Share with your CIO
OpenAI is betting that the reason enterprise AI agents fail isn’t the model, it’s the deployment. OpenAI Presence is a managed agentic product, not a self-serve purchase, delivered through the company’s own Forward Deployed Engineers and a set of named global systems integrators. Each engagement starts with a single workflow, billing disputes, insurance claims, IT service requests, and is governed by a six-stage process covering security review, simulation, staged rollout, and continuous iteration via Codex. OpenAI’s own phone support line, running at 75% autonomous resolution, is the headline proof point.
What this means for your business
The dividing line this announcement draws isn’t between companies that can afford AI and those that can’t. It’s between organizations that have already discovered, expensively, that a production agent requires integration work, permissions architecture, and change management that no dashboard ships with, and those still about to find that out. If your team has a graveyard of stalled pilots, Presence is designed specifically for where those pilots died. If you haven’t started, the six-stage process OpenAI documents is a useful map of what any serious deployment actually costs, whether you buy Presence or not.
The Forward Deployed Engineer model, borrowed directly from Palantir, where engineers embed inside customer operations for months at a time, carries a structural tension OpenAI hasn’t resolved. Software scales; credentialed humans cleared into a bank’s core systems don’t. OpenAI is simultaneously the model vendor, the implementation partner, and the entity setting the grading criteria for whether the agent is performing. That’s a workable arrangement at low volume and a procurement problem at scale. The contract you sign needs to specify what happens when a policy is misapplied in production and who owns the remediation, because the current framing leaves accountability blurred by design.
The three named customers, BBVA Mexico in voice banking exploration, SoftBank in Japanese-language testing, IAG in severe-weather support, are design partners, not scaled deployments. OpenAI’s own figures are self-measured against self-defined benchmarks on a self-owned channel. None of that makes the product wrong, but any vendor who controls the outcome metric, the implementation, and the model configuration is holding three variables you’d normally want separated. The renewal decision to watch isn’t whether Presence works in pilot. It’s whether the pricing, which isn’t published, looks defensible once you can compare cost-per-resolved-contact against your existing contact-center contract.
Based on reporting from OpenAI Presence: enterprise AI agents, engineers included, originally published 2026-07-24 04:06:00.
