Multiple users report GitHub Copilot is 'slowly getting worse' with quality degradation across all models. Suggestions that used to be helpful now produce incorrect, hallucinated, or irrelevant code. Some describe it as 'unusable' compared to earlier versions. GitHub community discussions show widespread frustration.

GitHub Copilot has been slowing down steadily. Prompts that used to be answered immediately now take much longer - some users report having to 'wait for a coffee' before getting responses. Auto-completion feels sluggish compared to earlier versions.

Premium request quotas introduced in 2025 generate massive frustration. Using GPT-4.5 costs 50x multiplier (one interaction = 50 requests). Users report hitting limits in 6 days despite normal usage. Overages cost $0.04/request - one user estimated $40/month for moderate use. Pro+ costs $39/month but still has limits.

Pro+ at $39/month is steep for individual developers. The model multiplier system makes advanced models practically unusable without upgrading. Some users find ChatGPT provides similar capabilities for free, questioning Copilot's value proposition.

Copilot struggles with context in larger codebases. It makes assumptions based on limited file context, often ignoring project-wide patterns. Generated documentation contained 60% speculative content including fabricated API details. After corrections, 30% still remained speculative.

Agent Mode repeatedly edits code despite being asked not to. Using models like Claude Sonnet 3.7, the agent continuously modifies code through many iterations, adding 'nothing but junk' while generating warnings. Users report having to undo AI changes constantly.

Copilot is hit or miss - sometimes brilliant, sometimes useless. It may provide irrelevant code for your context or simply not work at all. The inconsistency makes it unreliable as a productivity tool since you never know if the next suggestion will help or waste time.

Copilot sometimes generates generic error handling that makes debugging harder. Code may not match your project's types and schemas. When issues occur, tracing back through AI-generated code adds complexity to troubleshooting.

Ongoing class-action lawsuit alleges Copilot reproduces licensed code without attribution. GitHub admits ~1% of suggestions may contain 150+ character matches to training data. Using Copilot-generated code may expose projects to license violations. Legal risk remains unclear.

Users report waiting weeks for support replies. When help arrives, it's often irrelevant or outdated information. Billing issues go unresolved, and some users describe unauthorized charges with no recourse. Enterprise customization requests face long delays.

Getting good results from Copilot requires learning specific prompting patterns. New users struggle to understand when suggestions are helpful vs harmful. The tool can make beginners 'lazy' and miss important learning milestones in coding education.

Some users report Copilot hanging or becoming unresponsive shortly after payment. Complaints include service issues that support doesn't address, leaving paying customers unable to use the product they paid for.