAI Presence sells enterprise AI agents with engineers attached

Details

OpenAI Presence is not available as a self-serve purchase. Deployments are led by OpenAI Forward Deployed Engineers alongside selected global systems integrators, making each engagement project-based rather than product-based. Work starts with a single narrow job — billing disputes, insurance claims, employee IT requests — with the agent restricted to only the permissions and knowledge that task requires. The customer defines escalation rules and approval thresholds; Codex then monitors live sessions post-launch and proposes iterative changes that the customer's team must test before rollout. The design is a direct response to a documented enterprise failure mode. Gartner has warned that over 40% of agentic AI projects will be cancelled by end of 2027, citing governance gaps and weak operational discipline rather than model capability. Presence addresses these through pre-launch simulations and graders, session audit logs, structured escalation context for human reviewers, and controlled rollout with rollback. OpenAI's own documentation lays out a six-stage process from scoping to post-launch iteration — an unusually candid admission of the labour required to make agents production-ready. The constraint is access: Presence is in limited general availability with no public pricing, no self-serve path, and case-by-case model and access decisions. The trade-off is deliberate — premium, engineer-led deployment over broad market reach. For enterprises that have already failed at DIY agentic integration, this model directly targets what they've been getting wrong: integration, permissions, and change management, not the model itself.