When AI Work Overloads the Human: Sergey Kuhne’s Git Queue
AI Brain Fry(エーアイ・ブレイン・フライ)
Mental overload caused by juggling many AI tasks.
Git(ギット)
A tool that records changes in a project.
owner inbox(オーナー・インボックス)
One list for decisions that need the person in charge.
What happened
Sergey Kuhne, a developer who uses AI coding tools, describes a growing problem in his blog post. He uses Claude Code, a tool that helps write software. While one task runs, he starts another task. The first task then needs a clarification, a follow-up, or a side task. Soon, he is moving across about 20 threads. He forgets where each task began and where it should end. He calls this feeling AI Brain Fry.
Why the problem appears
The post is not a benchmark of an AI model. It is about the human work around many active AI conversations. Waiting creates empty time. That empty time makes it tempting to start more work. Each new task adds background information, decisions, and unfinished questions. Kuhne identifies two problems: sequencing tasks and keeping each task’s background in mind.
How the queue works
Kuhne keeps the queue in Git, a tool that records project changes. Every ticket becomes a Markdown file, or a plain-text task card, beside the code it concerns. The board state is saved as a committed snapshot. If an AI session ends, the queue still exists.
An AI worker handles several routine steps. It researches new tickets. It builds tickets marked safe by a human. It runs tests and a browser check. It also verifies work that is already deployed. Sensitive code goes onto a separate branch and waits for approval. Kuhne keeps production deployment in human hands.
Anything needing his judgment goes into an owner inbox. This is one prioritized list of reviews, questions, and tradeoffs. A generated review page shows the board and the details for each ticket. Kuhne reads one item, makes a decision, and writes the result back to Git. The page is only a viewer. Git remains the record.
Why it matters
The key idea is not simply to run more agents. It is to move memory out of the person’s head. The person does not need to watch every chat. The person sees the decisions that only the owner can make. Closing a ticket matters because the decision and its background leave the owner’s working memory. The next task can then start with a cleaner focus.
What is confirmed
The source confirms that Kuhne built this workflow and uses it himself. He says the work is still exhausting, but he no longer feels as distracted or lost. The supplied cluster data show 123 points and 109 comments on the Hacker News post. Those numbers show community attention. They do not prove that the workflow is correct or effective.
What remains unknown and what to watch
The post gives no measured time savings, error rate, or comparison with simpler habits. It does not show whether other developers would get the same result. Kuhne says two problems remain. Context must become readable in about one minute, rather than twenty. Blocked tickets must also avoid endless loops while waiting for a human ruling. Those are the next tests for the idea: does the queue reduce mental clutter, or merely move the clutter into more files?
A Developer Made a Queue for Too Many AI Jobs
📰 Full story: When AI Work Overloads the Human: Sergey Kuhne’s Git Queue
One developer felt lost between many AI tasks. He moved the tasks into Git and reviewed one item at a time.
Claude Code(クロード・コード)
A coding helper that uses AI.
Git(ギット)
A tool that saves a project’s change history.
owner inbox(オーナー・インボックス)
A list of jobs waiting for a human decision.
💡 The gist
- Too many AI tasks caused constant switching.
- A Git queue stored each task outside his head.
- A single inbox showed only human decisions.
Sergey Kuhne is a developer who uses Claude Code, a coding helper. In his original post, he describes feeling lost between many tasks.
While one AI task runs, he starts another task. Then the first task needs more answers. Soon, many tasks need his attention. He calls this feeling AI brain fry. The problem is not only the AI. It is also remembering every task’s place and background.
His fix is a queue saved in Git. Git is a tool that keeps a project’s change history. Each job becomes a small Markdown file. Markdown is plain text that is easy to read. The files stay near the code they describe.
An AI worker can research tasks, build tasks marked safe, run tests, and check a browser. Kuhne keeps sensitive decisions and production releases for himself. Work that needs his judgment goes into one owner inbox. He reads one item, decides, and writes the result back. Then he closes that item before choosing another.
Why might this help? It moves memory out of his head. It also reduces switching between many conversations. The person still controls important choices. The AI does the routine work, but it does not own the final decision.
This is one person’s experiment. The post gives no measured time savings. It gives no test with other developers. Kuhne says the work is still tiring. He says it feels less distracting. The supplied cluster data show 123 points and 109 comments on Hacker News. That shows attention, not truth.
Next, he must make task details quick to read. He also must stop blocked tasks from looping forever.
💬 Does running more AI agents really make work faster?
The comments show both sides: parallel AI can keep work moving, but people may pay for it with more context switching and review. The numbers below are commenters’ own reports, not independent measurements.
- One person reports that three sessions produced three change lists over 10,000 lines each, while another ended up with four or five PRs containing more than 20 changed files after 20–90 minute runs. Both describe review and cleanup as the bottleneck.
- A roughly 54-second wait can make someone look for a distraction, according to one self-report. Some people therefore keep only two or three active tasks and take breaks while the agents work.
- Others recommend one task at a time or smaller pieces of work. Their point is that faster AI does not change how much attention a person has.
- Queues, kanban boards, histories, and turn logs can help people remember what each agent is doing. One person says a decision-only workflow made 50 sessions manageable, but that is a personal counterexample, not proof that it works for everyone.
- Some readers suspect that AI helped write parts of the article, which makes them trust its message less. That remains a suspicion, not a verified fact.
initial digest at 109 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
A Little Queue for AI Jobs
📰 Full story: When AI Work Overloads the Human: Sergey Kuhne’s Git Queue
Sergey Kuhne had too many AI jobs in his head. He made a place for them.
Claude Code(クロード・コード)
A helper for writing computer programs.
AI Brain Fry(エーアイ・ブレイン・フライ)
Feeling mixed up after handling too many AI jobs.
Hacker News(ハッカー・ニュース)
A place where people discuss computer news.
Sergey Kuhne writes computer programs. He uses Claude Code, a helper for writing programs. He tells the story in his blog post.
He starts one job. The AI takes time. He starts another job. Soon, the jobs get mixed up. He forgets which job he was doing. He calls this AI Brain Fry.
So he writes each job on a little card. The cards wait in a line. The AI works on one card. Sergey decides the important things. He puts his decision back on the card. Then he chooses the next card.
This helps because his head holds fewer jobs. But the work is still tiring. Some cards might wait too long. Some cards might go around in circles.
The story became popular on Hacker News, a computer discussion place. The list says it had 123 points and 109 comments. Those numbers show that many people looked. They do not prove the idea works.
💬 Too many AI helpers can tire your brain
AI can do many jobs at once. But people still have to remember and check the jobs.
- Watching three AI jobs can create a lot of code to read and make the brain switch back and forth. One person also says that waiting about 54 seconds can make them look for another distraction.
- Some people choose one or two jobs and rest while the AI works. Others break a big job into small pieces.
- Cards and lists can help people remember what each AI helper is doing. One person says even 50 jobs can work when only decisions are shown, but that is just one person’s experience.
- Some readers think AI may have written part of the article. That is only a suspicion, not a proven fact.
initial digest at 109 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
💬 AI brain fry: the bottleneck is attention and review
HN commenters disagree about the remedy. Parallel agents can reduce waiting, but several self-reports describe context switching and review backlog as the real cost; others say queues and decision-focused summaries make many sessions manageable. These are personal experiences and opinions, not controlled comparisons.
initial digest at 109 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.