Chat GPT and Claude can confidently skip half of your document and never tell you. If you're new here, I'm Dylan. I run an AI consultancy and this silent failure shows up in coaching sessions across nearly every industry I work with. You upload a long file into Chat GPT or Claude. You get a clean confident answer and you assume the model read everything. It usually didn't. I'll show you why this breaks down in a three chat workflow I use in desktop agents like Codex or Claude Cowork to work around this weakness. So, let's get into it. Now, this is the battle that we're all facing and it's something that we're likely going to continue facing until it's fundamentally fixed by the research labs like OpenAI, Google, and Anthropic. And that fundamental challenge is how much space is in AI's head and how much information we can put in there before its intelligence drops. And every single one of my clients that are doing meaningful work with AI hit this limit. And this manifests in three ways, especially if you're using the browser AIs. So, if you're using Chat GPT in the browser or Claude in the browser. The first one is you just give the AI a really large file, something that exceeds maybe 800 to 1,000 pages. When doing this, the AI may error out and say that it can't handle it. And the next one is if you give too many files to the AI. So, this could be Claude or GPT. If you give more than 10 files, it'll likely prevent you from putting any more files in. You can only do 10 at a time. And I'll talk a bit more about this later, but depending on the size of the file, you may be only get up to like three files or four files. But the thing here is these two are actually good use cases because the AI explicitly told you that it couldn't handle it. But the worst case, which happens way more often than you would expect, is the AI takes all of it and it gives you an answer back. But the issue is that answer, the AI only partially read the documents you provided, so the answer is likely only partially true. And this is the big thing that we want to avoid. I'm going to walk you through a three chat workflow that I use with Codex and Claude Cowork on my desktop to mitigate many of the failures that we see here. But before we get into that chat workflow, I want to walk you through more why this is such a problem. So, you've probably seen this chart many times in my previous videos. And what this chart shows is the drop-off of intelligence when you fill up the AI's head. So, on one axis, we have intelligence, low and high. On the other axis, we have how much space we filled up in the AI's head. Once you fill the AI's head up to around 50 to 60% depending on the model you're using, the intelligence drops like a rock. So, it's important to be aware of this so you don't necessarily have a long-running conversation that's too long or you put too many files in the AI's head in the wrong way where it tries to hold all of it in one spot and then answer your question at the same time. By doing that, you're going to get probably a poor answer. Quick pause in your regular programming. This video is brought to you by me, as always. Two quick things. First off, below is a 30-day AI insight series. You'll get 30 insights in your inbox for how you can apply AI to your business and your work. The second thing is if you'd like to work with me, below are a series of offerings to see if there's a good fit between the two of us. Now, let's get back to the video. And another important thing to call out here, it's not all file types are created equal. Some files are much smaller than others. So, if you have a plain text file, even if it's a thousand pages, if it's plain text, it's much smaller on the AI's head and it can hold that effectively. And as we go up these file types, it fills up the AI's head faster. So, if it's a clean PDF, that's okay. If it's a slide deck like a PowerPoint, that can fill up the AI's head pretty quickly depending on the complexity of the slides and how many there are. If it's a spreadsheet that has many columns, rows, and tabs, that again can fill up the AI's head really quickly. And videos are on the high end where if you put this in there, it probably can only hold maybe between a minute and 10 minutes of video depending on the model you're using. So, it's not just how many files, but also the type of file you're sharing with AI. And as I mentioned, different models can hold different amounts of information based off of the model you're using and also where you're using it as well, which is interesting. So, for example, if I'm using Opus 4.7, if I use it inside of Claude Cowork on the desktop agent, I likely have around 200,000 tokens that I can use. And a token basically is a part of a word. And if I'm using Opus 4.7 from Claude inside of Claude Code, I can switch on the ability to use a million tokens. So, it's not just the model you're using, but also where you're using that model. And then GPT 5.5, if you use t
