Why Your AI Gets Dumber the Longer You Talk to It (And how to fix it)

If you use AI every day, you may have noticed a phenomenon:
When you talk to it for too long in a single conversation, it gets significantly dumber.

This phenomenon is called context rot.
It happens when the AI takes in too much context and uses too many tokens in one conversation, so it starts to forget.
More precisely, it usually forgets what's in the middle of the conversation.
It usually remembers the beginning and the end well, but the middle gets lost, and it loses a lot of context.

So when you use AI, it's really important to avoid having too long a conversation with it.
Because the longer the conversation, the worse the AI gets and the worse the results.
And this is a mistake I see most people make:
Doing everything in one chat, or running several tasks in a row in the same session with an AI agent.

There are a few ways to fix this.
The first is to get into the habit of using one chat, or one session with your agent, for a single task.
If the task is big, like creating a course, divide it into smaller ones, for example one session per module instead of doing the whole course in a single session.

The other important thing, especially with AI agents, is to monitor the context window you've used, and so the number of tokens you've used.
If you use Claude, you'll see the models can have either 200,000 of context, or up to 1 million.
Whatever the size, you never want to go past 100,000 tokens.
Past 100,000 tokens, the results start getting worse and context rot starts to appear.
So even if the AI says it can hold 200,000 or 1 million, past 100,000 it will start forgetting things and getting less and less relevant.

So what I recommend is to monitor your context regularly.
Make sure you never go past 100k tokens.
And when you hit 100,000, it's simple:
Clear your session, or open a new one in your terminal, and start again from scratch.
That's why it's so important to monitor that window and to divide your task the right way, so your work always stays inside it.
Again, if you're working on a big program or a big task, you can't do everything in one session and go past that context window.

And if you get into the habit of monitoring your context window with an AI agent, you usually just run the /context command to see it.
I also use a plugin called Status Line that shows you the context window in real time, so it's really useful for this.
I recommend you install it too.

Here's the install command:
npx ccstatusline@latest
You run it in your terminal, it installs for Claude Code, and then in the settings you just select the context used among the options you want to display.
You can display plenty of options.

But it's super important to monitor this.
And really, if you tend to always use AI in a single session or window, for example creating all your newsletters in the same session, I really recommend you stop doing that now.
Because the AI will be much less relevant than if you use one session or chat window per task or subtask.

So that was my tip of the day, and it's an important one to put in place.
(And of course, if you're a Sovereign Creator OS Pro member, you've probably already learned it in the module on managing context like a professional, in the AI agents section)

That's all for today's email.
Thanks for reading...
And welcome back to the Ark.
Noah.

P.S. I currently have open spots for one-on-one coaching calls. If you want me to support you in building your own AI second brain system and developing your creator business with it, including all the strategy, the monetization, and the systems to set up, then reply "second brain" to this email, and I'll send you the details of my coaching.