How I Fight AI Brain Fry
- Source
- sergeykuhne.com
- Author
- kyunez
- Date
Context
Sergey Kuhne's answer to what he calls AI brain fry is to stop keeping work in his head and in open chat tabs, and to run it through a queue that hands him one decision at a time. In a blog post, he describes the problem for people who run coding agents such as Claude Code or Codex from a terminal. While one agent works, you start another task; the first reply needs a clarification and a follow-up, each wanting its own tab; soon you are hopping among 20 threads and lose track of where each began. Rather than the common advice to take breaks or limit scope, he treats it as two problems, the order in which work gets done and the each item needs.
In his setup, every ticket is a markdown file committed to the git repository next to the code it concerns, so nothing important lives only in a chat window and the queue survives if a session dies. An agent works through the board in cycles: it researches new tickets, builds the ones a human has marked safe to build, runs the tests and a browser check, and verifies what is already deployed. It builds changes to sensitive code on a branch and asks first, and it never deploys to production, which Kuhne keeps for himself. Anything that needs his judgment, such as reviewing a change, answering a question or ruling on a tradeoff, a script gathers into one prioritized list he calls his owner inbox. He reads it on a single Claude artifact page, rebuilt from git each cycle with a review page per ticket, and his verdict is written back onto the ticket in git. The page only displays; git holds the record.
The step he credits most is closing each item. He takes one, loads its context, makes a decision or requests an action, and puts it back in the queue, so it stops occupying his memory and he can give the next item his full . He says the work is still exhausting but he no longer feels distracted or lost. This is one developer's account of his own workflow, not a measured result, and he names what still does not work: serving each ticket's context so that loading it takes a minute rather than twenty, and keeping tickets from looping while they wait on a ruling only he can give.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
Checking sign-in…
Loading comments…

