Developer cancels Claude subscription over token limits and copy-paste support

A long-time developer who built AI deeply into their workflow documents the reasons for cancelling a Claude Pro subscription after several weeks of use. Initial enthusiasm faded as token allowances seemed to deplete unusually fast, including a case where two trivial Haiku queries spiked usage to 100 percent after a ten-hour idle period. A support ticket about the spike was answered with what appeared to be a templated response about Pro plan usage limits, then closed with a notice that further replies might not be monitored.

The author also flags experience regressions: a previously-undisclosed monthly usage warning appearing without explanation, the weekly reset window shifting unannounced, and conversation cache loss forcing repeated codebase reloads that bill tokens twice for the same context. Quality complaints centre on Claude Opus proposing a workaround during a refactor, including a generic JSX initializer in ui-events.js to avoid editing each slider, which the model conceded was lazy when challenged. The author acknowledges quality perception is subjective and operator-dependent.

The broader argument is structural rather than a pure quality rant: Anthropic appears to be making design and capacity trade-offs, including opaque limit tiers and templated support, that erode trust with paying technical users. The author notes Claude Code still ships working software and praises features like Claude Cowork, but concludes the support and metering experience no longer justifies the subscription.