Lovable Just Raised $400M at a $13.3 Billion Valuation. What It Means When Vibe Coding Crosses Into Enterprise Infrastructure.

On July 22, 2026, OpenAI launched Presence, a managed enterprise product for deploying AI voice and chat agents across customer-facing and internal business workflows. The product is already live at scale: OpenAI uses Presence-powered agents to handle its own English-language customer support, resolving 75% of inbound calls without human intervention.

Presence is available only through a limited general availability program. There is no self-serve option. There is no published price. Every deployment is led by OpenAI's own Forward Deployed Engineers or a small set of approved systems integrators. The reported enterprise entry point is approximately $10 million.

If that description sounds less like a software product and more like a consulting engagement, that is because it is. OpenAI has built a product that it sells the way Accenture sells an implementation project: the vendor's senior engineers embed in the customer's environment, configure the system to the customer's specific workflows, and remain engaged through go-live and the subsequent optimization cycle. The product and the professional services that deliver it are not separable.

This is a significant and underappreciated structural shift in how OpenAI goes to market. The company built the largest consumer AI application in history through self-serve distribution. ChatGPT accumulated users without sales conversations, without implementation projects, and without consulting engagements. Presence inverts every element of that model — and the inversion is deliberate.

What Presence Actually Does

Presence packages the infrastructure required to operate AI agents in enterprise production: policies, system connections, approved action sets, evaluation tools, guardrails, safety simulations, and a continuous improvement loop powered by OpenAI Codex that monitors agent performance and proposes workflow refinements.

The use cases supported at launch are defined and bounded: customer support (voice and chat), outbound sales qualification, and internal IT service desk requests. Each deployment begins with a single workflow. The expansion to additional workflows happens after the first deployment has stabilized and demonstrated measurable performance against the enterprise's defined success criteria.

The voice agents support real-time interruption — a user can cut off the agent mid-sentence and redirect — and the latency profile is designed for live phone calls, not asynchronous chat. The chat agents operate across web, mobile, and internal interfaces. The architecture handles both modalities through a common orchestration layer, which means a support workflow initially deployed as voice can be extended to chat without a separate implementation project.

OpenAI stated that it does not transfer the API credentials to the customer. The enterprise buyer receives the work product — resolved customer support calls, IT tickets cleared, qualified leads — not the underlying model access. This is the same structure as OpenAI's Daybreak cybersecurity program, where security research outputs flow to enterprises through vetted partners rather than directly via API.

The Forward Deployed Engineer Model

The FDE model is the defining commercial and operational characteristic of Presence. Understanding it requires understanding why OpenAI made this choice when it had a perfectly functional self-serve API that enterprises were already using to build agents on their own.

The honest answer is that self-built enterprise AI agents at production scale have a high failure rate. 88% of enterprise AI agent pilots fail to reach production by some estimates. The failure modes are consistent: agents that work in demos fail on edge cases at production volume; guardrail calibration that was adequate in testing proves too restrictive or too permissive with real users; system integrations that work in controlled environments break under production load; absence of evaluation frameworks means problems are not detected until they cause customer impact.

OpenAI's FDE model is a response to this failure rate. Rather than shipping a self-serve API and letting enterprises build agents that fail, OpenAI ships the implementation capability alongside the model capability. The FDE ensures that the guardrails are calibrated for the specific workflow, the system integrations are tested against real production data before go-live, the evaluation framework is defined and instrumented, and the improvement loop is operating before the customer is left to manage the deployment independently.

This is not a new pattern. Palantir built its enterprise market on a similar model: Palantir Forward Deployed Engineers embedded in customer organizations, building deployments of Palantir's data infrastructure against specific analytical problems. The FDE model produced Palantir's enterprise revenue but also limited Palantir's growth rate relative to what a self-serve product could have achieved. OpenAI is making the same explicit tradeoff: higher initial quality and lower initial scale in exchange for a track record of successful deployments that builds the credibility required to sell at larger enterprise scale later.

The $10 Million Entry Point and What It Signals

The reported $10 million enterprise entry point for Presence is not a price for a software license. It is a price for an implementation engagement that includes FDE time scoped to the specific workflow, system integration development and testing, guardrail and policy calibration, safety simulation and evaluation framework setup, go-live support and stabilization, and initial improvement loop operation.

At $10 million, Presence is priced in the same tier as a major enterprise consulting engagement from a tier-one SI firm. This is intentional. OpenAI is not competing with Salesforce Agentforce at $2 per resolved conversation. It is competing with the Accenture or IBM engagement that an enterprise would have commissioned to build an equivalent system using OpenAI models through the standard API, plus the internal engineering hours required to implement and maintain it.

The total cost of ownership comparison requires more detail than Presence has disclosed publicly, but the framing is important: the $10M entry point is an alternative to a build-it-yourself implementation using OpenAI APIs, not an alternative to a per-resolution software subscription. For enterprises that have already attempted to build enterprise AI agents in-house and experienced the production failure pattern, an OpenAI-delivered implementation may have a better cost-risk profile than another attempt at self-building.

Enterprise AI agent ROI payback periods average 14 months for organizations that reach production, and considerably longer for organizations whose implementations fail and require restart. The Presence value proposition is straightforward: a 75% resolution rate from day one of deployment, delivered and maintained by the model vendor, at a fixed engagement cost that removes the open-ended implementation risk from the enterprise buyer.

How Presence Sits in the Competitive Field

The enterprise AI agent market has at least four distinct purchasing patterns in 2026, and Presence competes differently against each of them.

Against Salesforce Agentforce: Agentforce is the default choice for organizations whose agent workflows are embedded in Salesforce CRM data. Customer service agents that need to query order history, case history, account health, and product entitlement data have the highest resolution rates when the agent is native to the system holding that data. Presence's advantage over Agentforce is ecosystem independence: it does not require the workflow to be Salesforce-data-native, which matters for organizations whose customer data lives in Zendesk, SAP, or a proprietary CRM. Salesforce's bundling strategy and the 34% AI add-on cost shock has also created friction for Agentforce adoption among organizations that feel the Salesforce pricing architecture is compounding their enterprise software costs.

Against Genesys and contact-center-native AI: Genesys Cloud builds its virtual agent on large action models (LAMs) designed to plan and execute multi-step tasks across front and back-office systems within the Genesys platform. For organizations running Genesys as their primary contact center platform, the Genesys-native AI agent capability avoids the integration complexity of introducing an external AI layer. Presence's advantage is model quality: OpenAI's frontier models are not matched by the models underlying Genesys AI in general capability, which matters for workflows with high ambiguity or complex multi-turn reasoning requirements.

Against custom API builds: The enterprise buyer who would otherwise assemble their own agent using OpenAI's APIs, a vector database, an orchestration framework, and internal engineering resources is the most natural Presence candidate. The comparison is not Presence's price versus a software subscription — it is Presence's price versus the internal engineering cost, the implementation timeline, and the operational risk of a self-built system. For organizations that have tried and failed at self-built agents, this calculus often resolves in Presence's favor.

Against competing managed AI services: Microsoft Azure AI Foundry, Google Cloud Vertex AI, and Anthropic's enterprise programs all offer managed AI deployment paths, though none has launched a comparable managed agent product with FDE delivery at the same scope as Presence. Anthropic's enterprise inference hooks and DLP integration address enterprise governance requirements but through the API layer rather than a managed FDE delivery model.

What the FDE Model Changes About AI Procurement

Presence represents OpenAI's largest single bet on a distribution model that is not self-serve. Every prior OpenAI product — ChatGPT, the API, operator partnerships, the GPT Store — has distributed at scale through customer self-activation. Presence is designed to grow through a channel that is fundamentally capacity-constrained: the number of FDEs OpenAI can deploy.

The business model implications are significant. A self-serve AI product's revenue scales with adoption and usage. An FDE-delivered product's revenue scales with the number of successful deployments OpenAI can execute, which is bounded by FDE headcount and project duration. OpenAI can only run so many $10M+ engagements simultaneously.

For enterprise system integrators, the Presence launch creates a specific commercial threat: OpenAI is now selling delivery capability, not just model access. The SI firms whose enterprise AI practices are built around OpenAI API implementations now have a competitive product from their own vendor. The mitigating factor is that OpenAI's FDE headcount cannot scale to the breadth of the enterprise market — which is why OpenAI's approved SI partner program matters. Getting on that approved list is a priority for every major SI firm.

1. Assess whether your AI agent use case fits the Presence profile. Presence is optimized for well-defined, high-volume, measurable workflows — call center resolution, IT service desk, outbound qualification. Workflows with high variance in customer intent, complex multi-system dependencies, or specialized domain knowledge requirements are less likely to be served well by the current Presence framework, which is still in limited GA.

2. Understand the FDE availability timeline before committing. OpenAI's FDE capacity is constrained during limited GA. If there is a six-month queue to begin an implementation, the total timeline to agent deployment may not match your business need.

3. Get a total cost of ownership comparison in writing. The $10M entry point for implementation needs to be compared against the full cost of alternatives: Agentforce licensing plus implementation services, Genesys AI plus integration, a custom build with ongoing maintenance. The comparison should use the same time horizon — typically 36 months — and include the cost of ongoing optimization and support in all scenarios.

4. Evaluate your existing OpenAI contractual relationship. Organizations that already have significant OpenAI API spend have more leverage in Presence pricing negotiations than organizations approaching OpenAI as a new vendor. Presence pricing appears to be negotiated per engagement, which means the commercial relationship context matters.

5. Ask explicitly about model dependency and future pricing. Presence currently uses OpenAI models, with model selection determined by OpenAI based on the workflow. As new model generations are released, understand what happens to existing Presence deployments — are model upgrades included, and at what cost? The improvement loop that refines agent behavior: who controls it, and can the enterprise audit or override proposed changes? These governance questions have long-run commercial implications.

The Broader Shift: When AI Vendors Become Integrators

The OpenAI Presence model is the clearest evidence yet that the AI industry's leading vendors are no longer content to be infrastructure layers. Infrastructure vendors sell compute and APIs. Integrators sell outcomes. OpenAI, with Presence, is selling outcomes — the resolved customer support call, the cleared IT ticket, the qualified sales lead — rather than the API calls required to produce them.

The agentic GTM adoption gap documented across B2B sales organizations reflects the same dynamic: enterprises understand the potential value of AI agents but distrust their ability to realize it through self-directed implementation. Presence is a direct commercial response to that distrust.

The structural question for the next 24 months is whether the FDE model is a transitional phase — where OpenAI uses high-touch delivery to build deployment playbooks and then codifies those playbooks into a self-serve product — or whether managed delivery becomes a permanent, growing revenue stream alongside the API business. The Palantir precedent suggests both can be true: Palantir's FDE model never fully evolved into self-serve, but it did build a durable enterprise franchise that coexists with Palantir's platform business.

Enterprise AI token economics and the Jevons paradox will shape how Presence pricing evolves: as model costs continue declining, the economics of a managed deployment should improve on OpenAI's margin even as the competitive pressure on per-resolution pricing intensifies from Agentforce, Genesys, and the growing field of contact center AI specialists.

Why the Limited GA Phase Matters More Than the Product Roadmap

The limited general availability structure of Presence is not a standard product launch tactic. It is OpenAI's explicit signal that the company does not yet know how to scale this product. The FDE requirement is a quality control mechanism that limits total deployment count to the number of qualified FDEs in the building. Every Presence deployment in limited GA is also a training exercise: OpenAI is learning what configuration patterns work across different industries, workflow types, and enterprise data environments.

What emerges from the limited GA phase is not just a product but a deployment library — a set of validated playbooks for specific workflow categories that can then be packaged into a more scalable offering. The six months of limited GA before broader availability is when those playbooks get written. Organizations that participate in limited GA get more hands-on OpenAI FDE attention than any future customer will receive, because they are the research subjects for the playbook documentation.

This creates a specific buying signal for enterprise decision-makers: if your organization's use case is one that Presence is specifically designed to address — customer support, IT service desk, outbound sales — getting into the limited GA program now gives you both early implementation advantage and a level of OpenAI FDE investment that the production release will not replicate. The trade-off is committing to a product that is still being defined, at a price that has not been publicly disclosed, with a vendor relationship that will evolve significantly as the product matures.

Takeaway: OpenAI Presence is the company's most consequential distribution experiment since ChatGPT — and it runs in the opposite direction. Instead of frictionless self-serve, Presence requires OpenAI Forward Deployed Engineers, has no self-serve path, reports a $10M entry point, and limits availability to a controlled GA program. The 75% resolution rate on OpenAI's own support line validates the product's technical capability. Enterprise buyers should treat Presence as a bespoke professional services engagement — not a software purchase — and assess it against the full build-or-buy TCO for their specific agent workflow, with the FDE availability constraint and model dependency governance questions answered before any commitment.