Detailed Analysis
Anthropic's Claude platform is facing a visible wave of user frustration following what appears to be a policy change restricting third-party and automated access to Claude subscriptions — a shift that has prompted public backlash directed at Boris Cherny (@bcherny), a prominent Claude team member, across social media. The focal tweet from @codyplof — "Actively working on improving reliability and usability. Bear with us" — serves as Anthropic's public acknowledgment of ongoing friction, even as the replies reveal the depth of discontent: users demanding refunds, reporting broken credits, describing abandoned internal projects, and migrating to competitors including OpenAI, Gemini, Qwen, and open-weight models like MiniMax. The common thread across the complaints is a perception that Anthropic has unilaterally closed off workflows that paying subscribers had built and relied upon — particularly those using tools like OpenClaw, a third-party Claude interface — without providing adequate alternatives or compensation.
The usage limit controversy sits at the center of the backlash. Multiple users report that the $20/month Claude Max plan imposes session and message caps — one user citing a limit of eight messages before hitting a five-hour cooldown — rendering the tier functionally inadequate for the agentic and automated coding workflows that Claude Code was marketed to support. The frustration is compounded by the perception that competing platforms like OpenAI's Codex permit similar automated uses. Anthropic's apparent reasoning — limiting token burn from high-volume automated users exploiting subscription plans at API-equivalent consumption rates — is economically rational, as one commenter analogizes to ISPs throttling the top traffic users in the early internet era. However, the execution of this policy shift, particularly the reported failure of promised credits and the lack of a clear refund pathway, has amplified grievances well beyond what the policy change itself might have generated.
The reliability dimension of the backlash is distinct but intersecting. Beyond the policy dispute, users surface concrete technical frustrations: memory leaks in terminal environments running multiple concurrent Claude Code sessions, inadequate UX for managing multi-agent workflows, and the absence of features like image insertion in Claude Cowork. These complaints align with findings from Anthropic's own research: a large-scale survey of over 80,000 Claude.ai users across 159 countries identified AI error-making — not job displacement — as users' primary concern. Anthropic's internal data simultaneously shows progress, with human intervention rates per task falling 33% over six months and autonomous task completion roughly doubling, yet the gap between internal benchmarks and the lived experience of power users remains tangible.
The competitive context sharpens the stakes considerably. Several users in the thread explicitly cite the policy change as the moment they shifted internal project infrastructure away from Anthropic's ecosystem, with one noting a formal ban on Claude.ai subscriptions in favor of Gemma, Qwen, and OpenAI. Open-weight models are invoked not as aspirational alternatives but as practical present-day replacements — a notable shift from the conventional wisdom that Claude's reasoning capabilities maintained a meaningful lead. Anthropic's challenge is that its most vocal and technically sophisticated users are also its most capable of migrating, and their public statements function as social proof that accelerates churn among the broader developer community. The company's framing — that current models are "the worst today that they'll ever be" — is intended as a promise of continuous improvement, but in the short term it reads to frustrated users as an admission of current inadequacy rather than a credible near-term remedy.
The episode illustrates a structural tension that increasingly defines frontier AI companies: the economics of subscription pricing versus API consumption make it difficult to serve both casual users and power developers under the same plan tier, yet attempting to stratify those populations through policy enforcement generates exactly the kind of high-visibility developer anger that can reshape competitive positioning quickly. Anthropic's public response — a brief acknowledgment promising reliability and usability work — reflects a communication posture calibrated for damage containment rather than strategic clarity. Whether the company can translate its documented internal productivity gains and reduced intervention rates into a developer-facing value proposition that neutralizes the current discontent will be a meaningful test of its ability to retain the technical community that has, until recently, been among its strongest advocates.
Read original article →