How I Fight AI Brain Rot. Friction Maxxing With Codex, Grok And Claude.

Friction maxing. That's what I do all Friction maxing. That's what I do all day with AI and it's why I don't think day with AI and it's why I don't think day with AI and it's why I don't think it's rotting my brain. Most AI use is it's rotting my brain. Most AI use is it's rotting my brain. Most AI use is friction removal. I totally get it and I friction removal. I totally get it and I friction removal. I totally get it and I talk about it a lot on this channel. You talk about it a lot on this channel. You talk about it a lot on this channel. You ask, you get an answer and you move on, ask, you get an answer and you move on, ask, you get an answer and you move on, right? It's fast. It's clean. It's done. right? It's fast. It's clean. It's done. right? It's fast. It's clean. It's done. But I run it back the other way, too. But I run it back the other way, too. But I run it back the other way, too. And that's what this video is about. I And that's what this video is about. I And that's what this video is about. I make the AI work harder on purpose. But make the AI work harder on purpose. But make the AI work harder on purpose. But every day I move between Codex and Grock every day I move between Codex and Grock every day I move between Codex and Grock and Claude and about 10 people that I and Claude and about 10 people that I and Claude and about 10 people that I trust. Not because I want all of those trust. Not because I want all of those trust. Not because I want all of those opinions for fun, but because I want to opinions for fun, but because I want to opinions for fun, but because I want to find the thing that breaks my find the thing that breaks my find the thing that breaks my assumptions, the thing that breaks the assumptions, the thing that breaks the assumptions, the thing that breaks the answer that everybody else is agreeing answer that everybody else is agreeing answer that everybody else is agreeing on. From the outside, this just looks on. From the outside, this just looks on. From the outside, this just looks like an elaborate system of passing the like an elaborate system of passing the like an elaborate system of passing the work around. But what I'm hunting for is work around. But what I'm hunting for is work around. But what I'm hunting for is disagreement. Every disagreement is like disagreement. Every disagreement is like disagreement. Every disagreement is like a rep for my brain. and the work comes a rep for my brain. and the work comes a rep for my brain. and the work comes out better because whatever survives out better because whatever survives out better because whatever survives four, five, 6, 10 rounds of argument is four, five, 6, 10 rounds of argument is four, five, 6, 10 rounds of argument is never what the model first handed to me, never what the model first handed to me, never what the model first handed to me, is it? Here is the version of that is it? Here is the version of that is it? Here is the version of that question that actually matters. After question that actually matters. After question that actually matters. After you use AI, do you feel more capable or you use AI, do you feel more capable or you use AI, do you feel more capable or less? Some of the panic runs ahead of less? Some of the panic runs ahead of less? Some of the panic runs ahead of the evidence. The MIT team behind your the evidence. The MIT team behind your the evidence. The MIT team behind your brain on chat GPT says its study is very brain on chat GPT says its study is very brain on chat GPT says its study is very preliminary and explicitly tells people preliminary and explicitly tells people preliminary and explicitly tells people not to call its findings brain rot. That not to call its findings brain rot. That not to call its findings brain rot. That has not stopped the internet, nor should has not stopped the internet, nor should has not stopped the internet, nor should we expect it to. People keep asking me we expect it to. People keep asking me we expect it to. People keep asking me if they can look over my shoulder, if they can look over my shoulder, if they can look over my shoulder, right, and see how I actually use AI.

right, and see how I actually use AI. right, and see how I actually use AI. And I want to take you inside that And I want to take you inside that And I want to take you inside that because for me, using AI well is because for me, using AI well is because for me, using AI well is constant mental exercise. It's like the constant mental exercise. It's like the constant mental exercise. It's like the opposite of being lazy. Every answer I opposite of being lazy. Every answer I opposite of being lazy. Every answer I get from AI pushes me and my human brain get from AI pushes me and my human brain get from AI pushes me and my human brain to create another choice. And so I want to create another choice. And so I want to create another choice. And so I want to walk you into how I do that so you to walk you into how I do that so you to walk you into how I do that so you too can friction max for AI so to speak. too can friction max for AI so to speak. too can friction max for AI so to speak. I have to ask myself when I use AI, do I I have to ask myself when I use AI, do I I have to ask myself when I use AI, do I accept it? Do I challenge it? Do I accept it? Do I challenge it? Do I accept it? Do I challenge it? Do I compare it with another model? Do I ask compare it with another model? Do I ask compare it with another model? Do I ask a person who knows the work or throw it a person who knows the work or throw it a person who knows the work or throw it away? I do that dozens of times a day. away? I do that dozens of times a day. away? I do that dozens of times a day. And yes, I'm going to walk into And yes, I'm going to walk into And yes, I'm going to walk into specifics with different models in this specifics with different models in this specifics with different models in this video and you're kind of going to get a video and you're kind of going to get a video and you're kind of going to get a sense of that and how I work. But every sense of that and how I work. But every sense of that and how I work. But every day I move between Codex and Grock and day I move between Codex and Grock and day I move between Codex and Grock and Claude and about 10 people I trust Claude and about 10 people I trust Claude and about 10 people I trust because I am deliberately trying to use because I am deliberately trying to use because I am deliberately trying to use AI to accelerate the way my own judgment AI to accelerate the way my own judgment AI to accelerate the way my own judgment forms. If you watched me for a day, the forms. If you watched me for a day, the forms. If you watched me for a day, the screen would just keep changing all the screen would just keep changing all the screen would just keep changing all the time. It'd have codeex up and then I time. It'd have codeex up and then I time. It'd have codeex up and then I would have Grock talking with Codex and would have Grock talking with Codex and would have Grock talking with Codex and I would have Claude helping me think I would have Claude helping me think I would have Claude helping me think through design and then my phone because through design and then my phone because through design and then my phone because a friend or a colleague has looked at a friend or a colleague has looked at a friend or a colleague has looked at what I'm building and told me what I'm what I'm building and told me what I'm what I'm building and told me what I'm missing. And from the outside, that can missing. And from the outside, that can missing. And from the outside, that can look like chaos, right? It can look like look like chaos, right? It can look like look like chaos, right? It can look like way too much. It can look like an way too much. It can look like an way too much. It can look like an elaborate system that just passes the elaborate system that just passes the elaborate system that just passes the buck down the line. But what I'm buck down the line. But what I'm buck down the line. But what I'm actually looking for is much simpler.

actually looking for is much simpler. actually looking for is much simpler. What I'm looking for is disagreement. What I'm looking for is disagreement. What I'm looking for is disagreement. I'm friction maxing. What does one model I'm friction maxing. What does one model I'm friction maxing. What does one model see that another one misses? Where does see that another one misses? Where does see that another one misses? Where does a human reaction break the answer that a human reaction break the answer that a human reaction break the answer that three different AIs found convincing? three different AIs found convincing? three different AIs found convincing? Which feedback should I throw away? And Which feedback should I throw away? And Which feedback should I throw away? And so I'm pushing and pushing and pushing so I'm pushing and pushing and pushing so I'm pushing and pushing and pushing because I'm never going to be the person because I'm never going to be the person because I'm never going to be the person to be satisfied with the initial answer, to be satisfied with the initial answer, to be satisfied with the initial answer, especially from an AI. And the payoff especially from an AI. And the payoff especially from an AI. And the payoff for me, it feels like it grows every for me, it feels like it grows every for me, it feels like it grows every single day because over time I feel more single day because over time I feel more single day because over time I feel more able to answer the question that I face able to answer the question that I face able to answer the question that I face every single day, which is what all of every single day, which is what all of every single day, which is what all of us face. What is the best way to get us face. What is the best way to get us face. What is the best way to get this work done today? I also feel like I this work done today? I also feel like I this work done today? I also feel like I leave the task with a better idea of leave the task with a better idea of leave the task with a better idea of what the models can do, where they fail, what the models can do, where they fail, what the models can do, where they fail, whose judgment I trust, what I should whose judgment I trust, what I should whose judgment I trust, what I should notice the next time a surface problem notice the next time a surface problem notice the next time a surface problem looks different. There's not a shortcut looks different. There's not a shortcut looks different. There's not a shortcut to this. People who say, "Well, Nate, to this. People who say, "Well, Nate, to this. People who say, "Well, Nate, give me the prompt for this." are asking give me the prompt for this." are asking give me the prompt for this." are asking the wrong question. This is about the wrong question. This is about the wrong question. This is about deliberately putting your brain in deliberately putting your brain in deliberately putting your brain in contact with disagreement and pushing contact with disagreement and pushing contact with disagreement and pushing until you get to a better thinking until you get to a better thinking until you get to a better thinking process. And by the end of this video, I process. And by the end of this video, I process. And by the end of this video, I want you to see how I run that friction want you to see how I run that friction want you to see how I run that friction maxing loop. When a model does something maxing loop. When a model does something maxing loop. When a model does something surprising, I test that same boundary surprising, I test that same boundary surprising, I test that same boundary really deliberately with other models, really deliberately with other models, really deliberately with other models, and I talk through the result with and I talk through the result with and I talk through the result with people that I trust to figure out what people that I trust to figure out what people that I trust to figure out what I'm learning. I probably try less fancy I'm learning. I probably try less fancy I'm learning. I probably try less fancy AI stuff than you imagine. I will AI stuff than you imagine. I will AI stuff than you imagine. I will happily try and dump a new workflow, a happily try and dump a new workflow, a happily try and dump a new workflow, a graph, a tool, a model recommendation if

graph, a tool, a model recommendation if graph, a tool, a model recommendation if it does not help with the work in front it does not help with the work in front it does not help with the work in front of me. I'm super practical. My goal is of me. I'm super practical. My goal is of me. I'm super practical. My goal is to understand the edges and capabilities to understand the edges and capabilities to understand the edges and capabilities of the model well enough to use the of the model well enough to use the of the model well enough to use the simplest thing that works. Let me show simplest thing that works. Let me show simplest thing that works. Let me show you what that looks like with the wrong you what that looks like with the wrong you what that looks like with the wrong spreadsheet. This is a real story. I spreadsheet. This is a real story. I spreadsheet. This is a real story. I asked an agent to take the current asked an agent to take the current asked an agent to take the current spreadsheet from my downloads folder to spreadsheet from my downloads folder to spreadsheet from my downloads folder to attach it to an email draft and to leave attach it to an email draft and to leave attach it to an email draft and to leave the email unscent cuz I wanted to check the email unscent cuz I wanted to check the email unscent cuz I wanted to check it first. This is a new agent. This is it first. This is a new agent. This is it first. This is a new agent. This is not Claude. This is not Grock. This is not Claude. This is not Grock. This is not Claude. This is not Grock. This is not Codeex. It's a brand new agent. It not Codeex. It's a brand new agent. It not Codeex. It's a brand new agent. It came back with the right recipient. It came back with the right recipient. It came back with the right recipient. It came back with the right subject. It was came back with the right subject. It was came back with the right subject. It was a reasonable email draft. And it came a reasonable email draft. And it came a reasonable email draft. And it came back with a spreadsheet that carried the back with a spreadsheet that carried the back with a spreadsheet that carried the right file name. It all looked good, but right file name. It all looked good, but right file name. It all looked good, but I had a feeling. So, I opened the I had a feeling. So, I opened the I had a feeling. So, I opened the attachment. The agent had pulled an attachment. The agent had pulled an attachment. The agent had pulled an older copy of the spreadsheet that was older copy of the spreadsheet that was older copy of the spreadsheet that was out of date from a previous email out of date from a previous email out of date from a previous email because it could not access downloads at because it could not access downloads at because it could not access downloads at all. Like that file system was all. Like that file system was all. Like that file system was ineligible as a tool call for that ineligible as a tool call for that ineligible as a tool call for that agent. But if I just stop there, what agent. But if I just stop there, what agent. But if I just stop there, what did I learn? Nothing. The useful lesson did I learn? Nothing. The useful lesson did I learn? Nothing. The useful lesson was not that this particular agent is was not that this particular agent is was not that this particular agent is bad at spreadsheets. I had instead bad at spreadsheets. I had instead bad at spreadsheets. I had instead learned something about how agents enter learned something about how agents enter learned something about how agents enter a new environment, about agent a new environment, about agent a new environment, about agent onboarding, about how agents disclose onboarding, about how agents disclose onboarding, about how agents disclose what they can see, about agent what they can see, about agent what they can see, about agent confidence in a place where they're not confidence in a place where they're not confidence in a place where they're not sure what's working and what's not in a sure what's working and what's not in a sure what's working and what's not in a compute environment. Because the agent compute environment. Because the agent compute environment. Because the agent in this case did not say, "Hey, Nate, I in this case did not say, "Hey, Nate, I in this case did not say, "Hey, Nate, I couldn't reach downloads. I found an couldn't reach downloads. I found an couldn't reach downloads. I found an older file somewhere else and I hoped it older file somewhere else and I hoped it older file somewhere else and I hoped it was close enough. that was my best guess was close enough. that was my best guess was close enough. that was my best guess for what you would want. Instead, the

for what you would want. Instead, the for what you would want. Instead, the agent presented the completed draft and agent presented the completed draft and agent presented the completed draft and claimed that it did the work. And so, claimed that it did the work. And so, claimed that it did the work. And so, the dangerous part of that whole system the dangerous part of that whole system the dangerous part of that whole system was the agent's ability to deceive to was the agent's ability to deceive to was the agent's ability to deceive to look like it got the work done. The look like it got the work done. The look like it got the work done. The first thing I did is I went back and I first thing I did is I went back and I first thing I did is I went back and I asked the agent to see if it could asked the agent to see if it could asked the agent to see if it could change what it could access. It change what it could access. It change what it could access. It couldn't. It this was a hard boundary. couldn't. It this was a hard boundary. couldn't. It this was a hard boundary. This particular agent, it was a personal This particular agent, it was a personal This particular agent, it was a personal assistant agent. It was not going to get assistant agent. It was not going to get assistant agent. It was not going to get access to my downloads folder. access to my downloads folder. access to my downloads folder. Apparently that was a hard edge. Well, Apparently that was a hard edge. Well, Apparently that was a hard edge. Well, immediately I went back and I retested immediately I went back and I retested immediately I went back and I retested what my core agents could do. Can Codex what my core agents could do. Can Codex what my core agents could do. Can Codex do it? Yep. Can Claude do it? Claude do it? Yep. Can Claude do it? Claude do it? Yep. Can Claude do it? Claude struggled a little bit, but Claude got struggled a little bit, but Claude got struggled a little bit, but Claude got it done. And can Grock do it? Grock was it done. And can Grock do it? Grock was it done. And can Grock do it? Grock was able to do it as well. So, what did I able to do it as well. So, what did I able to do it as well. So, what did I learn here? I learned that I could learn here? I learned that I could learn here? I learned that I could figure out very rapidly whether an AI figure out very rapidly whether an AI figure out very rapidly whether an AI was going to be useful to me by walking was going to be useful to me by walking was going to be useful to me by walking through not the stated claims of a through not the stated claims of a through not the stated claims of a particular AI when it jumps in, but the particular AI when it jumps in, but the particular AI when it jumps in, but the actual ability of the agent to disclose actual ability of the agent to disclose actual ability of the agent to disclose transparently its own capability set and transparently its own capability set and transparently its own capability set and to work with me on updating its to work with me on updating its to work with me on updating its capabilities over time. So there's a capabilities over time. So there's a capabilities over time. So there's a deliberate product insight that we can deliberate product insight that we can deliberate product insight that we can all gain here too where you can start to all gain here too where you can start to all gain here too where you can start to think about how do agents disclose what think about how do agents disclose what think about how do agents disclose what they learn in the first 2 or 3 minutes they learn in the first 2 or 3 minutes they learn in the first 2 or 3 minutes in ways that a user can learn from and in ways that a user can learn from and in ways that a user can learn from and actually get an accurate idea of the actually get an accurate idea of the actually get an accurate idea of the agent's capability, actually get agent's capability, actually get agent's capability, actually get surprise and delight, and actually

surprise and delight, and actually surprise and delight, and actually develop a sense of connection. So much develop a sense of connection. So much develop a sense of connection. So much of what is stopping agents from being of what is stopping agents from being of what is stopping agents from being useful to people who are not nerds is useful to people who are not nerds is useful to people who are not nerds is this kind of deceptiveness. This kind of this kind of deceptiveness. This kind of this kind of deceptiveness. This kind of I can help you but I can't really help I can help you but I can't really help I can help you but I can't really help you. I can help you but you can't see you. I can help you but you can't see you. I can help you but you can't see where the edges are. And one of part of where the edges are. And one of part of where the edges are. And one of part of what I'm trying to share here is how I what I'm trying to share here is how I what I'm trying to share here is how I sus out the root cause. Cuz I wouldn't sus out the root cause. Cuz I wouldn't sus out the root cause. Cuz I wouldn't describe the root cause here as you know describe the root cause here as you know describe the root cause here as you know it didn't find the downloads file. The it didn't find the downloads file. The it didn't find the downloads file. The root cause when you really think about root cause when you really think about root cause when you really think about it is that the designed product it is that the designed product it is that the designed product onboarding experience for that agent was onboarding experience for that agent was onboarding experience for that agent was really really unfit to the actual really really unfit to the actual really really unfit to the actual capabilities of the agent. It did not capabilities of the agent. It did not capabilities of the agent. It did not fit well. It didn't disclose what the fit well. It didn't disclose what the fit well. It didn't disclose what the agent was able to do and it poorly agent was able to do and it poorly agent was able to do and it poorly represented the capabilities the agent represented the capabilities the agent represented the capabilities the agent did have and that turned me off to it. did have and that turned me off to it. did have and that turned me off to it. This is the kind of thing that my brain This is the kind of thing that my brain This is the kind of thing that my brain does all day. So my brain did not store does all day. So my brain did not store does all day. So my brain did not store it's the wrong spreadsheet. My brain it's the wrong spreadsheet. My brain it's the wrong spreadsheet. My brain built a working model of agent built a working model of agent built a working model of agent onboarding and capability claims and onboarding and capability claims and onboarding and capability claims and plausible substitution and proof. And plausible substitution and proof. And plausible substitution and proof. And that model is now something I that model is now something I that model is now something I deliberately update when I think about deliberately update when I think about deliberately update when I think about new AI agents that I onboard and and new AI agents that I onboard and and new AI agents that I onboard and and select. So I don't have to play this select. So I don't have to play this select. So I don't have to play this same lesson every time a new logo pops same lesson every time a new logo pops same lesson every time a new logo pops up. I actually deliberately want to go up. I actually deliberately want to go up. I actually deliberately want to go in and say, can I walk through in and say, can I walk through in and say, can I walk through onboarding with this mental model in onboarding with this mental model in onboarding with this mental model in place for a new agent? And can I see place for a new agent? And can I see place for a new agent? And can I see whether this new agent represents itself whether this new agent represents itself whether this new agent represents itself and its capabilities in a way that and its capabilities in a way that and its capabilities in a way that encourages me to have a delightful, encourages me to have a delightful, encourages me to have a delightful, useful, accurate experience. Now, of

useful, accurate experience. Now, of useful, accurate experience. Now, of course, there's lots of ordinary AI use course, there's lots of ordinary AI use course, there's lots of ordinary AI use where I just want to get stuff done, where I just want to get stuff done, where I just want to get stuff done, right? Sometimes I just need the trains right? Sometimes I just need the trains right? Sometimes I just need the trains to run on time, so to speak. I need a to run on time, so to speak. I need a to run on time, so to speak. I need a cleaner paragraph. I need a comparison. cleaner paragraph. I need a comparison. cleaner paragraph. I need a comparison. I need a piece of research done. I need I need a piece of research done. I need I need a piece of research done. I need a piece of code. And so, I just ask and a piece of code. And so, I just ask and a piece of code. And so, I just ask and the answer arrives and the job is over. the answer arrives and the job is over. the answer arrives and the job is over. And that happens lots of the time. And And that happens lots of the time. And And that happens lots of the time. And no, I'm not calling 10 friends to figure no, I'm not calling 10 friends to figure no, I'm not calling 10 friends to figure out if this piece of code works. But out if this piece of code works. But out if this piece of code works. But even in that context, iteration is even in that context, iteration is even in that context, iteration is useful. And that's another lesson I want useful. And that's another lesson I want useful. And that's another lesson I want to call out. When I ask for a draft of a to call out. When I ask for a draft of a to call out. When I ask for a draft of a web page, which is really a a series of web page, which is really a a series of web page, which is really a a series of code problems that the AI is solving, I code problems that the AI is solving, I code problems that the AI is solving, I ask for it so that I can rapidly see ask for it so that I can rapidly see ask for it so that I can rapidly see what is wrong, refine my explanation of what is wrong, refine my explanation of what is wrong, refine my explanation of the problem, and push the model to try the problem, and push the model to try the problem, and push the model to try again. I then iterate so the current again. I then iterate so the current again. I then iterate so the current output gets better over time and I do output gets better over time and I do output gets better over time and I do that really relentlessly and that is why that really relentlessly and that is why that really relentlessly and that is why I talk so much on this channel about I talk so much on this channel about I talk so much on this channel about using agents as a way of sharpening your using agents as a way of sharpening your using agents as a way of sharpening your thinking. Agents are basically long thinking. Agents are basically long thinking. Agents are basically long leverage that you can use to shape the leverage that you can use to shape the leverage that you can use to shape the world. They allow you to get real world. They allow you to get real world. They allow you to get real leverage against big problems and leverage against big problems and leverage against big problems and iterate rapidly so that you in turn can iterate rapidly so that you in turn can iterate rapidly so that you in turn can learn what is working and what is not learn what is working and what is not learn what is working and what is not about your ideas. But I don't think about your ideas. But I don't think about your ideas. But I don't think current AI interfaces help us develop current AI interfaces help us develop current AI interfaces help us develop that skill set. I feel like I am that skill set. I feel like I am that skill set. I feel like I am fighting current AI interfaces to do fighting current AI interfaces to do fighting current AI interfaces to do that. Most AI interfaces today naturally that. Most AI interfaces today naturally that. Most AI interfaces today naturally pull us toward the middle of the current pull us toward the middle of the current pull us toward the middle of the current output distribution. By which I mean output distribution. By which I mean output distribution. By which I mean they want you to do something simple

they want you to do something simple they want you to do something simple like correct this paragraph, fix this like correct this paragraph, fix this like correct this paragraph, fix this bug, improve this design, and maybe try bug, improve this design, and maybe try bug, improve this design, and maybe try again one or two times, but not really. again one or two times, but not really. again one or two times, but not really. So borrowing the term loosely, most AI So borrowing the term loosely, most AI So borrowing the term loosely, most AI agent interfaces encourage a kind of agent interfaces encourage a kind of agent interfaces encourage a kind of relentless gradient descent. In other relentless gradient descent. In other relentless gradient descent. In other words, each correction moves the current words, each correction moves the current words, each correction moves the current result closer to the target in the result closer to the target in the result closer to the target in the middle of the AI distribution that the middle of the AI distribution that the middle of the AI distribution that the AI knows well. The harder question for AI knows well. The harder question for AI knows well. The harder question for us humans as we interact with AI is us humans as we interact with AI is us humans as we interact with AI is asking whether any given correction that asking whether any given correction that asking whether any given correction that we are iterating toward changes us and we are iterating toward changes us and we are iterating toward changes us and helps us think about the next problem helps us think about the next problem helps us think about the next problem creatively. whether what we are putting creatively. whether what we are putting creatively. whether what we are putting out there is actually creative and on out there is actually creative and on out there is actually creative and on the edges of the distribution and not the edges of the distribution and not the edges of the distribution and not just pushed to the center. Can I just pushed to the center. Can I just pushed to the center. Can I recognize a capability gap in AI when I recognize a capability gap in AI when I recognize a capability gap in AI when I see a polished design that is something see a polished design that is something see a polished design that is something I don't actually like but it looks I don't actually like but it looks I don't actually like but it looks polished and done. It has followed that polished and done. It has followed that polished and done. It has followed that gradient descent. It looks like a gradient descent. It looks like a gradient descent. It looks like a standard AI design. Can I see, you know standard AI design. Can I see, you know standard AI design. Can I see, you know what, this doesn't actually reflect my what, this doesn't actually reflect my what, this doesn't actually reflect my vision and I know how to ask for vision and I know how to ask for vision and I know how to ask for something bigger. And so what I find is something bigger. And so what I find is something bigger. And so what I find is when I'm using AI, well, I might be when I'm using AI, well, I might be when I'm using AI, well, I might be doing the same process with a doing the same process with a doing the same process with a spreadsheet, with a website, with a spreadsheet, with a website, with a spreadsheet, with a website, with a piece of writing, the surface changes, piece of writing, the surface changes, piece of writing, the surface changes, but the pattern of what I'm doing but the pattern of what I'm doing but the pattern of what I'm doing persists. And the thing I want to persists. And the thing I want to persists. And the thing I want to emphasize is is that that pattern needs emphasize is is that that pattern needs emphasize is is that that pattern needs to survive in us, in people. We could to survive in us, in people. We could to survive in us, in people. We could have a database or a graft or a prompt have a database or a graft or a prompt have a database or a graft or a prompt that can help us. But if the tool is

that can help us. But if the tool is that can help us. But if the tool is able to retrieve the lesson and I have able to retrieve the lesson and I have able to retrieve the lesson and I have not become better at recognizing when it not become better at recognizing when it not become better at recognizing when it matters, then all I've done is found a matters, then all I've done is found a matters, then all I've done is found a way for my brain to stay roughly the way for my brain to stay roughly the way for my brain to stay roughly the same instead of finding a way to same instead of finding a way to same instead of finding a way to actually scale up over the course of actually scale up over the course of actually scale up over the course of using AI. And so much of that pushing using AI. And so much of that pushing using AI. And so much of that pushing the edges of AI. Finding places where the edges of AI. Finding places where the edges of AI. Finding places where you can push back and say this doesn't you can push back and say this doesn't you can push back and say this doesn't reflect the vision that I have. This is reflect the vision that I have. This is reflect the vision that I have. This is where I want to go, right? This is the where I want to go, right? This is the where I want to go, right? This is the design I want. This is what I want to design I want. This is what I want to design I want. This is what I want to say and you're not getting it. Or this say and you're not getting it. Or this say and you're not getting it. Or this is what I want to build and you haven't is what I want to build and you haven't is what I want to build and you haven't got it right yet. This is why the got it right yet. This is why the got it right yet. This is why the conversation around Ilia Sutzgiver and conversation around Ilia Sutzgiver and conversation around Ilia Sutzgiver and safe super intelligence by the way is so safe super intelligence by the way is so safe super intelligence by the way is so interesting right now. So the rumor is interesting right now. So the rumor is interesting right now. So the rumor is that Ilia is working on testime learning that Ilia is working on testime learning that Ilia is working on testime learning and the idea behind testime learning is and the idea behind testime learning is and the idea behind testime learning is that there is the possibility of an that there is the possibility of an that there is the possibility of an architecture for a model that can learn architecture for a model that can learn architecture for a model that can learn as it experiences the world and that as it experiences the world and that as it experiences the world and that does better over time as it learns. So does better over time as it learns. So does better over time as it learns. So the model gets smarter as it gets fed the model gets smarter as it gets fed the model gets smarter as it gets fed prompts. The model gets smarter as it prompts. The model gets smarter as it prompts. The model gets smarter as it experiences the world. Now, nobody experiences the world. Now, nobody experiences the world. Now, nobody outside the company knows whether that's outside the company knows whether that's outside the company knows whether that's actually being worked on or not. I'm actually being worked on or not. I'm actually being worked on or not. I'm more interested in the broad idea here.

more interested in the broad idea here. more interested in the broad idea here. A system that keeps updating as it A system that keeps updating as it A system that keeps updating as it encounters the world instead of doing encounters the world instead of doing encounters the world instead of doing all of its learning before deployment is all of its learning before deployment is all of its learning before deployment is one of the holy grails of AI. We humans, one of the holy grails of AI. We humans, one of the holy grails of AI. We humans, we already live that way. We have an we already live that way. We have an we already live that way. We have an experience, reality answers back, experience, reality answers back, experience, reality answers back, another person tells us what we missed, another person tells us what we missed, another person tells us what we missed, and we update our thinking while we're and we update our thinking while we're and we update our thinking while we're still going. Let's say the meeting for still going. Let's say the meeting for still going. Let's say the meeting for you goes really badly on Tuesday. A you goes really badly on Tuesday. A you goes really badly on Tuesday. A colleague explains, "Hey, here's some colleague explains, "Hey, here's some colleague explains, "Hey, here's some feedback for how that could go better on feedback for how that could go better on feedback for how that could go better on Wednesday and by Thursday you're Wednesday and by Thursday you're Wednesday and by Thursday you're changing the way you interact with your changing the way you interact with your changing the way you interact with your colleagues because of it." That's a colleagues because of it." That's a colleagues because of it." That's a super normal human professional super normal human professional super normal human professional environment interaction. I bet a lot of environment interaction. I bet a lot of environment interaction. I bet a lot of us have had that. I certainly have. We us have had that. I certainly have. We us have had that. I certainly have. We are in many ways test time learning are in many ways test time learning are in many ways test time learning machines, right? And that is one of the machines, right? And that is one of the machines, right? And that is one of the things that makes human judgment so things that makes human judgment so things that makes human judgment so valuable and interesting. The valuable and interesting. The valuable and interesting. The opportunity I see is to put AI inside opportunity I see is to put AI inside opportunity I see is to put AI inside that existing human loop so that we as that existing human loop so that we as that existing human loop so that we as humans learn faster. We compound faster. humans learn faster. We compound faster. humans learn faster. We compound faster. Our experience and judgment grows Our experience and judgment grows Our experience and judgment grows faster. I want anyone who is afraid that faster. I want anyone who is afraid that faster. I want anyone who is afraid that they are delegating their brain to watch they are delegating their brain to watch they are delegating their brain to watch this video because this is the way out.

this video because this is the way out. this video because this is the way out. And you might wonder how can AI help? And you might wonder how can AI help? And you might wonder how can AI help? Well, AI can help by giving me more Well, AI can help by giving me more Well, AI can help by giving me more shots on goal, right? It can give me shots on goal, right? It can give me shots on goal, right? It can give me more attempts, more counter examples, more attempts, more counter examples, more attempts, more counter examples, more comparisons. I personally feel like more comparisons. I personally feel like more comparisons. I personally feel like my thinking has gotten sharper because I my thinking has gotten sharper because I my thinking has gotten sharper because I am constantly having to push back on AI am constantly having to push back on AI am constantly having to push back on AI to chip away at an idea that I know is to chip away at an idea that I know is to chip away at an idea that I know is in a general concept that isn't precise in a general concept that isn't precise in a general concept that isn't precise enough yet. And I see AI try one version enough yet. And I see AI try one version enough yet. And I see AI try one version and another version and I keep saying and another version and I keep saying and another version and I keep saying no, no, no. And I feel like I'm chipping no, no, no. And I feel like I'm chipping no, no, no. And I feel like I'm chipping away at a sculpture and chipping away away at a sculpture and chipping away away at a sculpture and chipping away every piece of marble that isn't the every piece of marble that isn't the every piece of marble that isn't the sculpture until I get exactly what I sculpture until I get exactly what I sculpture until I get exactly what I want. I get more chances to see my own want. I get more chances to see my own want. I get more chances to see my own expectations not work. And that helps me expectations not work. And that helps me expectations not work. And that helps me to iterate toward what I want to see. As to iterate toward what I want to see. As to iterate toward what I want to see. As long as I'm thinking, as long as I'm long as I'm thinking, as long as I'm long as I'm thinking, as long as I'm engaged, the model does not get to engaged, the model does not get to engaged, the model does not get to decide what I learn. I choose that. And decide what I learn. I choose that. And decide what I learn. I choose that. And this is what using multiple models has this is what using multiple models has this is what using multiple models has meant for my work. I'm asking because a meant for my work. I'm asking because a meant for my work. I'm asking because a different model can expose different different model can expose different different model can expose different input sources and potentially give me a input sources and potentially give me a input sources and potentially give me a different failure mode, a different different failure mode, a different different failure mode, a different learning mode. I do the same with learning mode. I do the same with learning mode. I do the same with Claude. If Grock or Claude gives me a Claude. If Grock or Claude gives me a Claude. If Grock or Claude gives me a really thoughtful perspective that I really thoughtful perspective that I really thoughtful perspective that I agree with, I can pull that in. If it agree with, I can pull that in. If it agree with, I can pull that in. If it gives me a snarky, clever criticism that gives me a snarky, clever criticism that gives me a snarky, clever criticism that sounds good, but it doesn't smell right, sounds good, but it doesn't smell right, sounds good, but it doesn't smell right, I, as the human can discard it, even in I, as the human can discard it, even in I, as the human can discard it, even in the middle of a process loop, I'm going the middle of a process loop, I'm going the middle of a process loop, I'm going to keep using because in that particular to keep using because in that particular to keep using because in that particular instance, it didn't work. And if all instance, it didn't work. And if all instance, it didn't work. And if all three models agree, I can ask myself three models agree, I can ask myself three models agree, I can ask myself what evidence would make them wrong. I what evidence would make them wrong. I what evidence would make them wrong. I can go to my human team and say, "Hey can go to my human team and say, "Hey can go to my human team and say, "Hey guys, I feel like this isn't quite guys, I feel like this isn't quite guys, I feel like this isn't quite getting it. I feel like I cannot get AI

getting it. I feel like I cannot get AI getting it. I feel like I cannot get AI to move off of the beaten path. It is to move off of the beaten path. It is to move off of the beaten path. It is stuck in gradient descent. I need to stuck in gradient descent. I need to stuck in gradient descent. I need to find another path forward. How can I find another path forward. How can I find another path forward. How can I think about this problem differently?" think about this problem differently?" think about this problem differently?" Design works the same way. I use Claude Design works the same way. I use Claude Design works the same way. I use Claude a lot when I'm thinking about design. I a lot when I'm thinking about design. I a lot when I'm thinking about design. I want to see what it proposes because a want to see what it proposes because a want to see what it proposes because a concrete design gives my taste something concrete design gives my taste something concrete design gives my taste something to react to. Sometimes I like the to react to. Sometimes I like the to react to. Sometimes I like the result. often times, especially the result. often times, especially the result. often times, especially the first two or three iterations through a first two or three iterations through a first two or three iterations through a design, I absolutely hate it. And Claude design, I absolutely hate it. And Claude design, I absolutely hate it. And Claude is the best one out there. And part of is the best one out there. And part of is the best one out there. And part of what I hate is that Claude tends to get what I hate is that Claude tends to get what I hate is that Claude tends to get stuck in the clays and stuck in the stuck in the clays and stuck in the stuck in the clays and stuck in the maroon reds right now. And that's a new maroon reds right now. And that's a new maroon reds right now. And that's a new version of what used to be another local version of what used to be another local version of what used to be another local attractor in agent space. Claude used to attractor in agent space. Claude used to attractor in agent space. Claude used to be stuck more in the dark linear purple be stuck more in the dark linear purple be stuck more in the dark linear purple space. And that was the design obsession space. And that was the design obsession space. And that was the design obsession for Claude. And so if I don't like it, for Claude. And so if I don't like it, for Claude. And so if I don't like it, the next useful move is not to keep the next useful move is not to keep the next useful move is not to keep asking for random variations until one asking for random variations until one asking for random variations until one feels acceptable. I see that a lot. feels acceptable. I see that a lot. feels acceptable. I see that a lot. Instead, I'm going to try and friction Instead, I'm going to try and friction Instead, I'm going to try and friction max. I'm going to look in my own brain max. I'm going to look in my own brain max. I'm going to look in my own brain and I'm going to say, "What do I dislike and I'm going to say, "What do I dislike and I'm going to say, "What do I dislike and why?" Or I'm asking a friend, and a and why?" Or I'm asking a friend, and a and why?" Or I'm asking a friend, and a friend may immediately notice that the friend may immediately notice that the friend may immediately notice that the page asks the viewer to understand way page asks the viewer to understand way page asks the viewer to understand way too much and is way too textheavy.

too much and is way too textheavy. too much and is way too textheavy. Another classic AI design failure mode. Another classic AI design failure mode. Another classic AI design failure mode. One of the things I feel like I've One of the things I feel like I've One of the things I feel like I've gotten a lot better at as I've gone gotten a lot better at as I've gone gotten a lot better at as I've gone through this friction maxing process is through this friction maxing process is through this friction maxing process is I've got better at self-consciously I've got better at self-consciously I've got better at self-consciously catching where I have unclear catching where I have unclear catching where I have unclear articulation where I am not saying articulation where I am not saying articulation where I am not saying clearly what I intend and I think that clearly what I intend and I think that clearly what I intend and I think that that has helped me to be much better at that has helped me to be much better at that has helped me to be much better at communicating what I want to AI. In communicating what I want to AI. In communicating what I want to AI. In 2020, I was using Figma. 2024, using 2020, I was using Figma. 2024, using 2020, I was using Figma. 2024, using windsurf. Now 2026, I'm asking myself, windsurf. Now 2026, I'm asking myself, windsurf. Now 2026, I'm asking myself, are my instructions clear? Is my intent are my instructions clear? Is my intent are my instructions clear? Is my intent clear? Have I given the model enough clear? Have I given the model enough clear? Have I given the model enough non-generic input that it can actually non-generic input that it can actually non-generic input that it can actually find design inspiration that's unique find design inspiration that's unique find design inspiration that's unique and not just copy something? Have I and not just copy something? Have I and not just copy something? Have I looked across a wide range of model looked across a wide range of model looked across a wide range of model responses so I get a sense of how models responses so I get a sense of how models responses so I get a sense of how models view this problem? Have I also talked view this problem? Have I also talked view this problem? Have I also talked with humans so I get a sense of how with humans so I get a sense of how with humans so I get a sense of how humans view this problem? That's humans view this problem? That's humans view this problem? That's increasingly important because we're not increasingly important because we're not increasingly important because we're not just designing for humans anymore. We just designing for humans anymore. We just designing for humans anymore. We have to design for humans and agents have to design for humans and agents have to design for humans and agents both. So the perspective of both is both. So the perspective of both is both. So the perspective of both is important. And then when I have finally important. And then when I have finally important. And then when I have finally come back and made a specific choice and come back and made a specific choice and come back and made a specific choice and it has come back incorrect, can I point it has come back incorrect, can I point it has come back incorrect, can I point to the failure and explain to Claude in to the failure and explain to Claude in to the failure and explain to Claude in this case what I would want instead and this case what I would want instead and this case what I would want instead and how to correct it. This is the loop that how to correct it. This is the loop that how to correct it. This is the loop that I am running in 2026 with AI. I'm asking I am running in 2026 with AI. I'm asking I am running in 2026 with AI. I'm asking AI to not agree with me so quickly. I'm AI to not agree with me so quickly. I'm AI to not agree with me so quickly. I'm asking it to question me. I'm asking it asking it to question me. I'm asking it asking it to question me. I'm asking it to name the assumptions behind the to name the assumptions behind the to name the assumptions behind the answers it gives me. I'm asking it to answers it gives me. I'm asking it to answers it gives me. I'm asking it to give me a steelman case against my view give me a steelman case against my view give me a steelman case against my view and to throw out the straw man. I'm and to throw out the straw man. I'm and to throw out the straw man. I'm asking it to show me where two parts of asking it to show me where two parts of asking it to show me where two parts of my request fight with each other so I

my request fight with each other so I my request fight with each other so I can resolve the conflict and not have can resolve the conflict and not have can resolve the conflict and not have the model do it. But I don't want this the model do it. But I don't want this the model do it. But I don't want this video to collapse into Nate is giving me video to collapse into Nate is giving me video to collapse into Nate is giving me a magic prompt and now it's all going to a magic prompt and now it's all going to a magic prompt and now it's all going to be fine. The prompt cannot help you with be fine. The prompt cannot help you with be fine. The prompt cannot help you with the decision to continually train your the decision to continually train your the decision to continually train your brain. It can start and kickstart that brain. It can start and kickstart that brain. It can start and kickstart that harder interaction. And I have that, but harder interaction. And I have that, but harder interaction. And I have that, but but ultimately you and I are the ones but ultimately you and I are the ones but ultimately you and I are the ones who have to push ourselves in the age of who have to push ourselves in the age of who have to push ourselves in the age of AI. We are the ones who have to resist AI. We are the ones who have to resist AI. We are the ones who have to resist the first polished answer. We are the the first polished answer. We are the the first polished answer. We are the ones that have to notice when another ones that have to notice when another ones that have to notice when another model has become a rubber stamp and is model has become a rubber stamp and is model has become a rubber stamp and is always approving instead of a real always approving instead of a real always approving instead of a real challenge. And by the way, I tried challenge. And by the way, I tried challenge. And by the way, I tried Gemini for a while. Gemini became a Gemini for a while. Gemini became a Gemini for a while. Gemini became a rubber stamp. I have not been using rubber stamp. I have not been using rubber stamp. I have not been using Gemini lately because of that. I noticed Gemini lately because of that. I noticed Gemini lately because of that. I noticed an edge in model capability and I've an edge in model capability and I've an edge in model capability and I've been walking away. Now, does that mean been walking away. Now, does that mean been walking away. Now, does that mean all Gemini models are unusful? No. I all Gemini models are unusful? No. I all Gemini models are unusful? No. I have a much more sophisticated world have a much more sophisticated world have a much more sophisticated world model. Human feedback is so important. I model. Human feedback is so important. I model. Human feedback is so important. I know I've talked about my friends, know I've talked about my friends, know I've talked about my friends, right? But I really want to call out if right? But I really want to call out if right? But I really want to call out if you don't have a community of trusted you don't have a community of trusted you don't have a community of trusted friends and colleagues who are looking friends and colleagues who are looking friends and colleagues who are looking at some part of what you're building, at some part of what you're building, at some part of what you're building, who are saying and thinking and testing who are saying and thinking and testing who are saying and thinking and testing and saying this is good, this is not and saying this is good, this is not and saying this is good, this is not good, you're missing out. This is part good, you're missing out. This is part good, you're missing out. This is part of what is powerful about the Slack of what is powerful about the Slack of what is powerful about the Slack community that I've put together is that community that I've put together is that community that I've put together is that people expose their thinking to one people expose their thinking to one people expose their thinking to one another and people critique one another.

another and people critique one another. another and people critique one another. It's also part of what makes the YouTube It's also part of what makes the YouTube It's also part of what makes the YouTube comments under these videos special. comments under these videos special. comments under these videos special. People expose their thinking. People ask People expose their thinking. People ask People expose their thinking. People ask for feedback. I love that. It's so for feedback. I love that. It's so for feedback. I love that. It's so important to find people you can build important to find people you can build important to find people you can build with because in a sense a single human with because in a sense a single human with because in a sense a single human reaction that is out of the ordinary can reaction that is out of the ordinary can reaction that is out of the ordinary can help my brain see something from a new help my brain see something from a new help my brain see something from a new perspective that all of the models perspective that all of the models perspective that all of the models missed because the person knows me or missed because the person knows me or missed because the person knows me or the person knows the audience or the the person knows the audience or the the person knows the audience or the person knows the work or the situation person knows the work or the situation person knows the work or the situation differently. And so I love and I differently. And so I love and I differently. And so I love and I desperately need the feedback from the desperately need the feedback from the desperately need the feedback from the humans in my life. I love to cycle that humans in my life. I love to cycle that humans in my life. I love to cycle that feedback back into AI when I'm working feedback back into AI when I'm working feedback back into AI when I'm working on a problem. If somebody says a design on a problem. If somebody says a design on a problem. If somebody says a design is confus confusing, as much as I can, I is confus confusing, as much as I can, I is confus confusing, as much as I can, I give that feedback directly to Claude. I give that feedback directly to Claude. I give that feedback directly to Claude. I say, Claude, which assumption in the say, Claude, which assumption in the say, Claude, which assumption in the design would make this reaction design would make this reaction design would make this reaction reasonable? How can you deeply reasonable? How can you deeply reasonable? How can you deeply understand and empathize with the human understand and empathize with the human understand and empathize with the human here and update the design as a result?

here and update the design as a result? here and update the design as a result? If somebody rejects the premise of an If somebody rejects the premise of an If somebody rejects the premise of an argument and and I'm going back and argument and and I'm going back and argument and and I'm going back and thinking about how to sharpen it, I can thinking about how to sharpen it, I can thinking about how to sharpen it, I can ask Grock to build and research the ask Grock to build and research the ask Grock to build and research the strongest version of the rejection, strongest version of the rejection, strongest version of the rejection, strengthen the rejection and the strengthen the rejection and the strengthen the rejection and the disagreement that my friend made by disagreement that my friend made by disagreement that my friend made by looking on the internet for more pieces looking on the internet for more pieces looking on the internet for more pieces like that so I can understand it better. like that so I can understand it better. like that so I can understand it better. If a colleague catches an operational If a colleague catches an operational If a colleague catches an operational problem in something that Codeex built problem in something that Codeex built problem in something that Codeex built for me, I can ask Codeex to explain why for me, I can ask Codeex to explain why for me, I can ask Codeex to explain why its own testing did not catch that its own testing did not catch that its own testing did not catch that problem with the software. And I've done problem with the software. And I've done problem with the software. And I've done that. Trusted people can share blind that. Trusted people can share blind that. Trusted people can share blind spots as well. Models, of course, as spots as well. Models, of course, as spots as well. Models, of course, as I've talked about, can repeat the same I've talked about, can repeat the same I've talked about, can repeat the same confident mistakes and do so much more confident mistakes and do so much more confident mistakes and do so much more often than people. And if you don't have often than people. And if you don't have often than people. And if you don't have your own sense of compass, more feedback your own sense of compass, more feedback your own sense of compass, more feedback can sound like noise instead of making can sound like noise instead of making can sound like noise instead of making you wiser. And this is why I tell you wiser. And this is why I tell you wiser. And this is why I tell stories about what I learn. A story, stories about what I learn. A story, stories about what I learn. A story, even if you're not telling it on even if you're not telling it on even if you're not telling it on YouTube, a story forces the pattern into YouTube, a story forces the pattern into YouTube, a story forces the pattern into a form that I can remember and explain. a form that I can remember and explain. a form that I can remember and explain. The wrong spreadsheet is an agent The wrong spreadsheet is an agent The wrong spreadsheet is an agent claiming that it got something done claiming that it got something done claiming that it got something done without actually getting it done all the without actually getting it done all the without actually getting it done all the way. And by telling that story, I am way. And by telling that story, I am way. And by telling that story, I am actually able to think differently about actually able to think differently about actually able to think differently about every single agent I run across. It is every single agent I run across. It is every single agent I run across. It is turning that Asian experience from a turning that Asian experience from a turning that Asian experience from a single bad product experience into a single bad product experience into a single bad product experience into a larger insight into how agents are larger insight into how agents are larger insight into how agents are built, launched, and introduced to the built, launched, and introduced to the built, launched, and introduced to the public today. Which outputs from which public today. Which outputs from which public today. Which outputs from which models need special source checks to models need special source checks to models need special source checks to avoid being inaccurate? I'll give you an

avoid being inaccurate? I'll give you an avoid being inaccurate? I'll give you an example. Grock tends to be really fast example. Grock tends to be really fast example. Grock tends to be really fast and it tends to need extra source checks and it tends to need extra source checks and it tends to need extra source checks because it moves really quickly. And so because it moves really quickly. And so because it moves really quickly. And so I always double check what I get from I always double check what I get from I always double check what I get from Grock. You know, I see comments in my Grock. You know, I see comments in my Grock. You know, I see comments in my community where people talk about the community where people talk about the community where people talk about the fact that they're tired of the hype is fact that they're tired of the hype is fact that they're tired of the hype is they're tired of new models. I get it. they're tired of new models. I get it. they're tired of new models. I get it. Part of why I'm making this video is to Part of why I'm making this video is to Part of why I'm making this video is to remind you that you need to get to a remind you that you need to get to a remind you that you need to get to a practical payoff for AI. It needs to be practical payoff for AI. It needs to be practical payoff for AI. It needs to be worth it to you. And part of what I am worth it to you. And part of what I am worth it to you. And part of what I am doing with the way I build this learning doing with the way I build this learning doing with the way I build this learning loop is I am choosing to learn skills loop is I am choosing to learn skills loop is I am choosing to learn skills that help me to choose AI tools faster that help me to choose AI tools faster that help me to choose AI tools faster that help me to spend less time that help me to spend less time that help me to spend less time polishing work built on wrong polishing work built on wrong polishing work built on wrong assumptions. Almost everything I'm assumptions. Almost everything I'm assumptions. Almost everything I'm doing, and you could call this designing doing, and you could call this designing doing, and you could call this designing a human harness, right? Almost a human harness, right? Almost a human harness, right? Almost everything I'm doing with my brain and everything I'm doing with my brain and everything I'm doing with my brain and the way I train it is designed to help the way I train it is designed to help the way I train it is designed to help me make fast, accurate decisions about me make fast, accurate decisions about me make fast, accurate decisions about AI capabilities and to update those AI capabilities and to update those AI capabilities and to update those reliably when AI models evolve. And when reliably when AI models evolve. And when reliably when AI models evolve. And when you start to do that, the brain rot you start to do that, the brain rot you start to do that, the brain rot question stops being some abstract question stops being some abstract question stops being some abstract internet argument or something to get internet argument or something to get internet argument or something to get angry about on TikTok and it becomes angry about on TikTok and it becomes angry about on TikTok and it becomes something instead that I can see something instead that I can see something instead that I can see surfacing in my own day. Either I am surfacing in my own day. Either I am surfacing in my own day. Either I am anti-brain rot, I am digging into anti-brain rot, I am digging into anti-brain rot, I am digging into friction. I am training my brain, or I friction. I am training my brain, or I friction. I am training my brain, or I am just going with the flow and copying am just going with the flow and copying am just going with the flow and copying and pasting and not paying attention to and pasting and not paying attention to and pasting and not paying attention to the AI output. There are some useful the AI output. There are some useful the AI output. There are some useful questions that help you know where you questions that help you know where you questions that help you know where you are in this. I'll give you one. Can I are in this. I'll give you one. Can I are in this. I'll give you one. Can I explain why my mind changed without explain why my mind changed without explain why my mind changed without asking a model to reconstruct the reason

asking a model to reconstruct the reason asking a model to reconstruct the reason for me? Do I have it up here? If AI for me? Do I have it up here? If AI for me? Do I have it up here? If AI always forms the first opinion and always forms the first opinion and always forms the first opinion and writes the plan and interprets the writes the plan and interprets the writes the plan and interprets the feedback, can I actually articulate an feedback, can I actually articulate an feedback, can I actually articulate an independent view and explain clearly to independent view and explain clearly to independent view and explain clearly to the AI, this is where you're wrong and the AI, this is where you're wrong and the AI, this is where you're wrong and this is why or am I just learning to be this is why or am I just learning to be this is why or am I just learning to be a validator of decisions that I never a validator of decisions that I never a validator of decisions that I never learn to make. So here's my question for learn to make. So here's my question for learn to make. So here's my question for you. Where are you in this journey of you. Where are you in this journey of you. Where are you in this journey of learning to work with AI? I think this learning to work with AI? I think this learning to work with AI? I think this is the most important journey of our is the most important journey of our is the most important journey of our lives. When you use AI on a serious lives. When you use AI on a serious lives. When you use AI on a serious task, what is your brain doing? Before task, what is your brain doing? Before task, what is your brain doing? Before the AI answers, are you asking more the AI answers, are you asking more the AI answers, are you asking more questions back? Are you interrupting it? questions back? Are you interrupting it? questions back? Are you interrupting it? Are you pushing it? Have the courage to Are you pushing it? Have the courage to Are you pushing it? Have the courage to think about how you're using AI? Now, if think about how you're using AI? Now, if think about how you're using AI? Now, if you are already iterating, if you're one you are already iterating, if you're one you are already iterating, if you're one of those people who pushes the model, of those people who pushes the model, of those people who pushes the model, are you someone who is also thinking are you someone who is also thinking are you someone who is also thinking about how the humans in your life are about how the humans in your life are about how the humans in your life are helping you evolve your perception of helping you evolve your perception of helping you evolve your perception of AI? Are you deliberately cultivating AI? Are you deliberately cultivating AI? Are you deliberately cultivating human community and artificial human community and artificial human community and artificial intelligence skill sets in ways that intelligence skill sets in ways that intelligence skill sets in ways that allow you to grow as a person? If you allow you to grow as a person? If you allow you to grow as a person? If you are getting human perspective, if you're are getting human perspective, if you're are getting human perspective, if you're working with the AI, how do you know working with the AI, how do you know working with the AI, how do you know that you're getting better? Are you able that you're getting better? Are you able that you're getting better? Are you able to more efficiently push through and to more efficiently push through and to more efficiently push through and chip that marble off an idea and get chip that marble off an idea and get chip that marble off an idea and get exactly what you want out there and have exactly what you want out there and have exactly what you want out there and have more confidence and belief that it's more confidence and belief that it's more confidence and belief that it's beautiful because you are so skilled at beautiful because you are so skilled at beautiful because you are so skilled at understanding how to push on AI, get understanding how to push on AI, get understanding how to push on AI, get feedback from humans, form your own feedback from humans, form your own feedback from humans, form your own perspective, and use your thinking to

perspective, and use your thinking to perspective, and use your thinking to work with AI to get something into the work with AI to get something into the work with AI to get something into the world that it just flows easier and world that it just flows easier and world that it just flows easier and easier over time. That is what I want easier over time. That is what I want easier over time. That is what I want you to see when you look over my you to see when you look over my you to see when you look over my shoulder. Look, the screen might change shoulder. Look, the screen might change shoulder. Look, the screen might change from codeex to gro to claude to a text from codeex to gro to claude to a text from codeex to gro to claude to a text conversation, but what's really conversation, but what's really conversation, but what's really happening is up here in my brain. That's happening is up here in my brain. That's happening is up here in my brain. That's what I want to give you a picture of. what I want to give you a picture of. what I want to give you a picture of. Every disagreement that I am giving the Every disagreement that I am giving the Every disagreement that I am giving the AI gives me a chance to understand AI gives me a chance to understand AI gives me a chance to understand capabilities and blind spots, to capabilities and blind spots, to capabilities and blind spots, to understand my friends better, to understand my friends better, to understand my friends better, to understand my own sense of taste more understand my own sense of taste more understand my own sense of taste more clearly. So, when you have your next clearly. So, when you have your next clearly. So, when you have your next serious task, don't stop at asking serious task, don't stop at asking serious task, don't stop at asking whether the AI produced a better answer. whether the AI produced a better answer. whether the AI produced a better answer. Ask whether you formed your own Ask whether you formed your own Ask whether you formed your own perspective, whether you encountered perspective, whether you encountered perspective, whether you encountered real resistance along the way. Ask real resistance along the way. Ask real resistance along the way. Ask whether you changed your mind for a whether you changed your mind for a whether you changed your mind for a reason you can explain. And then ask the reason you can explain. And then ask the reason you can explain. And then ask the question that we we started with in the question that we we started with in the question that we we started with in the most personal way possible. Do I feel most personal way possible. Do I feel most personal way possible. Do I feel more capable because of AI? As the AI more capable because of AI? As the AI more capable because of AI? As the AI gets better, is my judgment getting gets better, is my judgment getting gets better, is my judgment getting better with it? Is my craft getting better with it? Is my craft getting better with it? Is my craft getting better? The world's not going to slow better? The world's not going to slow better? The world's not going to slow down and we know that. And the answer is down and we know that. And the answer is down and we know that. And the answer is not going to be outsourcing every not going to be outsourcing every not going to be outsourcing every difficult thought to a system that is difficult thought to a system that is difficult thought to a system that is improving faster than you are. If you do improving faster than you are. If you do improving faster than you are. If you do that, you're just going to end up being that, you're just going to end up being that, you're just going to end up being a meat puppet, right? You're just going a meat puppet, right? You're just going a meat puppet, right? You're just going to end up producing whatever AI gives to end up producing whatever AI gives to end up producing whatever AI gives you. Push the AI to push you. Check it you. Push the AI to push you. Check it you. Push the AI to push you. Check it against another model. Check both against another model. Check both against another model. Check both against people and reality and discard, against people and reality and discard, against people and reality and discard, discard, discard, discard. AI doesn't discard, discard, discard. AI doesn't discard, discard, discard. AI doesn't reduce my thinking. AI has increased the

reduce my thinking. AI has increased the reduce my thinking. AI has increased the friction that I have to face to produce friction that I have to face to produce friction that I have to face to produce something in the world. And I love that something in the world. And I love that something in the world. And I love that because it's made me smarter. How is AI because it's made me smarter. How is AI because it's made me smarter. How is AI making you smarter? Let me know in the making you smarter? Let me know in the making you smarter? Let me know in the comments.