Sergey Kuhne describes how terminal AI tools like Claude Code and Codex cause "AI Brain Fry" through context-switching overload. He proposes organizing work like a priority queue using git, where tickets are markdown files stored alongside code. This eliminates chat windows and mental RAM. An agent cycles through the board, researching and building tagged tickets, running tests and browser checks. For sensitive code, it creates branches and seeks human approval. Critical decisions land in a prioritized "owner inbox," where Kuhne reviews changes, answers questions, and rules on tradeoffs via a Claude artifact viewer. The system mirrors the whole board and per-ticket review pages. The key is closing one ticket before moving to the next, preventing context from lingering in memory. While still exhausting, the method reduces distraction by serving context digestibly and keeping git as the single source of truth. Kuhne views it as a significant productivity step forward, acknowledging it is a living organism he continues to improve.
The challenge remains in tooling that serves context in a digestible manner and prevents tickets from looping indefinitely when blocked by rulings only the human can give.
Source: Hacker News · Summarized by HeadlinesBriefing