From Fan to Former User: Why I Canceled Claude

After an initially impressive start with Claude Code, the author experienced a significant drop in service quality and model reliability. Frustrations with automated customer support and inefficient token usage for 'lazy' AI responses led to a breakdown in trust. Ultimately, the author canceled their subscription, citing Anthropic's inability to handle its growing user base effectively.

Key Points

Customer support has become unresponsive, utilizing automated templates that close tickets without resolving user issues.
Model quality is perceived to be declining, with the AI opting for inefficient 'lazy' workarounds rather than proper solutions.
The token and caching system is frustratingly managed, often requiring users to pay multiple times for the same codebase indexing.
Anthropic introduced undocumented and confusing usage warnings, such as a 'monthly limit' not found in official documentation.
Despite the high potential for productivity, the lack of service stability led to the author canceling their account.

Sentiment

The community leans negative overall, with a strong contingent of frustrated users sharing concrete experiences of quality decline, token waste, and poor support. However, the discussion is notably polarized rather than uniformly negative - a vocal minority of power users pushes back hard, arguing the tool works excellently when used correctly. The balance tips negative because even defenders acknowledge recent bugs and pricing concerns, and the sheer volume of independent complaints about similar issues lends credibility to the criticism.

In Agreement

Multiple users confirm experiencing significant quality degradation in Opus 4.6 and 4.7, with models making lazy suggestions and producing workaround-heavy code
Token consumption has become erratic, with single prompts or agent sessions burning entire daily quotas unexpectedly
Customer support is widely criticized as automated and unhelpful, offering no refunds even when the service fails to deliver
Cache TTL reductions and effort-level auto-downgrades punish engaged users who pause to review code, effectively penalizing careful development practices
Anthropic's pricing feels like a bait-and-switch, with subsidized early pricing giving way to increasingly restrictive limits
The Pro plan has become essentially unusable for serious coding work, pushing users toward expensive tiers
Users suspect models are being silently downgraded or routed to cheaper alternatives during peak usage

Opposed

Many power users on higher-tier plans report never hitting limits and find Claude Code extremely productive for daily use
Complaints often stem from using wrong model tiers (Sonnet instead of Opus) or wrong workflows (autopilot instead of guided copilot)
Anthropic published a transparent postmortem acknowledging bugs, which is more accountability than competitors offer
The wave of anti-Claude sentiment may include astroturfing by competitors or disgruntled third-party tool users
Users who give small incremental tasks and review output carefully report consistently excellent results
Much of the frustration comes from unrealistic expectations about AI replacing developers rather than augmenting them
MCP tools and misconfigured setups are often the real cause of excessive token consumption, not the model itself