The author describes a personal workflow system designed to mitigate "AI Brain Fry" — the cognitive overload caused by juggling multiple concurrent AI coding agent sessions (Claude Code, Codex) in the terminal. The core problem is identified as broken sequencing and context management when developers switch between dozens of threads. The solution models work as a priority queue persisted in Git: every task is a Markdown file co-located with relevant code, and the board state is a committed snapshot, eliminating reliance on chat history or human memory. An autonomous agent cycles through the queue, researching new tickets, building those tagged safe, running tests and browser checks, and verifying deployments; sensitive changes trigger branch creation and human approval requests, but the agent never deploys to production. Decisions requiring human judgment — reviews, clarifications, tradeoffs — are aggregated into a single "owner inbox" rendered as a Claude Artifact, rebuilt from Git each cycle to show the full board and per-ticket review pages. The human processes one item at a time: loads context, decides or acts, then closes the ticket by returning it to the queue, freeing mental RAM. Key tooling challenges include serving digestible context (targeting minute-scale load times) and preventing tickets from stalling in blocked loops. The system remains iterative but reportedly yields a significant productivity gain with AI tools.