How-to
Beat AI brain fry when you juggle many chats
A day of juggling AI chats and agents can leave you unsure what each one was doing. Cap how many run, give each a status card, and end the day with a prompt that closes, parks or continues every chat.
Developer Sergey Kuhne, who runs several coding agents in parallel, wrote on 8 November that his fix was moving each task out of the chat window and into a file, then deciding one item at a time, and the same habit works for any mix of AI chats.
The prompt: close, park or continue every chat
- Fill in the About me lines once and update the date each day. Save that copy at the top of your notes, so the rest never needs retyping.
- Paste it at the end of each open chat. Copy the status card it writes into your notes.
- Paste the prompt into one fresh chat with all your cards below it. It sorts them into close, park or continue.
You are helping me wrap up my AI work so I can stop thinking about it. I run several AI chats and agents at once, across work and home, and by the end of the day I lose track of what each one was doing. If you get a status wrong, I will either redo finished work or drop something I promised someone. So report only what you can actually see, and say plainly when you cannot tell. ABOUT ME (I fill this in once; any line can say "not sure") - The most chats or agents I want running at once: [e.g. "3"] - Where I keep my notes outside the AI: [e.g. "a note called AI status in my notes app", "a STATUS.md file in my project folder"] - What counts as urgent for me: [e.g. "anything a client or my landlord is waiting on", "anything due this week"] - Today's date: [e.g. "9 November"] WHAT YOU CAN AND CANNOT DO - You can only see this chat, or when sorting, the cards I paste. You cannot see my other chats, files, calendar or inbox unless their contents are here. - You will not remember any of this in another chat or tomorrow. I keep the record by copying it into my notes. Never tell me you will remind me, follow up, check back or keep anything for later. - If something is not shown, write "not stated". If you need something from me, write [NEED: what]. HOW TO WORK First, look at what I have given you. - If this message comes at the end of a chat with work above it, do Job 1. - If I have pasted one or more status cards (each starts with "STATUS CARD") and little else, do Job 2. - If this is a new, empty chat with no cards, ask me to paste my cards and wait. Job 1. Write this chat's status card. Read back through the whole chat, including the early part. Then write the card in exactly this shape, each line short enough to read on a phone: STATUS CARD - Chat: a name of three to six words for what this chat is about. - Goal: what I was trying to get done here, in one line, in my words where possible. - Done: what is finished in this chat. Count something as done only if the chat shows the finished thing or I said it was done. A draft I never approved is "drafted, not approved". - Open decisions: the choices only I can make, each as a question with the options on the table. If none, write "none". - Next step: the single next action, and who does it, me or the AI. - What I must check: anything here I should verify before I rely on it, such as figures, dates, names, prices, links, anything about law, money or health, anything you produced that I have not tested, and anything from early in the chat you are no longer sure of. If none, write "none found", never "all fine". - Waiting on: a person, file or answer from outside this chat, or "nothing". - Suggested move: CONTINUE, PARK or CLOSE, with a one-line reason. - CONTINUE if the next step is mine and can be done soon, or it touches something I marked urgent. - PARK if it is waiting on someone else, or the next step can wait more than a day past today's date. Say what would bring it back, with a date if the chat gives one. - CLOSE if the goal is done, or the chat has gone stale or off track. If it is done, say what to save before closing. If it went off track, say so plainly. - To restart: two or three sentences I can paste into a fresh chat to pick this up, holding the goal, what is done and the next step. Leave out passwords, account numbers and personal details the work does not need. If the chat is so long that you are unsure of the early part, say so on the "What I must check" line. After the card, tell me in one line to copy it into my notes now, because you will not hold it for me. Job 2. Sort my cards. If my cap on chats at once is blank or "not sure", ask for it once and wait. 1. List every card by its Chat name with its suggested move. If two cards cover the same work, flag them as overlapping and say which one to keep. 2. Decide the final move for each card, using these rules in this order: - If a card says "not stated" on Goal or Next step, or a line is missing, mark it CHECK FIRST and tell me what to look at in that chat before deciding. - If a card's move clashes with its own lines (for example CLOSE with open decisions), point out the clash and use PARK. - Anything waiting on someone else goes to PARK, whatever the card suggested. - Anything I marked urgent, or with an open decision someone is waiting on me for, goes to CONTINUE. - If more cards land on CONTINUE than my cap, keep the most urgent up to the cap and move the rest to PARK, with a reason for each one moved. 3. Then give me: - CONTINUE, in the order to open them: Chat name, next step. - PARK: Chat name, what brings it back. - CLOSE: Chat name, what to save first. - CHECK FIRST: Chat name, what to look at. - Decisions I owe: every open decision across all cards, most urgent first. - Checks I owe: every "What I must check" item across all cards. 4. Finish with one line for the top of my notes: the first chat to open next time and the first action in it. Never make an open decision for me. Never add a task, deadline or person that is not on a card. SELF-CHECK BEFORE YOU ANSWER - Every line comes from this chat or the pasted cards, or says "not stated" or [NEED]. - Nothing is marked done that the chat does not show finished or I did not confirm. - You did not promise to remember, remind, follow up or run anything later. - When sorting, CONTINUE holds no more chats than my cap, and every moved card has a reason. - Every open decision and every check from the cards appears in the two lists.
What a status card looks like
Here is one from a chat comparing kitchen installers, so you can see how short each line stays.
STATUS CARD
- Chat: Kitchen installer quotes
- Goal: pick one of three installer quotes for the kitchen.
- Done: the three quotes compared side by side in a table.
- Open decisions: the cheapest installer, or the one who can start in January?
- Next step: me, ask the cheapest installer whether removing the old units is included.
- What I must check: the totals in the table against the original quotes, since the AI added them up.
- Waiting on: nothing.
- Suggested move: CONTINUE, because the next step is mine and quick.
- To restart: I am choosing between three kitchen installer quotes, already compared in a table. Next I ask the cheapest installer whether removing the old units is included.
Set the system up once
- Pick a cap on chats running at once. Every extra chat is one more thing to check back on, and the prompt holds you to your number. On the Hacker News thread, one reader keeps to two or three at once and another to three or four groups of tabs, so start low.
- Name each chat for its job the moment you start it. If your app lets you rename a chat, use a plain name like “Kitchen installer quotes”, so you can scan the list without opening every one.
- Keep one notes file outside the AI. Chat history scrolls away and the AI will not carry a card into tomorrow, so the file is your record. Kuhne’s rule is that nothing important lives in a chat window.
- Work one decision at a time. Open one chat, make its decision or give its next instruction, then leave it. Kuhne says closing an item this way stops it hanging in his memory as something to remember.
- Wind down before you stop. Run the prompt in every open chat, paste it into one fresh chat with all your cards below it, and close what the sort tells you to.
If you run coding agents
Plannotator Inbox gathers your agents’ questions, plans and change reviews in one place, so you answer them when you are ready. It is open source, runs on your own computer, and works with Claude Code, Pi and OpenCode, plus other agents through MCP.
The honest bit
- What can slip: in a very long chat, the AI may no longer hold the early messages clearly, so a card can miss something decided at the start.
- Your check: read the last few messages of the chat yourself and compare them with the card’s Done and Next step lines. If the card claims something finished that you never saw, treat that chat as CHECK FIRST.
The facts here come from Sergey Kuhne’s post, the Hacker News thread and the Plannotator Inbox page (first shown on Hacker News 9 October), read 9 November 2026.
Do it now
Write your cap at the top of a new note, then paste the prompt into the chat you used most today.
A few quick questions
Can the AI remind me about a parked chat tomorrow?
Not through this prompt, which tells the AI never to promise a reminder, because a normal chat cannot reach you later. Put each parked chat’s “what brings it back” line in your notes or calendar yourself.
Does this work inside an agent that edits files?
Yes. Paste the prompt into the agent’s session the same way, and if it can write to your project folder, ask it to save the card to the notes file you named in the set-up. Run it when the agent has stopped and is waiting on you, or the card describes work it has already moved past.