OpenAI Confirms Where the Market Is Heading: The Future of AI Is Deployment, Not the Model
Two weeks ago, we wrote about Ode with Anthropic — the first confirmation that the world's largest AI model maker was betting hundreds of millions of dollars on deployment, not the model. Today, OpenAI takes the next step: it has launched Presence, a product that lets large companies put AI agents to real work, with guardrails, approvals, and human escalation built in. This is no longer a thesis or a business plan — it's a live product, with production numbers and named enterprise customers. Two of the world's biggest AI model makers have reached the same conclusion in just two weeks: competitive advantage is built in deployment.
- OpenAI launched Presence — not a new company, like Anthropic's Ode, but a product built directly into the OpenAI platform, for enterprise AI agents with guardrails and human approval.
- It already runs OpenAI's own support line: it resolves 75% of calls without human help, with a 15-percentage-point drop in human handoffs in just 10 days.
- It's the second confirmation, in two weeks, that the world's biggest AI model makers are betting on deployment, not the model — from Ode (Anthropic) to Presence (OpenAI).
From thesis to product
On July 22, OpenAI officially introduced Presence: unlike Ode, which is a separate joint venture, Presence is a product built directly into the OpenAI platform. Every deployment starts with a concrete job — resolving a billing issue, an internal IT request, an insurance claim. For each agent, the client company defines:
- what information it can access
- what actions it can take
- when it needs approval
- when it escalates to a person
The work doesn't stop at launch. Production sessions and escalations reveal where the agent has gaps, and Codex — using a dedicated Presence plugin — proposes updates that the team tests and approves before rolling them into production. It's a continuous improvement loop, not a product shipped and forgotten.
Presence isn't self-serve yet. It's available through a limited general-availability program, with deployments led by OpenAI's Forward Deployed Engineers and select systems integrators — in practice, an enterprise consulting relationship, not a subscription you switch on from a dashboard.
The proof is in-house
Before selling it to others, OpenAI tested it on itself. Presence runs the company's English-language phone support line — 1-888-GPT-0090 — where it verifies caller identity, uses account context, and takes approved actions. Within weeks, it met or exceeded the benchmarks OpenAI uses to grade human support quality. Today it resolves 75% of inbound calls without any human involvement.
For a director evaluating an AI vendor, the number that matters most is the speed of improvement, not 75%: the Codex-powered optimization loop cut human handoffs by 15 percentage points in just 10 days working with the launch team. That's not a static result — it's a system that corrects itself quickly as it sees real cases.
Three big customers, one pattern
Beyond its own support line, OpenAI has named three enterprise customers already testing Presence — worth stating precisely: they're testing partners, not full deployments at scale yet.
BBVA is exploring AI-powered voice support for everyday banking in Mexico. Daniel Ordaz, Head of AI Transformation at BBVA Mexico, describes the bank as a "design partner" for Presence. SoftBank Corp. is testing natural Japanese-language conversations — its frontline teams rated them highly for being "natural and accurate," according to Tadahisa Murakami, a VP at SoftBank. IAG is exploring how Presence could support customers during high-demand periods, such as severe weather — says Julie Batch, CEO of Retail Insurance Australia, part of the group.
Three different industries — banking, telecom, insurance — three different continents, the same move. The fact that OpenAI is testing Presence across all three at once says more about the product's direction than the size of any single pilot.
What this means for your company
As with Ode, a direct contract with Presence isn't a realistic option for a company in Romania — the entry threshold is built for American enterprise clients, with budgets to match. But the pattern matters for any AI conversation inside your company: two of the world's biggest model makers reached the same conclusion two weeks apart — the advantage doesn't come from which model you pick, it comes from the work of connecting it to your real processes, with policies, guardrails, and a person who takes over when it matters.
It's also a signal of caution for specialized integrators like MassAI: competitive pressure is rising exactly where we operate. When the world's biggest model makers decide to sell deployment directly, not just model access, differentiation no longer comes from wiring an API into a CRM. It comes from how well you know your client's business and how fast you fix what isn't working. Here's what an operational AI agent actually looks like in practice — a system that resolves real requests, with human control where it matters, not just a chatbot that answers questions.
The difference, in short: an assistant that answers generic questions is a model. An agent that verifies a customer's identity, applies your company's policy, and resolves the request end to end — escalating to a person when needed — is deployment.
If Anthropic confirmed this direction through a new company, and OpenAI through a new product, it becomes harder and harder to see deployment as a secondary step. For the industry's leaders, it's starting to become the main product.
MassAI editorial policy: our analyses combine primary sources and independent publications. When we cite estimates or commercial reports, we explicitly state the source and context.
Sources: ↗ Official OpenAI Presence announcement · ↗ VentureBeat · ↗ The Register · ↗ The New Stack
See what MassAI agents can do →