What Is OpenAI Presence? OpenAI's Managed Enterprise Agents, Explained

OpenAI Presence is OpenAI's enterprise platform for deploying voice and chat AI agents that handle customer and internal workflows: answering questions, verifying callers, looking things up in company systems, taking approved actions, and escalating to a person when needed. It was announced on July 22, 2026 and is available to eligible enterprises through a limited general availability programme. The defining feature is how it ships. OpenAI's own Forward Deployed Engineers, or a select group of global systems integrators, build and roll out each deployment. There is no self-serve version, and you get in by contacting your OpenAI account team.

This post explains what Presence is, how a deployment works in practice, what OpenAI has and hasn't proven, and what the launch says about how enterprise AI agents are being sold now. Everything about Presence here comes from OpenAI's announcement and launch-week reporting from VentureBeat, AI News, and No Jitter, with links so you can check the sources. The product is weeks old and its terms will change.

Key takeaways

• Presence is a managed deployment, not a product you log into. OpenAI engineers or a partner integrator scope one workflow, build the agent, test it in simulation, and run a staged rollout.
• It targets real-time voice and chat for customer support, outbound sales, and internal helpdesks. Voice is the headline: OpenAI has been running it on its own phone support line.
• OpenAI's proof point is that line, which it says resolves 75% of inbound issues without human help. The number is OpenAI's own and unverified.
• Three design partners are named (BBVA, SoftBank, IAG). All are described as exploring, not running at scale.
• Pricing is undisclosed and negotiated per deployment. Access depends on workflow fit, implementation readiness, and OpenAI's delivery capacity.
• The launch confirms a wider pattern: for high-stakes enterprise agents, vendors are selling engineers alongside software. The open question for buyers is how much of their agent work needs that treatment.

What does OpenAI Presence actually do?

OpenAI's own description is "a battle-tested product that helps enterprises deploy trusted AI agents that can answer questions, resolve issues, use company systems, take approved actions, and escalate to people when needed." Strip the adjectives and the shape is a conversational agent with system access and a defined set of things it is allowed to do.

A typical Presence interaction, as OpenAI frames it, runs like this. A caller or chat user makes a request. The agent works out what they need, verifies who they are, looks up the relevant records in the company's systems, applies the company's policy, takes an approved action (issue a refund, reschedule, update a record), and hands off to a human when the case falls outside its remit. The channels at launch are real-time voice and chat. Contact-centre integration, according to AI News's reading of the deployment documentation, is handled case by case during limited availability.

The product is made up of several parts, which OpenAI lists as: policies and standard operating procedures that define how the agent behaves; guardrails and an explicit set of approved actions; simulation and evaluation tools for testing before launch; production monitoring with escalation analysis; and a Codex-powered improvement process.

That last piece is the most distinctive thing in the announcement. Production sessions, escalations, and quality signals accumulate as policies, products, and user behaviour change. Codex, using a Presence plugin, investigates those signals and proposes updates, which the customer's team reviews and tests before rollout. There is a designed loop for finding where the agent is failing and fixing it, with a human approving each change.

How does a Presence deployment work?

This is where Presence differs most from anything you can sign up for. Per OpenAI: "Deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators. Presence is not yet available as a self-serve product."

A Forward Deployed Engineer (FDE) is a vendor engineer embedded with the customer, building and tuning the product against that customer's systems, data, and rules. The model was popularised by Palantir, and VentureBeat's launch coverage drew the comparison directly. With Presence, that team is the delivery mechanism rather than an optional add-on.

AI News, which reviewed the deployment documentation, describes the process in more detail. Each engagement starts with a single workflow: billing disputes, insurance claims, IT requests. The agent receives only the knowledge and system access that workflow needs. The customer defines guardrails, escalation rules, and human handoff points. The rollout follows six stages: scoping the outcome, a security, privacy, and legal review, simulation, acceptance testing, staged rollout, and post-launch iteration. The documentation is explicit that "a Presence agent does not become production-ready simply by ingesting documents."

Access is gated on three things: "workflow fit, implementation readiness and available delivery capacity." The third is the one to notice. Engineers embedded in enterprise systems do not scale the way inference does, so the number of Presence deployments OpenAI can run at once is bounded by headcount and partner capacity, not demand.

Who is using it so far?

OpenAI names three design partners.

BBVA is exploring voice support for banking customers in Mexico, with a focus on personalisation. SoftBank Corp. is testing natural Japanese-language conversations. IAG, the Australian insurer, wants reliable support during high-demand periods such as severe weather events, when call volumes spike and human capacity runs out.

All three are described as exploring or testing. None is presented as running Presence at production scale, and there are no customer metrics in the announcement.

The scaled deployment OpenAI points to is its own. OpenAI's English-language support number, 1-888-GPT-0090, runs on Presence. OpenAI says it "met or exceeded benchmarks" for human support quality within weeks, now "resolves 75% of inbound issues without human assistance," and that the improvement loop "reduced human handoffs by 15 percentage points in just 10 days."

Those are strong numbers, and they deserve two caveats. They are OpenAI's internally measured results on OpenAI's own channel, where the vendor controls the product, the policies, and the definition of "resolved." And OpenAI's support traffic (account questions, billing, API access) is a comparatively well-structured domain. Whether a bank or an insurer sees similar figures is exactly what the design partnerships are meant to find out.

How much does OpenAI Presence cost?

OpenAI has not published pricing. Its statement is that "broader pricing and availability details will follow as the limited GA program expands." VentureBeat reported that OpenAI had not responded to two pricing enquiries by publication.

Given the delivery model, expect a services-shaped contract rather than a per-seat or per-token subscription: scoping and engineering time, plus usage. Every deployment is individually scoped, so there is no reference price to benchmark against yet.

What models and integrations does it use?

Presence runs on OpenAI models. Which models, and which versions, is not stated. AI News notes that the documentation describes the model configuration as "subject to change as that workflow evolves," which matters for any team that has built evaluation suites against a pinned model version. VentureBeat reports that customers can connect third-party models and services through APIs for guardrails, tools, and other parts of the workflow, so the core agent is OpenAI's but the surrounding pieces need not be.

Integration with company systems (CRM, billing, ticketing, telephony) is part of what the FDEs and integrators build during deployment. There is no published connector catalogue, because connections are made per engagement.

What's still unknown

A few gaps are worth listing plainly, because they shape whether Presence is even an option for a given company.

Pricing and contract terms are undisclosed. Geographic availability is unstated beyond the three partner countries (Mexico, Japan, Australia) and OpenAI's own US line. Data handling is governed by each deployment's architecture and contract rather than a published policy. Channel coverage during limited GA is voice or chat, with contact-centre integration case by case. Model versions are not pinned. Delivery capacity is finite, and OpenAI now competes for enterprise agent work with some of the same integrators it relies on to scale, which No Jitter and AI News both flag as a tension to watch.

None of this is unusual for a product five weeks into limited availability. It does mean that "can we use Presence" is, for most companies today, a question for an OpenAI account team rather than something you can answer from the website.

What Presence tells you about enterprise AI agents

The more useful reading of the launch is what it says about how enterprise agents are being sold.

OpenAI already offered the same underlying capability two other ways: direct API access for teams that build their own agents, and ChatGPT workspace agents for self-serve use. AI News put it well: OpenAI now offers "broadly the same capability three ways, separated less by what the technology can do than by who does the work." Presence is the version where OpenAI does the work.

That is a direct response to how agent projects have actually gone. Gartner predicted in mid-2025 that more than 40% of agentic AI projects would be cancelled by the end of 2027, mostly through unclear value, weak risk controls, and escalating costs rather than model limitations. The failures were in integration, permissions, testing, and change management. Presence puts engineers on precisely those problems and charges for it.

For a customer-facing voice channel at contact-centre scale, that is a defensible trade. The stakes are high, the work is concentrated in one workflow, and the cost of an agent that misapplies policy to a caller is real. Paying the model vendor to own the outcome makes sense.

The question a buyer should ask next is how much of their agent work looks like that. In most companies, the answer is: one or two workflows. The rest is spread across the business. The RevOps lead who needs 3,000 accounts scored by Thursday. The associate who needs 200 documents screened against the same rubric. The ops manager who wants every inbound lead enriched and routed before a human sees it. The finance team matching invoices. The analyst who wants a monitoring brief every morning. Nobody can scope, simulate, and stage-roll-out each of those with an engineer attached. The economics do not work, and the delivery capacity does not exist.

That work needs a platform the process owners can build on themselves. This is where we should be open about our own interest. QX Labs is built for that second category: you brief agents in plain English, connect them to your stack through 1,000+ integrations, run row-scale work in parallel with Grids, put repeatable processes on triggers and schedules with Flows, and ground everything in your documents with Knowledge Vaults. The owner of a process builds the automation, sees the estimated cost before running it, and edits it the day the policy changes.

The two models are not mutually exclusive, and the deployment style is not really the dividing line. Larger QX customers with bespoke or complex requirements get forward-deployed support from our team too, alongside the dedicated account management, custom integrations, and single-tenant options on the Enterprise plan. The difference is what sits underneath. Presence scopes one trusted agent for one channel and puts engineers around it. A platform gives a team the primitives to run dozens of processes, with engineers available for the ones that warrant them. Which mix a company needs depends on how concentrated its agent work is, and most companies find it is far less concentrated than the contact centre.

When is Presence the right call?

Presence belongs on your shortlist if you run a high-volume customer channel, especially voice, where a single workflow (claims, billing disputes, account servicing) carries most of the value and a failure would be public. It also fits organisations that have no one internally to own an agent in month six and would rather pay the vendor to own it, and regulated enterprises that want the model vendor as the accountable counterparty.

It is the wrong tool if your agent work is spread across many small processes, if you need to know the price before starting a procurement, if you want to pin or choose models, or if you want to build and change automations without waiting on delivery capacity. And if your problem is a single deterministic integration, a rule-based tool is cheaper than either approach.

FAQ

Is OpenAI Presence available now?

Partly. Presence entered limited general availability on July 22, 2026 for eligible enterprise customers. Deployments are led by OpenAI Forward Deployed Engineers or select global systems integrators, and access depends on workflow fit, implementation readiness, and available delivery capacity. To explore it, OpenAI says to contact your OpenAI account team. There is no self-serve sign-up.

Can I use OpenAI Presence without OpenAI's engineers?

Not at launch. OpenAI states Presence "is not yet available as a self-serve product." Every deployment is scoped and built with OpenAI's Forward Deployed Engineers or a partner integrator, following a staged process of scoping, security review, simulation, acceptance testing, and rollout. OpenAI has not said when, or whether, a self-serve version will follow.

What is the difference between OpenAI Presence and the OpenAI API?

The API gives your developers models and tools to build agents yourself; you own the integration, testing, and operations. Presence packages the agent, guardrails, simulations, monitoring, and improvement loop as a deployed product, with OpenAI or an integrator doing the build. Same underlying capability, different answer to who does the work.

What is a Forward Deployed Engineer?

A Forward Deployed Engineer is a vendor engineer embedded with a customer to build and tune the product against that customer's systems, data, and rules. Palantir popularised the model. OpenAI uses FDEs to scope, simulate, test, and roll out Presence agents, which is why Presence is sold as a deployment rather than a login.

Does OpenAI Presence support languages other than English?

OpenAI's own Presence deployment is its English-language support line. Among the design partners, SoftBank is testing Japanese-language conversations and BBVA is exploring voice support for banking customers in Mexico. OpenAI has not published a supported-language list, so treat language coverage as something to confirm during scoping.

Is QX Labs an alternative to OpenAI Presence?

Not for customer-facing voice at contact-centre scale; that is Presence's speciality. For the wider set of agent work (research, enrichment, document extraction, lead routing, monitoring, internal Q&A), yes. QX is self-serve with a free plan, shared across a team, connects to 1,000+ apps, and runs on the model you choose.

Where to go from here

If your highest-value agent use case is a customer channel, talk to your OpenAI account team and ask the questions above about pricing, model versioning, and data handling. If most of your agent work is spread across sales, ops, finance, and research, the cheapest way to find out what a platform can do is to give one a real job. Start free on QX and brief an agent this afternoon, or book a demo and we will build one with you. Our guide to the best AI agent platforms and our post on AI employees versus AI agents cover the neighbouring decisions.