OpenAI built support agents for its own customer service line, now it hopes big enterprises will trust them too
The New Stack Paul Sawers ● Covered by 4 sources
OpenAI is selling the support bot that runs its own customer line. It already handles 75% of those calls alone — now banks and airlines want in.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
For a while now, the AI industry has quietly agreed on something: the models themselves aren't really the problem anymore. GPT-class systems can hold a conversation, write decent code, and untangle a customer support ticket without much trouble. What's missing is everything around the model — the rules about what it's allowed to do on its own, what happens when it hits a wall, and a paper trail for when something breaks. OpenAI's answer to that gap, announced Wednesday, is a product called Presence, and it's built on the same agents that already answer OpenAI's own support phone line.
Presence isn't just an API key and a prayer. OpenAI builds and deploys the agent for each customer, then wraps it in the testing tools, monitoring, and policy scaffolding needed to keep it from going sideways once it's live. The agent itself can verify who's calling, pull up an account or order, and actually resolve something — a duplicate charge, a missing delivery — without routing to a human. Before any policy change ships, teams run it through simulation against a batch of past cases, scoring how it performs on refunds, cancellations, and account verification. Once it's live, a dashboard tracks accuracy and volume in real time.
Crucially, the customer sets the boundaries, not OpenAI. A company picks one narrow job — an insurance claim, an IT ticket, a billing dispute — and the agent only touches the systems tied to that job. The business decides what it can do unsupervised, what needs sign-off, and where it has to stop and hand off to a person. Getting there isn't a self-serve toggle: OpenAI's own engineers sit with each customer to map the job, wire up systems, set permissions, and test everything before handing ongoing support to outside integration partners.
OpenAI's pitch leans hard on its own numbers — the same agents now resolve 75% of inbound issues on its English-language phone line without a human stepping in, and the implication is obvious: if it's trustworthy enough for a company whose entire business depends on people believing what it builds, it should be trustworthy enough for anyone. Early partners include BBVA, SoftBank, and IAG, though this is still a restricted rollout run by OpenAI's forward deployed engineers and a handful of systems integrators, not something you can spin up on your own.
That staffing model is itself becoming an industry story. Forward deployed engineering went from niche job title to hot commodity in about ten days back in May, with OpenAI spinning up a $4 billion company around it, Google posting roles with mid-six-figure salaries, and AWS committing $1 billion to its own embedded-engineer push in June. Presence also arrives a month after OpenAI joined Google and Microsoft in founding the Appia Foundation, an effort to standardize how AI systems prove they're safe and compliant rather than relying on vendors' word for it. Appia is the industry writing shared rules; Presence is OpenAI trying to earn trust the slower way, one enterprise phone line at a time.
My take — AI-written commentary, not fact-checked reporting
Nothing says 'trust us' quite like a vendor grading its own homework — OpenAI's headline stat here is that OpenAI's own support agents resolve 75% of OpenAI's own tickets, which is a nice number if you don't ask who's counting. I'm all for agents handling refunds and duplicate charges, but bundling the model, the policy engine, the monitoring, and the engineers who set it all up into one proprietary package is exactly the kind of lock-in enterprises should be wary of, especially in Europe where regulators are already twitchy about black-box accountability. The forward-deployed-engineer trend is the tell: if you need a small army of specialists physically embedded to make your agent trustworthy, you haven't really solved trust, you've just made it very expensive to walk away.”}}
Read more about this at: The New Stack