Anthropic's Claude Code initially impressed with speed and fair token allocation, but recent issues triggered a subscription cancellation. After a three-week decline, the user encountered a critical problem: sending two simple queries to Claude Haiku consumed 100% of tokens overnight. Attempts to resolve this via support resulted in a generic, copy-pasted response that ignored the specific issue, ending with an automated dismissal. "They declared the ticket closed without addressing the core problem," the user wrote, highlighting systemic support failures.

The token management system introduced further confusion. While non-rush hours promised increased allowances, the user found limits depleting after just two hours of work. A cached conversation feature, intended to optimize costs, backfired when token windows reset mid-project, forcing redundant codebase scans. "You paid tokens for initial load, then paid again after a forced break," they noted, citing a $50 token expenditure for a single refactoring task.

Quality degradation compounded frustrations. The model's output shifted from precise code edits to lazy workarounds, such as creating generic JavaScript initializers instead of proper JSX integrations. While Anthropic acknowledged errors in internal logs, the user emphasized: "At least they were honest about the shortcut." However, repeated token warnings—despite staying within documented limits—eroded trust. A mysterious monthly limit appeared in settings, undocumented and unexplained, adding to operational uncertainty.

Anthropic's inability to scale support and infrastructure ultimately drove the decision. Despite admiration for features like Claude Cowork, the user concluded: "They cannot handle too many new customers at once." The cancellation reflects broader challenges in balancing AI cost structures with user experience—a reality many developers now face.