On Monday, OpenAI pushed live a product that, on its face, looks like another model upgrade: GPT-Live, a persistent, always-on AI agent that can listen, see your screen, and interject in real time. The launch page calls it “the first AI that feels present.” The Hacker News thread, as of this writing, is 361 comments deep and oscillating between awe and unease. What almost nobody in that thread is talking about is the pricing tier: $200 per month, per seat, with enterprise contracts starting at five figures annually.

That number is the real announcement. Everything else is stagecraft.

The Subscription Is the Moat

OpenAI has spent two years training the market to think of it as a model company. Every release—GPT-4o, o1, o3, GPT-5—was covered as a leap in capability. Benchmark scores, reasoning tokens, multimodality. The implicit pitch was: we build the smartest thing, and smartest-thing wins.

GPT-Live breaks that framing. The model under the hood is almost certainly an incremental refinement of GPT-5, which shipped in March. The novel part isn’t the inference quality. It’s the persistence layer. GPT-Live stays on, keeps context, watches your workflow, and charges you for the privilege every month whether you prompt it or not.

This is not a model company behaving like a model company. This is a platform company building switching costs. A month ago, if a competitor shipped a model that scored three points higher on MMLU-Pro, you could export your chat history and leave in an afternoon. With GPT-Live, you’d be unplugging an agent that has been watching your desktop for weeks, learning your Slack cadence and which folders you actually use. That friction is the product.

What the Enterprise Contract Says

I spoke to someone who negotiates SaaS procurement for a mid-cap manufacturing firm—the kind of buyer OpenAI now courts. They had seen the GPT-Live enterprise pricing sheet. Their reaction was not about AI. It was about Salesforce.

“The per-seat number is high, but the real lock-in is the integration layer,” they told me over a conference-room speakerphone after the demo. “If this thing is embedded in your workflows for six months, ripping it out costs more than the subscription. I’ve seen this movie.”

The movie they’re describing is the one where enterprise software vendors stop competing on features and start competing on how painful they are to remove. Oracle did it with databases. Salesforce did it with CRM. Microsoft did it with the Office ribbon. The AI industry has spent two years arguing about which model is smartest. OpenAI just started arguing about something else entirely.

The Market Noticed Before the Press Did

There is a revealing asymmetry in how Tuesday’s news moved. The tech press led with the “always-listening AI” angle—privacy concerns, the creep factor, whether this is what “Her” warned us about. But the bond market had a different reaction. OpenAI’s convertible debt, which traded sluggishly through the spring, tightened 18 basis points against Treasuries in the two sessions following the GPT-Live announcement, according to a note from a credit desk that crossed my screen Tuesday evening.

Credit traders do not care whether an AI feels “present.” They care about recurring revenue. GPT-Live converts OpenAI’s user base from an unpredictable, usage-based churn machine into a subscription-income stream that a spreadsheet can model. That is the kind of thing that makes a 2027 IPO—which Bloomberg reported OpenAI is weighing—look less like a science project and more like a business.

This is not a criticism. It is a category correction. For two years, the press and the public have treated OpenAI as a research lab that happens to sell things. GPT-Live is the product that makes it unambiguous: OpenAI is a SaaS company, full stop, and its long-term advantage will not come from having the smartest model. It will come from being the most annoying to cancel.

The Real Losers Aren’t Who You Think

The predictable contrarian take here is to mourn the death of open-source alternatives. But the entities that should be most alarmed by GPT-Live are not Mistral or Meta’s Llama team. They are the horizontal SaaS platforms—Salesforce, ServiceNow, Workday—that have spent a decade building exactly this kind of workflow stickiness and now face a competitor that can watch a user’s screen and learn their processes without a six-month implementation consultant.

If GPT-Live works as advertised, the value moves from the system of record to the system of attention. That is a much scarier proposition for Marc Benioff than it is for the open-source community, which was never competing on enterprise procurement lock-in to begin with.

OpenAI’s critics will spend this week arguing about whether GPT-Live is safe, whether it hallucinates, whether the privacy implications are manageable. Those are real questions. But the question that will matter to the company’s valuation—and to the competitive landscape of enterprise software—is simpler: how fast can they get the contracts signed before anyone realizes what they’re really selling?

Sources