On July 22, 2026, OpenAI shipped a product you cannot buy online. Presence is a managed enterprise agent offering delivered through a limited general availability programme, and the company states plainly that it is not yet self-serve. Deployments are led by OpenAI Forward Deployed Engineers and a set of selected global systems integrators.
Access depends on three stated criteria. Workflow fit, AI implementation readiness, and available delivery capacity.
Read the middle one again. The company with the strongest models on the market has concluded that models are not the constraint, and it now screens buyers on their own readiness before it will take the engagement.
Every rival is covering this as a go-to-market story. It is a readiness story.
What OpenAI actually shipped
Presence is sold as a project rather than a product. Each engagement starts with a single job, such as resolving a billing dispute, handling an insurance claim, or clearing an employee IT service request.
The agent receives only the knowledge and system access that one job requires. The customer writes the rules governing what it may do, when it needs sign-off, and when a person takes over. After launch, Codex reads production sessions and escalations, then proposes changes the customer tests and approves before rollout.
That is a sharp departure for a business built on API keys and seat licences. It is also an admission. Two years of enterprise deployments taught OpenAI that the hard part of a production agent sits in integration, permissions, and change management, and it decided to send engineers to do that work rather than ship a dashboard and call the gap a customer problem.
The reframe matters more than the product. For two years the enterprise question was which model to pick. OpenAI has now answered it commercially by saying the model was never the hard part, and it is willing to turn away revenue over the answer. If you want to know how you would score against that judgment, start with the free 60-second assessment and read the gap analysis rather than the number.
Hold the launch framing loosely. OpenAI describes the product as battle-tested, but the performance figures it published for its own support line were graded against its own criteria on its own channel and are not independently verified. The three named customers sit earlier in the cycle than the announcement implies. BBVA is exploring voice support for everyday banking in Mexico, SoftBank is testing Japanese-language conversations, and IAG is exploring support during high-demand events. Design partners are normal at limited availability. None of the three is presented as running Presence at scale.
Why AI implementation readiness became a gate
The commercial logic is not subtle once you look at the failure data. Gartner forecasts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, attributing the failures to escalating costs, unclear business value, and inadequate risk controls rather than to model capability.
A vendor that sells into unready buyers absorbs those cancellations. They become churn, case studies that never publish, and a support burden that scales with the customer count. Screening for AI implementation readiness moves that risk to the front of the funnel, where it is cheap.
There is a second constraint underneath it. Delivery capacity is a consulting problem, not a software problem. Software scales. Engineers cleared into a bank core system do not. Forward Deployed Engineer is a title borrowed from Palantir, where it describes staff embedded in customer operations for months at a time, and those economics look nothing like metered inference.
So OpenAI has to ration, and the criterion it chose to ration on is buyer readiness.
That should reframe how you think about a readiness exercise. It is no longer only an internal planning tool. It is becoming a procurement qualification. If you want to see where you would land against a gate like this, our free 60-second assessment scores the same organizational dimensions the six-stage process below tests.
The six stages, read as a readiness assessment
OpenAI documentation sets out a six-stage deployment process running from scoping business outcomes, through security, privacy, and legal review, then simulation and acceptance testing, staged rollout, and post-launch iteration. The documentation states directly that a Presence agent does not become production-ready simply by ingesting documents.
That is a consulting engagement written down as a product requirement. Read each stage as the question it is really asking about your organization.
Stage one: scope business outcomes
Can you name the outcome and produce its current baseline number? Most companies cannot state their current cost per resolved contact or their current handling time without a two-week data pull. If the baseline does not exist, the agent has nothing to be measured against.
Stage two: security, privacy, and legal review
Do you have a named owner and a decision path, or do you have a queue? Review cycles kill agent projects more often than model quality does, and the difference between a four-week review and a seven-month one is entirely organizational.
Stage three: simulation
Do you hold enough representative historical cases to simulate against, in a form somebody can actually retrieve? Simulation is where most teams discover that their case history lives in three systems with inconsistent labels.
Stage four: acceptance testing
Have you defined what a correct outcome means, in writing, before launch? This is the stage that exposes whether the business and the technical team ever agreed on the goal.
Stage five: staged rollout
Can you roll back? Is there a defined blast radius, and does someone have the authority to pull the agent without convening a committee?
Stage six: post-launch iteration
Who owns the agent in month four? If the answer is a team rather than a person, or a person with no allocated time, the agent will drift and nobody will notice for a quarter.
Not one of those six is a model question. All six are organizational. That is exactly what an AI implementation readiness assessment measures, and it is why the decision-rights work in our AI agent readiness checklist sits upstream of any platform choice.
What to do if OpenAI never calls you back
Most companies will not be in limited availability, and that changes nothing about the requirement.
The same six stages will be applied to you by whichever systems integrator you hire, or they will be applied to you by month five of a deployment that is not working. The only variable is whether you find the gaps before or after you sign.
Running the sequence yourself first is dramatically cheaper. It also changes the negotiation. A buyer who arrives with a documented baseline, a named legal owner, a retrievable case history, and a written definition of a correct outcome is a different customer than one who arrives with a budget and an ambition.
The second kind of buyer pays for the discovery phase. The first kind skips it.
There is a quieter benefit as well. Every one of the six stages produces an artifact that outlives the vendor relationship. A documented baseline survives a platform switch. A named review owner survives a reorganization. If the deployment fails, you still hold the work, which is not true of a proof of concept that ends when the trial credits run out.
Accountability moves when the vendor is also the integrator
One more thing belongs in your evaluation, and it is easy to miss during a demo.
When the model vendor is also the implementation partner, the lines of accountability for a policy misapplied in production need to be written into the contract rather than assumed. Trade coverage of the launch raised this point directly, and it is a fair question to put to any vendor running this model, not only to OpenAI.
The model configuration is a related gap. OpenAI says Presence uses its models with configuration selected for the workflow and subject to change as that workflow evolves. That is defensible engineering, because pinning a production agent to a frozen model version ages badly. Teams that spent the past year building evaluation suites against specific versions will still want the contract to state what they are being held to when the configuration moves.
Knowing which of those questions to ask is itself a readiness marker. Organizations that have never mapped their own decision rights do not think to ask who is accountable when an agent applies a policy incorrectly.
What to build first
Pick one workflow you already intend to hand to an agent. Not a portfolio. One.
Write down the outcome you want and the current baseline number for that outcome. Give yourself one week. If you cannot produce the baseline in a week, you have found your first gap, and it is not a technology gap.
Then name the person who owns that agent in month four, with hours allocated. Those two artifacts, a baseline and an owner, clear more of the six stages than any platform selection will.
Agent programs rarely fail on the model. They fail because nobody could name the outcome, the baseline, or the owner, and the six-stage process makes that visible on day one instead of month five. The free 60-second assessment returns a prioritized gap analysis and a 90-day sequence across those organizational dimensions. Run it against the one workflow you picked, then go find your baseline number.
Frequently Asked Questions
What is AI implementation readiness?
AI implementation readiness is an organization capacity to take an AI system from selection to production and keep it running. It covers whether outcomes and baselines are defined, whether review and approval paths have named owners, whether historical case data is retrievable, whether acceptance criteria are written down, and whether someone owns the system after launch. It is distinct from model capability and is measured organizationally rather than technically.
Why does OpenAI screen buyers on implementation readiness?
OpenAI states that access to Presence depends on workflow fit, implementation readiness, and available delivery capacity. Because deployments are led by its own Forward Deployed Engineers rather than sold self-serve, delivery capacity is limited and every unready customer consumes engineering time that cannot be recovered. Gartner forecasts that more than 40 percent of agentic AI projects will be canceled by the end of 2027 on cost, unclear value, and weak risk controls, which is the risk the screen is designed to filter.
How is AI implementation readiness different from general AI readiness?
General AI readiness asks whether an organization can adopt and sustain AI at all, covering strategy, data, skills, and culture. AI implementation readiness asks a narrower and harder question about a specific workflow, namely whether this particular deployment can be scoped, reviewed, simulated, tested, rolled out, and owned. An organization can score well on general readiness and still fail the implementation gate on a single workflow.
What does the six-stage Presence process require from the customer?
The six stages are scoping business outcomes, security and privacy and legal review, simulation, acceptance testing, staged rollout, and post-launch iteration. Each one requires something from the customer rather than from the vendor, including a measurable baseline, a named review owner, retrievable historical cases, written acceptance criteria, rollback authority, and an accountable owner after launch. OpenAI documentation states that an agent does not become production-ready simply by ingesting documents.
Can a company improve its readiness before buying an agent platform?
Yes, and it is far cheaper to do so before a contract is signed. Start with one workflow, produce its current baseline number, name the person accountable for the agent in month four, and write down what a correct outcome looks like. Those three artifacts clear more of the deployment sequence than any platform comparison, and they change the terms of the vendor conversation.
Be the First to Discover New AI Insights
Follow Elevates.AI on Google to stay updated with the latest AI readiness assessments, governance frameworks, implementation guides, buyer's guides, and enterprise AI best practices.
Follow Elevates.AI on GoogleThe AI Readiness Brief
Every Week, receive practical enterprise AI strategies, implementation frameworks, governance updates, and expert insights—all delivered in a 5-minute read.
| ✓ Enterprise AI Strategy | ✓ AI Readiness Frameworks |
| ✓ Governance & Compliance | ✓ Exclusive Guides & Resources |
Join 500+ AI Professionals
Enter your work email below to receive one high-value email every week. No spam. Unsubscribe anytime.