Generated by Codex with GPT 5.6 Sol High

TBPN surfaced the idea in its October 3 episode, “Sign-In With GPT, Anthropic Vs. the Pope, Paramount Rebrands as Skydance.” The hosts argue that OpenAI’s modest-sounding sign-in feature may matter more to small developers than the week’s flashier agent announcements. It lets eligible users authorize a third-party app to draw on usage already included in their ChatGPT plan, moving part of the inference bill from the app builder to the user who requested the work.

That changes the first question behind an AI product. Instead of asking whether a developer can afford every model call before revenue catches up, the product can ask whether users already value the underlying model enough to bring their own allowance. TBPN calls this “bring your own compute.” The phrase is not technically exact—the computation still runs on OpenAI’s infrastructure—but it captures the economic shift.

Identity and compute are separate permissions

OpenAI’s documentation describes two related but distinct capabilities. The first is identity: an app can let someone create or access an account with their ChatGPT identity. That flow shares basic profile information such as name, email address, and profile picture. It does not by itself expose conversations, memory, files, billing details, or an API key.

The second capability is optional plan usage. An eligible Plus or Pro user can separately allow a participating app to make supported AI requests against the usage included in that person’s ChatGPT plan. The user can set a weekly limit for each app and review its consumption in ChatGPT settings. If the plan limit is reached, the app may draw on ChatGPT credits only when the user has explicitly allowed that too.

This separation matters. “Sign in with ChatGPT” sounds like a familiar social-login button, but delegated model usage turns it into a payment and resource-control boundary as well. A person is authorizing both who they are and, in a second step, how much of a metered service another product may consume.

The launch is narrower than a universal billing layer. OpenAI says identity sign-in is rolling out to selected partners, while plan usage is available to open-source projects, local personal tools, and selected private or commercial clients. Apps can still charge for their own subscriptions, infrastructure, data, physical goods, or premium features. The user’s ChatGPT allowance covers eligible OpenAI inference, not the rest of the business.

A different route for small AI products

The conventional API model puts variable cost on the developer. A useful prototype can become financially dangerous precisely when it spreads: each new user creates another stream of model calls, while conversion to a paid plan may lag. Developers respond with credit systems, strict quotas, separate API-key setup, or an early paywall. Every defense adds friction before the user has experienced enough value to pay.

Delegated plan usage can remove several of those hurdles. An open-source coding tool, research utility, or language-learning app no longer needs every user to create an API account, copy a secret key, and establish a second billing relationship. The builder can concentrate on the “harness”—the workflow, context, tools, and interface around the model—while OpenAI handles identity, model access, and the usage meter.

This does not make inference free. It makes the cost legible to the user and places it inside a subscription the user already manages. Heavy use still consumes a finite allowance, and an app that burns through it must justify why its task deserves that share. Per-app caps are therefore part of the product design, not merely a safety setting.

The app-store bargain returns

TBPN’s larger question is whether this becomes the economic foundation for an AI app store. OpenAI can give small products easier onboarding and access to a large pool of paying model users. In return, it sits between the app and the scarce resource that powers it. Even without a formal revenue share, OpenAI gains demand, usage data, and leverage over which models, request types, and partners qualify.

Developers face the familiar platform bargain. Integration can sharply reduce acquisition and compute friction, but dependence on one provider’s eligibility rules, plan limits, pricing, and permissions can become structural. A product built around delegated ChatGPT usage may find multi-model support harder, while a competing model provider may struggle to displace the subscription users already bring.

OpenAI’s earlier GPT Store showed that distribution alone does not guarantee a durable application economy. This version is more concrete because it joins identity, authorization, and real compute consumption. If it works, the competitive advantage for many AI apps will shift away from merely reselling model calls and toward building a trusted workflow that users willingly spend their own allowance on. The important innovation is not the login button. It is turning a consumer AI subscription into portable infrastructure for software made by someone else.