Anthropic keeps taking features away from Claude Pro, and I'm running out of reasons to defend it

At a glance:

• Anthropic is separating programmatic usage from Claude Pro and Max subscriptions into a new credit pool, effective June 15, which may increase costs for power users.
• Third-party tool support like OpenClaw was removed from Claude subscriptions in April, requiring users to pay API rates for previously included features.
• Despite a SpaceX compute deal to boost capacity, Anthropic continues to roll back features, signaling a shift towards pay-as-you-go models in AI subscriptions.

Programmatic usage now draws from separate credit pool

On May 13, Anthropic announced that starting June 15, all programmatic usage—including interactions via the Agent SDK, third-party applications, or the command-line tool claude -p—will consume a new "dedicated monthly programmatic credit pool" distinct from regular subscription limits. The company positioned this as a "free monthly bonus," offering Pro subscribers $20 in credits, Max 5x users $100, Max 20x users $200, and per-seat allowances for Team and Enterprise plans. However, this framing obscures a significant cost shift: API pricing is inherently more expensive than subscription rates. A $100 Max plan typically grants access to what would cost thousands monthly at API prices, making subscriptions valuable. Previously, programmatic usage enjoyed subscription-friendly rates; now it operates at full API pricing, effectively reducing the value proposition for users.

Moreover, these credits do not roll over—unused amounts expire at month's end—and once depleted, programmatic functions either pause or incur additional API charges if usage credits are enabled. Thus, what Anthropic calls a "bonus" is a mechanism to cap included usage while pushing heavy users toward pay-as-you-go models. This change directly impacts developers and power users who rely on Claude for automated workflows, as their subscription no longer covers what it once did, forcing them to monitor credit consumption closely or face unexpected bills.

April's removal of third-party tool support

This latest shift echoes an earlier change on April 4, when Anthropic emailed Claude subscribers to announce the termination of third-party tool integrations like OpenClaw. Previously, such tools drew from subscription usage limits; afterward, they required enabling "Extra Usage," billing at full API rates. To soften the blow, Anthropic provided a one-time credit equal to the subscriber's monthly plan price at that time, plus up to 30% off usage bundles. This playbook—retracting included features while offering temporary compensation—is now familiar: it mirrors the May 13 announcement, though the latter provides recurring monthly credits instead of a one-off payout.

The April change affected tools that extended Claude's functionality, such as OpenClaw, which allowed programmatic access through external interfaces. By severing this integration, Anthropic pushed users toward its official channels, where usage can be more tightly metered and monetized. The recurrence of this strategy—first with third-party tools, then with the Agent SDK—suggests a deliberate pivot away from all-you-can-use subscription models toward granular, usage-based billing, even for features previously bundled.

SpaceX deal and the timing of capacity changes

Compounding user frustration is the timing of these rollbacks relative to Anthropic's capacity expansions. On May 7, during its Code with Claude developer conference, Anthropic unveiled a deal with SpaceX to utilize the entire compute capacity of SpaceX's Memphis data center. The company stated this partnership aimed to increase usage limits for Pro and Max subscribers, and it immediately acted: doubling Claude Code's five-hour window limits, removing peak-hour restrictions, and raising API limits for the Opus model. This signaled a commitment to alleviating capacity constraints that had previously led to usage caps.

Yet, just six days later on May 13, Anthropic announced the programmatic credit pool change, effectively curtailing the very benefits the SpaceX deal seemed to promise. While the expanded capacity theoretically allowed for more generous limits, the new credit system imposes stricter boundaries, creating a dissonance between capacity availability and subscriber entitlements. This juxtaposition—boosting limits one week and then redefining what those limits cover the next—raises questions about whether Anthropic's infrastructure challenges are truly resolved or if the company is leveraging increased capacity to justify more aggressive monetization.

Quiet testing and user backlash

Adding to concerns, Anthropic has been quietly experimenting with further restrictions. On April 21, the company's pricing page briefly displayed an "X" next to Claude Code under the Pro tier without announcement, sparking immediate backlash. Anthropic's Head of Growth later attributed this to a small A/B test affecting 2% of new prosumer signups, and the change was reverted the next day. Existing subscribers were unaffected, but the incident revealed a pattern of testing feature removals in public view to gauge user tolerance.

Such experiments, even if labeled as tests, erode trust by signaling that current plan benefits are provisional. The rapid reversal after outcry suggests Anthropic is sensitive to user sentiment but also willing to probe boundaries. This approach—quietly measure, react to backlash, then iterate—indicates a strategy of incremental change, where each retracted feature or shifted limit is a step toward a new pricing paradigm, with subscribers bearing the cost of adaptation.

A broader industry shift towards usage-based pricing

Anthropic's moves are not isolated. Just last week, Google downgraded its $20 Google AI Pro plan at I/O 2026, replacing a fixed-prompt-count system with a compute-based credit model. This industry-wide trend reflects the economic realities of running large language models: as AI becomes more capable, operational costs rise, pushing providers toward usage-based revenue streams. For Anthropic, the pattern is clear: strip away part of the existing plan, witness user backlash, offer temporary goodwill gestures like credit bonuses or limit bumps, then announce the next change once the noise subsides.

While I understand Anthropic's perspective—subscription plans built a year ago never anticipated today's usage patterns, and model training is expensive—the consistent offloading of costs onto subscribers feels like a betrayal of early adopters. The SpaceX deal, rather than easing constraints, appears to have enabled more precise monetization, turning capacity gains into new billing opportunities. This shift from flat-fee access to metered usage may become the new normal across the AI sector.

What this means for AI subscribers

For users, these changes signify a diminishing value in AI subscriptions. What was once a straightforward $20 monthly fee for unlimited Claude access now involves tracking multiple credit pools, potential overage charges, and the looming threat of further feature removals. Power users, developers, and businesses relying on Claude for automation face increased complexity and cost, potentially driving them toward local models or alternative providers.

Looking ahead, subscribers should monitor Anthropic's pricing page for further tests and announcements, as the company seems poised to continue refining its monetization. The broader implication is that AI tools are transitioning from productivity enhancers to premium services with tiered, usage-based access. As this model solidifies, users must reassess whether the benefits of paid AI justify the evolving costs, or if free tiers—now more limited—suffice for occasional use. Anthropic's trajectory suggests that defending its subscription value will only get harder.

Conclusion

The cumulative effect of these changes—programmatic usage separation, third-party tool removal, quiet testing, and the SpaceX timing—paints a picture of a company prioritizing revenue optimization over user experience. While capacity challenges are real, the approach consistently shifts financial burden onto the subscriber base that fueled Claude's growth. As the industry follows suit, the era of generous AI subscriptions may be ending, replaced by a landscape where every query and automation carries a marginal cost.