A longtime Claude Code user has decided to retire it as their main coding harness and move their development setup to Pi. The decision came after an accumulation of small frustrations, and the author notes that Anthropic announced AGENTS.md support just a day after the choice was made, which was too little, too late.
The author's biggest complaints center on the VSCode integration. Only one session can be connected to the IDE at a time, so text selections are not received by a second tab working on another task. The official GUI extension allows multiple tabs, but it lacks support for half of the built-in slash commands, has no custom status line, and mixes subagent output into the main chat thread in a confusing way.
The author also criticizes the pace of development. Claude Code saw 244 releases this year, with over three quarters changing the system prompt or tool descriptions. Those changes can alter how the model performs, especially when they conflict with user prompts or repository conventions. The default Bash tool description alone runs almost 3,000 tokens, which the author sees as unnecessary overhead compared with Pi.
The post closes on a more measured note. The author acknowledges that Anthropic has built a capable product and that the criticism reflects a personal workflow rather than a universal verdict.
Source: Hacker News · Summarized by HeadlinesBriefing