I watched the backlash unfold across social media with a mixture of sympathy and frustration. Anthropic's decision to block third-party applications from accessing Claude through consumer max subscriptions generated immediate outrage. Users who had built workflows around tools like Cursor, Windsurf, and various API wrappers felt blindsided. The narrative quickly solidified: Anthropic was being greedy, punishing loyal customers, and breaking functionality that people depended on. However, having spent considerable time thinking about the economics of AI services, I found myself on the opposite side of the argument. The restriction was not merely justified — it was inevitable.

The core issue is deceptively simple. Consumer subscriptions and API access are fundamentally different products with fundamentally different pricing structures. When I pay for a max subscription, I am paying for personal, interactive use of Claude through Anthropic's web interface or official apps. The pricing reflects human usage patterns: thinking time between prompts, reading responses, occasional intensive sessions balanced by quiet periods. API access, by contrast, is priced for programmatic use — automated systems, integrations, and applications that can send requests continuously without human latency in the loop.

Third-party tools that routed their requests through consumer subscriptions were exploiting a gap in enforcement rather than a feature. They consumed computational resources at API-level intensity while paying consumer-level prices. This is not a sustainable arrangement for anyone except the free-riders.

Consider what happens when a developer integrates Claude into an application using a back-door method. The application might send dozens or hundreds of requests per hour, far exceeding what any individual user would generate through manual interaction. Each request consumes the same GPU resources, the same inference compute, the same electricity. Anthropic's pricing for API access accounts for this intensity. Consumer pricing does not. The arbitrage was always temporary — a grace period while Anthropic focused on growth rather than enforcement.

I understand why users felt aggrieved. The tools built on this access were genuinely useful. Cursor's AI-assisted coding features became essential to many developers' workflows, much as Claude Code has become central to mine. Various browser extensions and automation tools extended Claude's capabilities in creative ways. When that access disappeared, real productivity was lost. The emotional response makes sense. Nevertheless, the underlying expectation — that consumer subscriptions should subsidise commercial application usage indefinitely — was never reasonable.

The comparison to other services clarifies the issue. Netflix does not allow me to use my personal account to stream content in my coffee shop. Spotify does not permit me to use my individual subscription to provide background music for a business. These restrictions exist because personal and commercial use represent different value propositions with different cost structures. The AI industry is simply catching up to distinctions that other subscription services established long ago.

Additionally, the unrestricted third-party access created problems beyond revenue leakage. Quality of service for legitimate users suffered when shared infrastructure handled disproportionate loads from automated tools. Security concerns emerged as third-party applications with varying levels of trustworthiness gained access to the same systems as verified users. Support costs increased as Anthropic fielded complaints about issues originating in third-party implementations they could not control. The externalities affected everyone.

Some critics argued that Anthropic should simply offer a middle-tier pricing option — something between consumer subscriptions and full API access. This sounds reasonable until you examine the complexity it creates. Pricing tiers require enforcement mechanisms. Enforcement mechanisms require technical controls. Technical controls create exactly the kind of restrictions that users were complaining about in the first place. There is no policy-free solution to the problem of misaligned incentives.

The API exists precisely for users who need programmatic access. It offers granular usage tracking, proper authentication, rate limiting, and pricing that reflects actual resource consumption. Developers building applications on Claude should use it. The fact that a back door existed — and that Anthropic initially tolerated its use — does not create an entitlement to its continuation. Technical debt in access controls is not a feature.

I also want to address the accusation of corporate greed that surfaces whenever a technology company enforces its terms of service. I have questioned subscription value myself in the past, so I understand the frustration. Anthropic operates expensive infrastructure. Training and running large language models requires substantial capital investment. The company has obligations to its investors, employees, and long-term mission. Allowing systematic underpayment for commercial-scale usage is not generosity — it is unsustainable business practice that ultimately threatens the service everyone depends on. A company that cannot capture appropriate value for its products will not continue providing them.

Furthermore, the timing of the restriction likely reflects maturation rather than sudden avarice. Early-stage companies frequently prioritise growth over monetisation. They tolerate edge cases and workarounds because the user base is still developing and rigid enforcement might discourage adoption. As the product matures and usage patterns stabilise, enforcement catches up to policy. This is normal. The alternative — never enforcing terms of service — benefits only those who learned to exploit the gaps.

The path forward for affected users is clear, if inconvenient. Those building applications should use the API. Those using third-party tools should pressure those tools to implement proper API integration. The additional cost reflects the actual value being consumed. Complaints that API pricing is too expensive are complaints that the market price exceeds what users wish to pay — a common sentiment that does not constitute an argument for subsidised access.

I recognise that this position places me against the prevailing sentiment among the users most directly affected. Technical communities tend toward libertarian instincts about access and ownership. The feeling that one's subscription should entitle unlimited use in any manner one chooses runs deep. However, feelings do not change economics. Consumer subscriptions were never intended to subsidise third-party commercial applications. The gap in enforcement was a grace period, not a promise. The door has closed because it was never really open.

Anthropic made the correct decision. The implementation may have been abrupt, and better communication might have softened the transition. Those criticisms are fair. Yet the underlying policy — that API-level usage requires API-level pricing — is both economically necessary and ethically defensible. I suspect the loudest critics will, with time, acknowledge this. Or they will move to other providers and discover that the same economic constraints apply everywhere. The laws of computational economics do not bend to user preferences. They did not bend here, and they will not bend elsewhere.