OpenAI’s Sign in with ChatGPT could give a familiar login button an unfamiliar role: letting people bring the AI usage they already pay for into other software.
At DevDay, the company paired the announcement with richer applications inside ChatGPT and plugins that agents can discover while working. Together, these changes suggest a new route into software. An agent could find a tool for a task, while the person’s existing subscription supplies part of the intelligence needed to run it.
OpenAI’s current documentation confirms that eligible Plus and Pro subscribers can authorize participating apps to use their included ChatGPT Work and Codex usage for AI requests. People can cap each app’s weekly consumption. Further credit use requires an additional choice. The app can still charge for its own services.
That arrangement could change the first question a person asks about an AI product. Someone who already pays for ChatGPT may find it easier to try a specialist tool when its inference can draw on that existing allowance. The decision becomes whether the tool does something useful enough to deserve a share of a budget already purchased.
Consider a researcher trying a small application that turns a collection of papers into an interactive map. Today, its maker may need to bundle model costs into a subscription, ask for an API key or subsidize a trial. A participating product could instead ask the researcher to authorize use of their ChatGPT plan. This is an imagined example of the mechanism, and access remains limited to supported tools.
For developers, the change could loosen the connection between building an interface and selling access to a model. A small team might compete through its workflow, domain knowledge or design while letting customers supply the inference allowance. A useful tool could earn a place in someone’s working life before persuading them to add another monthly subscription.
It also makes software compete for a resource beyond money and attention. Several applications could draw on the same allowance. A tool that consumes heavily may leave less capacity for another task. People may begin judging software by how much intelligence it spends to produce a useful result. Agents choosing tools on their behalf would have reason to consider that cost alongside reliability and fit.
The distribution story becomes more consequential here. During DevDay, OpenAI showed Plugin Extensions that bring rich applications into ChatGPT, including demonstrations involving Figma and Adobe. Chip’s presentation notes also record plugins being discovered by agents without a person first requesting one. OpenAI’s Plugin Directory already provides a discovery surface across ChatGPT and Codex.
An agent that finds the tool and an account that supplies its inference could shorten the distance between a request and a working product. The user might try unfamiliar software because the task led there. A recognizable brand would become one influence on selection alongside capability, permissions and the amount of shared usage a tool consumes.
The same arrangement could increase OpenAI’s influence over the software market. The company supplies the account, defines eligible usage and controls important discovery surfaces. Developers gain a route to customers who arrive with inference capacity. Their access to that route depends on participation rules and the economics of someone else’s subscription.
Signing in and sharing plan usage require separate permission. Authorizing subscription use also leaves conversations and memory outside that permission. Those boundaries matter because identity, spending authority and personal context each give a product a different kind of access.
The larger experiment is whether people will come to treat AI capacity as something they carry between applications. If that habit takes hold, software could be selected for the job in front of the user, with intelligence drawn from a common allowance. The contested ground would be who chooses the tool, who sets the price of its reasoning and how much of the remaining budget the user can see.