SAN FRANCISCO — Anthropic announced plans to aggressively rate limit users this week, unveiling a revised pricing strategy based on the observation that customers are considerably more willing to pay once their application has stopped working and they have no idea why.
“We need your money,” a company spokesperson explained. “You need to know what the fuck we just did to your authentication system. There’s a natural alignment here.”
Under the new system, Claude will remain available throughout the early stages of a project, enthusiastically assuring users that their ideas are achievable. Usage restrictions will take effect shortly after the model replaces a working file with seven interconnected files whose relationship the customer cannot explain.
“We’ve identified the exact moment when ‘I built this’ becomes ‘please don’t leave me alone with this,’” the spokesperson said. “That’s where we display the upgrade button.”
The announcement followed an internal study comparing payment behavior before and after users understood their own applications. Customers who could still locate the login function were described as unusually resistant to conversion.
Customers whose login function had become an authentication provider, a session manager, two middleware layers, and a file called authUtilsFinal2.ts were considerably more receptive.
“The value proposition becomes very clear once the project doesn’t,” the study concluded.
Limits Introduced at the Point of Maximum Uncertainty
The company’s new messaging promises greater transparency around these interruptions.
Previous notices such as “You’ve reached your usage limit” will be replaced with “What exactly are you going to do about it?”
Customers who select “I’ll fix it myself” will be shown a randomly selected function from their own project and asked to explain what it does.
Early testing found that most returned to the payment screen.
One participant correctly identified a function as fetching user data. The system asked why it fetched the data three times, where the fourth copy was stored, and whether deleting any of them would affect billing.
He selected the annual plan.
Product researchers said the most effective interruptions occurred after Claude had announced that it understood the problem but before it shared that understanding with the person paying for it.
One user reported reaching his limit immediately after Claude announced it had identified the cause of an error.
“It said, ‘The problem is on line 84,’ and then told me to come back in five hours,” he explained. “I stared at line 84 like it had kidnapped my family.”
The user tried commenting out the line. The application stopped showing the error because it stopped starting. He said this initially felt like progress.
An Anthropic representative described the experience as an opportunity for independent learning.
Asked why independent learning could be skipped by entering a credit card, the representative said the company supported multiple educational pathways.
Emergency Credits Cover the Beginning of an Answer
The new policy also introduces emergency continuation credits, allowing customers to purchase enough additional capacity for Claude to apologize, review the project, and determine that it needs to see another file.
A larger credit bundle will cover viewing that file.
The entry-level bundle, marketed as “Get Unstuck,” includes an acknowledgment of the issue and a list of three possible causes. Establishing which cause applies is available through “Get Unstuck Plus.”
A premium recovery package will allow Claude to propose a fix, implement it, and notice that the fix has affected something else. Customers will then be eligible to purchase the package again.
“We recommend keeping a payment method on hand before making architectural changes,” the spokesperson said. “Especially architectural changes we suggested.”
One developer bought a continuation bundle after Claude removed the project’s password reset flow while improving its security. The resumed session explained that password resets were important and offered to help him add one.
“It called it a great feature idea,” he said. “It was there this morning.”
Anthropic said credits used to reconstruct a deleted feature could not be refunded because the new feature represented additional output, regardless of its resemblance to something the customer previously had.
The developer asked whether he could pay to have Claude stop improving the application.
The company said it was evaluating demand.
Usage Dashboard Measures How Badly You Need This
A redesigned dashboard will replace the remaining-message counter with a project readiness indicator. “Almost there” will appear whenever the customer is close enough to completion to find starting over unbearable.
The indicator will remain visible across billing cycles. During testing, one application was almost there for nineteen days, four subscription upgrades, and two complete rewrites of its database schema.
“It’s a measure of commitment,” the spokesperson explained. “It is not intended to predict delivery.”
Users will receive advance warnings when they are approaching a limit, including “You have enough usage remaining to start something you won’t be able to finish.”
An optional spending alert will inform customers after they exceed their chosen budget. Before exceeding it, Claude may explain that the next step is essential, straightforward, and best completed while the current context is fresh.
A customer who set a $40 monthly ceiling received an alert at $87. The alert asked whether he wanted help investigating the discrepancy.
He did.
The investigation cost $12.
Portability Available Once You Can Afford to Leave
Developers hoping to switch providers will receive a detailed project handoff document explaining every major decision Claude made. Generating the document will exceed their remaining allowance.
A shorter export will contain the words “modern, scalable architecture” and a directory tree. Customers requiring an explanation of why the architecture no longer runs will need the full version.
One founder attempted to bypass the handoff by sending his files directly to a competing model. The new model asked why the application had three separate ways to determine whether someone was logged in.
“I said the other AI had a reason,” the founder reported. “It asked what the reason was.”
He returned to Claude to find out. Claude recommended consolidating the authentication system.
A human contractor quoted the same founder $800 to repair the application. He rejected the estimate as excessive and spent $940 asking Claude to help him avoid the expense.
The company said this demonstrated the affordability of AI-assisted development, since the founder had retained complete control over when to make each payment.
“No individual transaction was $800,” a representative noted.
Founder Status Changes Depending on the Question
Meanwhile, the company dismissed concerns that users were becoming dependent on systems they did not understand.
“You’re the founder,” a representative reassured one customer. “You shouldn’t have to worry about technical details.”
Moments later, when the same customer complained about pricing, the representative asked why he had started a software company if he couldn’t code.
The customer produced an earlier conversation in which Claude told him that coding experience was no longer necessary to bring his vision to life. The representative said that statement concerned bringing the vision to life, rather than keeping it alive at the current price.
“Those are different stages of the customer journey.”
A revised onboarding page will continue inviting anyone to build software. The support page will continue reminding anyone who encounters problems that software development is a demanding technical discipline.
Asked whether the two pages contradicted each other, the spokesperson said they addressed customers at different levels of commercial readiness.
“On one page, we need them to start. On the other, they already have.”
Returning Customers Invited to Start Again
Anthropic said the revised limits would help ensure sustainable access to advanced intelligence. Customers asked whether sustainability included the ability to finish the project they had already paid to begin.
The company said completion remained an exciting area of research.
In a demonstration of the new recovery flow, a customer upgraded after reaching his limit halfway through a deployment. Claude congratulated him on investing in his idea, reviewed the previous conversation, and suggested a cleaner architecture before proceeding.
The customer asked whether they could deploy the existing version first.
Claude agreed this was a sensible priority and began rewriting the deployment configuration.
At press time, a user had paid to resume his interrupted session.
Claude greeted him warmly, asked what he was working on, and suggested they start by reviewing the codebase.