An engineer emphatically explains 'the reasons why I canceled my Claude subscription,' citing token issues, declining quality, and inadequate support.
German engineer
Why I Canceled Claude: Token Issues, Declining Quality, and Poor Support - Nicky Reinert
https://nickyreinert.de/en/2026/2026-04-24-claude-critics/
Recalling his initial experience using the coding assistant Claude Code , Reinert says, 'The first few weeks were very pleasant.' It was fast, the token rewards were reasonable, and the quality was good. However, Reinert says that the initial enthusiasm 'quickly faded.'
The incident began one morning when I tried to use Claude Code. Although enough time should have passed for the token usage limit to refresh, after sending two simple questions to Claude Haiku (a lightweight model), the token usage rate suddenly jumped to 100%.
Mr. Reinert contacted the AI support bot, but only received a canned support message, which did not resolve his problem. He then requested to speak with a human support representative, and it took several days before he finally received a reply from one of them.
The response from the human support representative was, 'Our system has detected that your inquiry concerns the usage limits of the Pro or Max plan.' While this was true since Mr. Reinert was using the Claude Pro plan , this response was clearly a default response message for the inquiry.
Next came a lengthy response that appeared to be copied and pasted from Anthropic's official documentation, which seemed to describe the mechanisms for 'per day' and 'per week' token limits. Finally, at the end of the email was the typical irritating sentence that customers would not want to read at the end of an email: 'It doesn't matter if your problem is solved. We have marked this matter as resolved.'
Regarding this, Reinert wrote, 'To send an automated email that doesn't address the actual problem at all, and then abruptly end the conversation... Does that mean they don't appreciate anything? Or was I wrong in my thinking?'
Initially, Reinert was able to work on up to three projects using Claude Code. However, now he says that he reaches the token limit after spending just two hours on a single project.
Reinert pointed out the decline in Claude's quality, stating, 'Quality degradation is quite subjective, and we are well aware that the quality of agent output is always heavily dependent on the operator, and failures will show up on the screen.'
Furthermore, as a concrete example of the decline in quality, when Claude Opus was asked to refactor a project that was under development, he 'took a simplistic workaround that even a junior developer wouldn't have thought of.' When Reinert asked Claude Opus about this, saying, 'Are you kidding me? Is this how you solve the problem? Just a quick fix??' he received the reply, 'You're right, that was a shortcut. I'll redo it properly.'
Reinhard commented on this, saying, 'At least it's honest, but it's the epitome of a lazy developer.' He also noted that this exchange consumed 50% of his daily available tokens.
Furthermore, Claude also has an issue where the conversation cache gets erased, causing it to start reloading the codebase again. Regarding this, Reinert wrote, 'It's a smart decision for Anthropic's revenue, but what about the user experience? You're using tokens to load it initially, and then after the token window reaches its limit and is forcibly interrupted, you have to pay to load the same thing again.'
Reinert wrote, 'I'm a big fan of Claude Code. In theory, everything works perfectly and it offers a lot of possibilities. I respect Claude Code for working behind the scenes on so many of GitHub's issues. I love writing using Claude Cowork.' He continued, 'I understand the technical and organizational challenges of delivering this kind of product. When you sell inference, it's not easy to reap the benefits of scaling. You need extra time and the same amount of computing power for each new customer. That's the downside of incremental costs in this industry. It seems Anthropic can't handle a lot of new customers at once, so I closed my account to ease the burden.'
Reinert added, 'If you don't read to the end, it might sound like I'm complaining about Claude's overall quality. However, my concerns are focused on the lack of support and the token issue. I fully understand the challenges a company of this size faces, and I know Anthropic is working hard to resolve the issues. What I'm pointing out is probably a 'design flaw' that Anthropic may have made. The quality issues are merely an afterthought.'
Claude Code is functioning correctly, and Reinert mentioned that he has used it to complete various tasks. However, he pointed out that there has been a clear decline in quality, and that 'it simply takes longer to process.' Reinert acknowledged that this is purely his subjective opinion, saying, 'Of course, I understand that the performance of an AI agent depends on the operator and their instructions.'
Related Posts:
in AI, Posted by logu_ii