I guess we're writing loops now?

Here's your monthly reminder that you Here's your monthly reminder that you shouldn't be prompt encoding agents shouldn't be prompt encoding agents shouldn't be prompt encoding agents anymore. You should be designing loops anymore. You should be designing loops anymore. You should be designing loops that prompt your agents. I don't know that prompt your agents. I don't know that prompt your agents. I don't know about y'all, but this memo didn't make about y'all, but this memo didn't make about y'all, but this memo didn't make it to me. Of course, I've seen loops it to me. Of course, I've seen loops it to me. Of course, I've seen loops before. Things like the Ralph loop before. Things like the Ralph loop before. Things like the Ralph loop really helped me think about how agents really helped me think about how agents really helped me think about how agents can do more over time, but it also can do more over time, but it also can do more over time, but it also massively increased the error rate of massively increased the error rate of massively increased the error rate of the changes that I was having my agents the changes that I was having my agents the changes that I was having my agents make. They were really cool, but they make. They were really cool, but they make. They were really cool, but they didn't seem that productive, and I found didn't seem that productive, and I found didn't seem that productive, and I found myself going back to the usual, which myself going back to the usual, which myself going back to the usual, which was asking the model to make a plan, was asking the model to make a plan, was asking the model to make a plan, reading the plan, saying, "Yeah, that reading the plan, saying, "Yeah, that reading the plan, saying, "Yeah, that looks good. Go do this part." and then looks good. Go do this part." and then looks good. Go do this part." and then the next part, then the next part, the next part, then the next part, the next part, then the next part, another agent review it, then bringing another agent review it, then bringing another agent review it, then bringing the feedback back to the first agent, the feedback back to the first agent, the feedback back to the first agent, and just the usual looping of work, but and just the usual looping of work, but and just the usual looping of work, but I was the one running the loop. I was I was the one running the loop. I was I was the one running the loop. I was the one doing the hand-holding and the one doing the hand-holding and the one doing the hand-holding and bringing things from part one to part bringing things from part one to part bringing things from part one to part two, and making sure all of my agents two, and making sure all of my agents two, and making sure all of my agents had the context they needed to build had the context they needed to build had the context they needed to build well. And Pete, as always, is a bit well. And Pete, as always, is a bit well. And Pete, as always, is a bit ahead of the curve. I have been a huge ahead of the curve. I have been a huge ahead of the curve. I have been a huge fan of him since way before the open fan of him since way before the open fan of him since way before the open claw chaos, because he knew how to think claw chaos, because he knew how to think claw chaos, because he knew how to think about building with agents in a about building with agents in a about building with agents in a fundamentally different way that made fundamentally different way that made fundamentally different way that made him way more productive. I think of Pete him way more productive. I think of Pete him way more productive. I think of Pete as an experimental figure ways, where as an experimental figure ways, where as an experimental figure ways, where rather than being the role model we rather than being the role model we rather than being the role model we should all be copying, he's the person should all be copying, he's the person should all be copying, he's the person figuring out what the future looks like figuring out what the future looks like figuring out what the future looks like in a weird jank duct tape version now, in a weird jank duct tape version now, in a weird jank duct tape version now, and we can all learn from that and see and we can all learn from that and see and we can all learn from that and see where things are going. At least, that's where things are going. At least, that's where things are going. At least, that's how I used to think about him, and how I used to think about him, and how I used to think about him, and honestly, I'll admit I still do in a lot honestly, I'll admit I still do in a lot honestly, I'll admit I still do in a lot of ways, of ways, of ways, but then I started building more with but then I started building more with but then I started building more with loops. I started getting my agents to loops. I started getting my agents to loops. I started getting my agents to prompt themselves. I started setting up prompt themselves. I started setting up prompt themselves. I started setting up systems where agents would review code, systems where agents would review code, systems where agents would review code, give feedback, adjust it, and then give feedback, adjust it, and then give feedback, adjust it, and then trigger re-reviews. I started building trigger re-reviews. I started building trigger re-reviews. I started building systems that would watch pull requests systems that would watch pull requests systems that would watch pull requests and watch existing issues on other repos

and watch existing issues on other repos and watch existing issues on other repos to tell me when updates happen. I to tell me when updates happen. I to tell me when updates happen. I started using Hermes agent to bring started using Hermes agent to bring started using Hermes agent to bring context to me instead of to go out and context to me instead of to go out and context to me instead of to go out and find it for me. And I've accepted now find it for me. And I've accepted now find it for me. And I've accepted now that Pete's right. We should still be that Pete's right. We should still be that Pete's right. We should still be writing prompts, though. More writing prompts, though. More writing prompts, though. More importantly, I would argue now that the importantly, I would argue now that the importantly, I would argue now that the majority of your agent runs should majority of your agent runs should majority of your agent runs should probably not be running with prompts probably not be running with prompts probably not be running with prompts that you wrote. That is a crazy thing that you wrote. That is a crazy thing that you wrote. That is a crazy thing for me to say, because it was one of for me to say, because it was one of for me to say, because it was one of those like I never thought I I see the those like I never thought I I see the those like I never thought I I see the day things, but now that I've explored day things, but now that I've explored day things, but now that I've explored it myself and I've shipped a lot of code it myself and I've shipped a lot of code it myself and I've shipped a lot of code using these types of loops, using these types of loops, using these types of loops, I have a lot of thoughts I want to I have a lot of thoughts I want to I have a lot of thoughts I want to share. share. share. But I have one other thing I want to But I have one other thing I want to But I have one other thing I want to share quickly first, which is today's share quickly first, which is today's share quickly first, which is today's sponsor. AI should be good at design. It sponsor. AI should be good at design. It sponsor. AI should be good at design. It knows all the things it needs to about knows all the things it needs to about knows all the things it needs to about code, designs, visuals, and more. But code, designs, visuals, and more. But code, designs, visuals, and more. But every time I try to have it redesign every time I try to have it redesign every time I try to have it redesign things that I'm working on, it just things that I'm working on, it just things that I'm working on, it just doesn't do it right. At least that was doesn't do it right. At least that was doesn't do it right. At least that was my experience before I started using my experience before I started using my experience before I started using today's sponsor, Magic Patterns. These today's sponsor, Magic Patterns. These today's sponsor, Magic Patterns. These guys really cracked good design flows guys really cracked good design flows guys really cracked good design flows with AI. They're not trying to replace with AI. They're not trying to replace with AI. They're not trying to replace your whole stack or be a full site your whole stack or be a full site your whole stack or be a full site generator. They're trying to work within generator. They're trying to work within generator. They're trying to work within the real constraints of your real code the real constraints of your real code the real constraints of your real code base on just the front end in order to base on just the front end in order to base on just the front end in order to get great designs out. The first thing get great designs out. The first thing get great designs out. The first thing that makes them different is the design that makes them different is the design that makes them different is the design system selector. Unlike other tools that system selector. Unlike other tools that system selector. Unlike other tools that will just generate a bunch of slop code, will just generate a bunch of slop code, will just generate a bunch of slop code, you can pick between existing real base you can pick between existing real base you can pick between existing real base systems like the one that they provide, systems like the one that they provide, systems like the one that they provide, a wireframe system, or even classics a wireframe system, or even classics a wireframe system, or even classics like Shadcn, Chakra, Mantine, and MUI.

like Shadcn, Chakra, Mantine, and MUI. like Shadcn, Chakra, Mantine, and MUI. Or you can create your own, and you can Or you can create your own, and you can Or you can create your own, and you can also import things from Figma, too, also import things from Figma, too, also import things from Figma, too, which is super helpful. You can then which is super helpful. You can then which is super helpful. You can then switch between different models. switch between different models. switch between different models. Obviously, the ones that we all love and Obviously, the ones that we all love and Obviously, the ones that we all love and know are these in design like the Opus know are these in design like the Opus know are these in design like the Opus Line, Gemini 31, but also their auto Line, Gemini 31, but also their auto Line, Gemini 31, but also their auto router has really impressed me. It was router has really impressed me. It was router has really impressed me. It was able to grab the real SVGs for the logos able to grab the real SVGs for the logos able to grab the real SVGs for the logos for the things that I wanted to put on for the things that I wanted to put on for the things that I wanted to put on there once I showed them where they there once I showed them where they there once I showed them where they were. I could open up the preview and were. I could open up the preview and were. I could open up the preview and send this to other people on my team, send this to other people on my team, send this to other people on my team, which has already been super helpful. I which has already been super helpful. I which has already been super helpful. I can also leave comments on any point on can also leave comments on any point on can also leave comments on any point on the screen to tell the agent what else I the screen to tell the agent what else I the screen to tell the agent what else I wanted to fix, which has been a wanted to fix, which has been a wanted to fix, which has been a lifesaver when you're working on these lifesaver when you're working on these lifesaver when you're working on these types of things. They even have a visual types of things. They even have a visual types of things. They even have a visual editor for when you want to edit fonts, editor for when you want to edit fonts, editor for when you want to edit fonts, content, and things yourself. So, when content, and things yourself. So, when content, and things yourself. So, when you notice the agent's just not getting you notice the agent's just not getting you notice the agent's just not getting something right, don't fight it. Change something right, don't fight it. Change something right, don't fight it. Change it yourself. This is a small thing, but it yourself. This is a small thing, but it yourself. This is a small thing, but it's one of my favorites. The ability to it's one of my favorites. The ability to it's one of my favorites. The ability to choose different frames to test your choose different frames to test your choose different frames to test your site in to see how it looks on like a site in to see how it looks on like a site in to see how it looks on like a mobile display or an iPad display is so mobile display or an iPad display is so mobile display or an iPad display is so helpful as you're trying to get these helpful as you're trying to get these helpful as you're trying to get these fine-tuned pieces right. I can't tell fine-tuned pieces right. I can't tell fine-tuned pieces right. I can't tell you how many times I had a design that you how many times I had a design that you how many times I had a design that seemed good, but as soon as I shipped it seemed good, but as soon as I shipped it seemed good, but as soon as I shipped it and opened it on my phone, it was awful. and opened it on my phone, it was awful. and opened it on my phone, it was awful. No more. Just do it here. Starting to No more. Just do it here. Starting to No more. Just do it here. Starting to see why companies like DoorDash, Vappy, see why companies like DoorDash, Vappy, see why companies like DoorDash, Vappy, Granola, and more are leaning so hard on Granola, and more are leaning so hard on Granola, and more are leaning so hard on what Magic Patterns has built. These what Magic Patterns has built. These what Magic Patterns has built. These guys get it. Design better with AI at guys get it. Design better with AI at guys get it. Design better with AI at soidev.link/magicpatterns.

soidev.link/magicpatterns. soidev.link/magicpatterns. So, this post by Pete is the one that So, this post by Pete is the one that So, this post by Pete is the one that started this new era of looping started this new era of looping started this new era of looping discourse. But, this is not the tweet discourse. But, this is not the tweet discourse. But, this is not the tweet that got me to go try loops. It was this that got me to go try loops. It was this that got me to go try loops. It was this one. one. one. Here's a simple loop. Tell Codex to Here's a simple loop. Tell Codex to Here's a simple loop. Tell Codex to maintain your repos. Wake up every 5 maintain your repos. Wake up every 5 maintain your repos. Wake up every 5 minutes and direct work to threads. That minutes and direct work to threads. That minutes and direct work to threads. That makes it easy to parallelize and steer makes it easy to parallelize and steer makes it easy to parallelize and steer work as needed. He uses an orchestrator work as needed. He uses an orchestrator work as needed. He uses an orchestrator skill combined with his triaging and skill combined with his triaging and skill combined with his triaging and auto review and computer use skills. So, auto review and computer use skills. So, auto review and computer use skills. So, some work can land autonomously. This some work can land autonomously. This some work can land autonomously. This helped a lot click for me. In helped a lot click for me. In helped a lot click for me. In particular, of your agent directing work particular, of your agent directing work particular, of your agent directing work to threads. I didn't realize Codex had a to threads. I didn't realize Codex had a to threads. I didn't realize Codex had a feature where a thread in Codex could feature where a thread in Codex could feature where a thread in Codex could spin up another thread in Codex. And now spin up another thread in Codex. And now spin up another thread in Codex. And now that I know it has that, I've been that I know it has that, I've been that I know it has that, I've been pushing it much harder. I want to pushing it much harder. I want to pushing it much harder. I want to contextualize this in a bit of a weird contextualize this in a bit of a weird contextualize this in a bit of a weird way. I'm going to reference the article way. I'm going to reference the article way. I'm going to reference the article Anthropic did about recursive Anthropic did about recursive Anthropic did about recursive self-improvement because they did a self-improvement because they did a self-improvement because they did a great job describing how our work has great job describing how our work has great job describing how our work has changed over time. Previously, a person changed over time. Previously, a person changed over time. Previously, a person would use a computer and they eventually would use a computer and they eventually would use a computer and they eventually would use that to build a chatbot or an would use that to build a chatbot or an would use that to build a chatbot or an AI model. Once we had the AI model, the AI model. Once we had the AI model, the AI model. Once we had the AI model, the person could use the computer to ask the person could use the computer to ask the person could use the computer to ask the chatbot questions and get outputs that chatbot questions and get outputs that chatbot questions and get outputs that they could then use in their code to they could then use in their code to they could then use in their code to make better software and eventually make better software and eventually make better software and eventually maybe make a better model, too. But, the maybe make a better model, too. But, the maybe make a better model, too. But, the loop was the person uses the computer, loop was the person uses the computer, loop was the person uses the computer, asks the chatbot a question, it gives a asks the chatbot a question, it gives a asks the chatbot a question, it gives a result to the person who then result to the person who then result to the person who then copy-pasted it into their code, and then copy-pasted it into their code, and then copy-pasted it into their code, and then asks another question. I know a lot of asks another question. I know a lot of asks another question. I know a lot of people use stuff like my chat service T3 people use stuff like my chat service T3 people use stuff like my chat service T3 chat as a way to just do code, but they chat as a way to just do code, but they chat as a way to just do code, but they would bring it code questions and then would bring it code questions and then would bring it code questions and then copy-paste the answers. It really kind

copy-paste the answers. It really kind copy-paste the answers. It really kind of emphasized the whole like coding is of emphasized the whole like coding is of emphasized the whole like coding is just copy-paste meme. Chatbots pushed it just copy-paste meme. Chatbots pushed it just copy-paste meme. Chatbots pushed it way further, but now we've gone far way further, but now we've gone far way further, but now we've gone far beyond that because copy-pasting is not beyond that because copy-pasting is not beyond that because copy-pasting is not the best use of our time. So, instead of the best use of our time. So, instead of the best use of our time. So, instead of copy-pasting the result from the chatbot copy-pasting the result from the chatbot copy-pasting the result from the chatbot into our codebase, we started to just into our codebase, we started to just into our codebase, we started to just use our IDEs or terminals and other use our IDEs or terminals and other use our IDEs or terminals and other tools to talk to the model and get it to tools to talk to the model and get it to tools to talk to the model and get it to edit the code directly. And that's where edit the code directly. And that's where edit the code directly. And that's where things have been for a while now. But things have been for a while now. But things have been for a while now. But then we had another big change with then we had another big change with then we had another big change with workflows and sub-agents. I know a lot workflows and sub-agents. I know a lot workflows and sub-agents. I know a lot of people haven't even made this move of people haven't even made this move of people haven't even made this move yet, and I was hesitant to do it myself. yet, and I was hesitant to do it myself. yet, and I was hesitant to do it myself. Obviously, tools like Claude Code and Obviously, tools like Claude Code and Obviously, tools like Claude Code and Cursor will do some amount of this to go Cursor will do some amount of this to go Cursor will do some amount of this to go explore and find things in your code explore and find things in your code explore and find things in your code base, but the idea of telling my agent base, but the idea of telling my agent base, but the idea of telling my agent to spin up five agents to go break up to spin up five agents to go break up to spin up five agents to go break up work was something I just wasn't that work was something I just wasn't that work was something I just wasn't that interested in, especially when I saw all interested in, especially when I saw all interested in, especially when I saw all the crazy [ __ ] people were doing trying the crazy [ __ ] people were doing trying the crazy [ __ ] people were doing trying to create different personas and roles to create different personas and roles to create different personas and roles for all of those workers where they had for all of those workers where they had for all of those workers where they had a skill that wrote down in markdown a skill that wrote down in markdown a skill that wrote down in markdown files, "This is the adversarial files, "This is the adversarial files, "This is the adversarial reviewer. This is the security reviewer. reviewer. This is the security reviewer. reviewer. This is the security reviewer. This is the grokker and finder. This is This is the grokker and finder. This is This is the grokker and finder. This is the exploration agent." That made no the exploration agent." That made no the exploration agent." That made no [ __ ] sense, and I would argue that [ __ ] sense, and I would argue that [ __ ] sense, and I would argue that still makes no [ __ ] sense. The idea still makes no [ __ ] sense. The idea still makes no [ __ ] sense. The idea of predefining personas to go do things of predefining personas to go do things of predefining personas to go do things in your code base fundamentally misses in your code base fundamentally misses in your code base fundamentally misses the cool part of agents and AI as a the cool part of agents and AI as a the cool part of agents and AI as a whole. It's dynamic. The agent can build whole. It's dynamic. The agent can build whole. It's dynamic. The agent can build the context it needs and do the things the context it needs and do the things the context it needs and do the things it needs to without having everything it needs to without having everything it needs to without having everything pre-built and hard-coded ahead of time.

pre-built and hard-coded ahead of time. pre-built and hard-coded ahead of time. Imagine a coding template for a project Imagine a coding template for a project Imagine a coding template for a project where every file's already created and where every file's already created and where every file's already created and you have to edit things in the existing you have to edit things in the existing you have to edit things in the existing files. It's stupid, and that's how I files. It's stupid, and that's how I files. It's stupid, and that's how I felt about most of the sub-agent stuff felt about most of the sub-agent stuff felt about most of the sub-agent stuff that people were doing. Workflows pushed that people were doing. Workflows pushed that people were doing. Workflows pushed me hard here, and the video I just me hard here, and the video I just me hard here, and the video I just recently published about the things I recently published about the things I recently published about the things I like about Claude Code goes a little like about Claude Code goes a little like about Claude Code goes a little more in depth there on the things I like more in depth there on the things I like more in depth there on the things I like about workflows. The idea of your agents about workflows. The idea of your agents about workflows. The idea of your agents constructing the method that they're constructing the method that they're constructing the method that they're going to use to tackle a problem was going to use to tackle a problem was going to use to tackle a problem was really enticing to me. But now I'm going really enticing to me. But now I'm going really enticing to me. But now I'm going a bit further. Closing the loop, where a bit further. Closing the loop, where a bit further. Closing the loop, where the model doesn't just pick and spin up the model doesn't just pick and spin up the model doesn't just pick and spin up what sub-agents it needs, it audits the what sub-agents it needs, it audits the what sub-agents it needs, it audits the work it does and then sends the result work it does and then sends the result work it does and then sends the result back to run again back to run again back to run again and again, and again, and again. and again, and again, and again. and again, and again, and again. I am not at the fully autonomous loop I am not at the fully autonomous loop I am not at the fully autonomous loop point yet. I am not claiming the same point yet. I am not claiming the same point yet. I am not claiming the same things people like Boris are claiming things people like Boris are claiming things people like Boris are claiming where they're writing the loop, and now where they're writing the loop, and now where they're writing the loop, and now the code is just happening by itself the code is just happening by itself the code is just happening by itself with no oversight. That is stupid. But I with no oversight. That is stupid. But I with no oversight. That is stupid. But I wanted a taste. I wanted to get an idea wanted a taste. I wanted to get an idea wanted a taste. I wanted to get an idea of how this could work, so I could play of how this could work, so I could play of how this could work, so I could play with it myself and see what benefits with it myself and see what benefits with it myself and see what benefits exist. So, I started to play a bit. I exist. So, I started to play a bit. I exist. So, I started to play a bit. I started to do stuff like this. I had started to do stuff like this. I had started to do stuff like this. I had Claude code spin up a PR for a pretty Claude code spin up a PR for a pretty Claude code spin up a PR for a pretty big refactor. I used sub agents a bunch big refactor. I used sub agents a bunch big refactor. I used sub agents a bunch to go address specific concerns and take to go address specific concerns and take to go address specific concerns and take over specific parts of the code base. I over specific parts of the code base. I over specific parts of the code base. I didn't even say how to break it up. I didn't even say how to break it up. I didn't even say how to break it up. I let Opus figure that out itself. Man, I let Opus figure that out itself. Man, I let Opus figure that out itself. Man, I miss Mythos right now. But one specific miss Mythos right now. But one specific miss Mythos right now. But one specific thing I did do was tell the agent to thing I did do was tell the agent to thing I did do was tell the agent to monitor the PR for comments, because I

monitor the PR for comments, because I monitor the PR for comments, because I have a lot of awesome code review tools have a lot of awesome code review tools have a lot of awesome code review tools that are watching my PRs when they're that are watching my PRs when they're that are watching my PRs when they're filed and leaving feedback. And I moved filed and leaving feedback. And I moved filed and leaving feedback. And I moved away from copy-pasting code out of away from copy-pasting code out of away from copy-pasting code out of chatbots into my code base, and instead chatbots into my code base, and instead chatbots into my code base, and instead I found myself copy-pasting the comments I found myself copy-pasting the comments I found myself copy-pasting the comments that things like Code Rabbit, Graphtile, that things like Code Rabbit, Graphtile, that things like Code Rabbit, Graphtile, and Macroscopic would leave and pasting and Macroscopic would leave and pasting and Macroscopic would leave and pasting those into the agent so that it would go those into the agent so that it would go those into the agent so that it would go address them. It wasn't great. So, what address them. It wasn't great. So, what address them. It wasn't great. So, what I started doing instead, and this was I started doing instead, and this was I started doing instead, and this was the first step into heavier looping for the first step into heavier looping for the first step into heavier looping for me, and I would highly recommend you me, and I would highly recommend you me, and I would highly recommend you guys try the same cuz it's actually guys try the same cuz it's actually guys try the same cuz it's actually really cool. Once you have your setup in really cool. Once you have your setup in really cool. Once you have your setup in such a way where you have different work such a way where you have different work such a way where you have different work trees that are monitoring and working trees that are monitoring and working trees that are monitoring and working around specific pieces of work, where around specific pieces of work, where around specific pieces of work, where this code is in a directory that is this code is in a directory that is this code is in a directory that is specific to this PR. That means I don't specific to this PR. That means I don't specific to this PR. That means I don't care about this directory. It's not care about this directory. It's not care about this directory. It's not blocking other work. Once you have this blocking other work. Once you have this blocking other work. Once you have this broken out, in this case I'm SSHing to broken out, in this case I'm SSHing to broken out, in this case I'm SSHing to another machine on my network that is another machine on my network that is another machine on my network that is running this code base that has this running this code base that has this running this code base that has this fork of this code base, this work tree fork of this code base, this work tree fork of this code base, this work tree for it. And then I told it, "Monitor the for it. And then I told it, "Monitor the for it. And then I told it, "Monitor the comments. Watch the PR. Wait for comments. Watch the PR. Wait for comments. Watch the PR. Wait for comments to come in, and when they come comments to come in, and when they come comments to come in, and when they come in, address them."

in, address them." in, address them." And it did it. And it's been doing it And it did it. And it's been doing it And it did it. And it's been doing it now for like 6-plus hours. It has made a now for like 6-plus hours. It has made a now for like 6-plus hours. It has made a ton of improvements through this. And ton of improvements through this. And ton of improvements through this. And then I had a taste, and then I got then I had a taste, and then I got then I had a taste, and then I got really excited to play more. I wanted to really excited to play more. I wanted to really excited to play more. I wanted to push the limits push the limits push the limits of how much I could land without having of how much I could land without having of how much I could land without having to do the follow-up prompting myself. to do the follow-up prompting myself. to do the follow-up prompting myself. And I'll be honest, I still found myself And I'll be honest, I still found myself And I'll be honest, I still found myself hopping over the Codex and saying, "Hey, hopping over the Codex and saying, "Hey, hopping over the Codex and saying, "Hey, can you review this code?" and then can you review this code?" and then can you review this code?" and then copy-pasting the results of that review copy-pasting the results of that review copy-pasting the results of that review over. I played a little bit more there over. I played a little bit more there over. I played a little bit more there where I told Claude, "Hey, when you're where I told Claude, "Hey, when you're where I told Claude, "Hey, when you're done, run Codex with this command to get done, run Codex with this command to get done, run Codex with this command to get it to give feedback and then address it to give feedback and then address it to give feedback and then address what it gives as feedback." And that what it gives as feedback." And that what it gives as feedback." And that worked pretty well, too. But this is worked pretty well, too. But this is worked pretty well, too. But this is still for traditional work where I have still for traditional work where I have still for traditional work where I have one PR that does one thing that is being one PR that does one thing that is being one PR that does one thing that is being watched by my agent to address the watched by my agent to address the watched by my agent to address the comments that come in. There's a lot of comments that come in. There's a lot of comments that come in. There's a lot of work that can't be broken down into just work that can't be broken down into just work that can't be broken down into just one PR. I recently ran into one of those one PR. I recently ran into one of those one PR. I recently ran into one of those pieces of work. I have been rebuilding pieces of work. I have been rebuilding pieces of work. I have been rebuilding the isolate layer inside of Lakebed to the isolate layer inside of Lakebed to the isolate layer inside of Lakebed to make it a little more financially make it a little more financially make it a little more financially reasonable to deploy the way I want to reasonable to deploy the way I want to reasonable to deploy the way I want to deploy it. I did a deep dive on deploy it. I did a deep dive on deploy it. I did a deep dive on performance and alternative runtime performance and alternative runtime performance and alternative runtime options for how we could architect this options for how we could architect this options for how we could architect this with I5. And it had really good with I5. And it had really good with I5. And it had really good suggestions. But one of the things it suggestions. But one of the things it suggestions. But one of the things it pointed out was that my data pointed out was that my data pointed out was that my data architecture had a lot of room to architecture had a lot of room to architecture had a lot of room to improve that could help performance even improve that could help performance even improve that could help performance even more than runtime changes. Here's where more than runtime changes. Here's where more than runtime changes. Here's where it gave that feedback. The isolate it gave that feedback. The isolate it gave that feedback. The isolate architecture may not be the first architecture may not be the first architecture may not be the first scaling bottleneck. Current subscription scaling bottleneck. Current subscription scaling bottleneck. Current subscription validations rerun every query

validations rerun every query validations rerun every query subscription for an app after each subscription for an app after each subscription for an app after each mutation. For hot apps, we should mutation. For hot apps, we should mutation. For hot apps, we should implement dependency-aware invalidation, implement dependency-aware invalidation, implement dependency-aware invalidation, mutation coalescing, per-app mutation coalescing, per-app mutation coalescing, per-app invalidating batches, shared results for invalidating batches, shared results for invalidating batches, shared results for identical subscription arguments, and identical subscription arguments, and identical subscription arguments, and back-pressuring maximum refresh back-pressuring maximum refresh back-pressuring maximum refresh frequency. This is when I realized there frequency. This is when I realized there frequency. This is when I realized there was a lot of work that needed to be was a lot of work that needed to be was a lot of work that needed to be done. So I asked up front, "From these done. So I asked up front, "From these done. So I asked up front, "From these features you think we should implement, features you think we should implement, features you think we should implement, which should be done separately and which should be done separately and which should be done separately and which should be done in tandem? Would it which should be done in tandem? Would it which should be done in tandem? Would it be realistic to do all of this in one be realistic to do all of this in one be realistic to do all of this in one PR?" It very quickly said, "No, I would PR?" It very quickly said, "No, I would PR?" It very quickly said, "No, I would not implement all of this. It's one not implement all of this. It's one not implement all of this. It's one project, but at least three PRs." project, but at least three PRs." project, but at least three PRs." Current implementation synchronously Current implementation synchronously Current implementation synchronously yada yada. And then it broke up what the yada yada. And then it broke up what the yada yada. And then it broke up what the different PRs could look like. I asked, different PRs could look like. I asked, different PRs could look like. I asked, "Should they be worked on separately or "Should they be worked on separately or "Should they be worked on separately or should they be stacked?" It said they should they be stacked?" It said they should they be stacked?" It said they should be mostly stacked, but there's should be mostly stacked, but there's should be mostly stacked, but there's some opportunity for parallelizing. some opportunity for parallelizing. some opportunity for parallelizing. I then told it to write an HTML plan, my I then told it to write an HTML plan, my I then told it to write an HTML plan, my beloved. Thank you again to our friend beloved. Thank you again to our friend beloved. Thank you again to our friend Thoric for introducing me to this Thoric for introducing me to this Thoric for introducing me to this wonderful pattern where it is so much wonderful pattern where it is so much wonderful pattern where it is so much easier to see what my agents want to do easier to see what my agents want to do easier to see what my agents want to do and read it in a way that I can even and read it in a way that I can even and read it in a way that I can even open on my phone. It's so nice. And it open on my phone. It's so nice. And it open on my phone. It's so nice. And it wrote these plans for each of the wrote these plans for each of the wrote these plans for each of the portions that it needed to complete. I portions that it needed to complete. I portions that it needed to complete. I also told it here after the plan, like also told it here after the plan, like also told it here after the plan, like please make the plans piece, to create a please make the plans piece, to create a please make the plans piece, to create a new thread with the first plan as a new thread with the first plan as a new thread with the first plan as a starting point. And it did. It created a starting point. And it did. It created a starting point. And it did. It created a PR by itself in a new thread to go PR by itself in a new thread to go PR by itself in a new thread to go implement that first plan. And then it implement that first plan. And then it implement that first plan. And then it landed. I did my usual thing where I had landed. I did my usual thing where I had landed. I did my usual thing where I had a bunch of back and forth review. I spun a bunch of back and forth review. I spun a bunch of back and forth review. I spun up another thread to review it. I copy up another thread to review it. I copy up another thread to review it. I copy pasted back and forth. It got into a pasted back and forth. It got into a pasted back and forth. It got into a good state. So I merged it. I then asked good state. So I merged it. I then asked good state. So I merged it. I then asked it to make a fresh thread for part two,

it to make a fresh thread for part two, it to make a fresh thread for part two, which it did. It only took a few which it did. It only took a few which it did. It only took a few seconds. But I realized I should be seconds. But I realized I should be seconds. But I realized I should be looping harder. This is the single looping harder. This is the single looping harder. This is the single message I have sent to an agent that has message I have sent to an agent that has message I have sent to an agent that has impacted my psychosis the most. Would it impacted my psychosis the most. Would it impacted my psychosis the most. Would it be possible to make a workflow of some be possible to make a workflow of some be possible to make a workflow of some form that first will spin up a separate form that first will spin up a separate form that first will spin up a separate thread to make the PR, second spin up thread to make the PR, second spin up thread to make the PR, second spin up another thread to review that PR when another thread to review that PR when another thread to review that PR when it's filed, three puts the thread from it's filed, three puts the thread from it's filed, three puts the thread from one in a loop reviewing comments until one in a loop reviewing comments until one in a loop reviewing comments until it gets all approvals, and then fourth it gets all approvals, and then fourth it gets all approvals, and then fourth the thread would merge the PR and the thread would merge the PR and the thread would merge the PR and trigger another one for the next piece. trigger another one for the next piece. trigger another one for the next piece. I didn't think it'd be able to do this, I didn't think it'd be able to do this, I didn't think it'd be able to do this, but I was curious how it would try. And but I was curious how it would try. And but I was curious how it would try. And it made a kind of broken diagram showing it made a kind of broken diagram showing it made a kind of broken diagram showing the workflow it had in mind. It said it the workflow it had in mind. It said it the workflow it had in mind. It said it would use a heartbeat attached to this would use a heartbeat attached to this would use a heartbeat attached to this thread pulling every 5 to 10 minutes. On thread pulling every 5 to 10 minutes. On thread pulling every 5 to 10 minutes. On each wake up it would read the each wake up it would read the each wake up it would read the implementation thread status, detect implementation thread status, detect implementation thread status, detect filed PRs, create a fresh review thread filed PRs, create a fresh review thread filed PRs, create a fresh review thread when a new PR has a new sha head, send when a new PR has a new sha head, send when a new PR has a new sha head, send actionable findings back, re-review actionable findings back, re-review actionable findings back, re-review after the fixes are pushed, yada yada, after the fixes are pushed, yada yada, after the fixes are pushed, yada yada, and then pull latest main before and then pull latest main before and then pull latest main before creating the next work tree. So I said, creating the next work tree. So I said, creating the next work tree. So I said, "Make the workflow and use it to file "Make the workflow and use it to file "Make the workflow and use it to file the remaining PRs." And it did it. This the remaining PRs." And it did it. This the remaining PRs." And it did it. This was Sunday at 2:29 a.m. And it was Sunday at 2:29 a.m. And it was Sunday at 2:29 a.m. And it eventually finished and broke everything eventually finished and broke everything eventually finished and broke everything in my editor pretty aggressively at 6:50 in my editor pretty aggressively at 6:50 in my editor pretty aggressively at 6:50 a.m.

a.m. a.m. I set this off before going to bed, and I set this off before going to bed, and I set this off before going to bed, and I woke up the next day with four stacked I woke up the next day with four stacked I woke up the next day with four stacked PRs PRs PRs reviewed to hell and back reviewed to hell and back reviewed to hell and back all merged. It was [ __ ] awesome. Do I all merged. It was [ __ ] awesome. Do I all merged. It was [ __ ] awesome. Do I think you should do this on real think you should do this on real think you should do this on real production code bases that have millions production code bases that have millions production code bases that have millions of users? Probably not. At least not of users? Probably not. At least not of users? Probably not. At least not yet. But god damn is it cool to spin up yet. But god damn is it cool to spin up yet. But god damn is it cool to spin up work in this way where complex work in this way where complex work in this way where complex multi-stage problems that need their own multi-stage problems that need their own multi-stage problems that need their own breakdowns, that need their own pull breakdowns, that need their own pull breakdowns, that need their own pull requests, that need their own reviews requests, that need their own reviews requests, that need their own reviews and cycles and loops. Because that's the and cycles and loops. Because that's the and cycles and loops. Because that's the craziest thing here. I asked the model craziest thing here. I asked the model craziest thing here. I asked the model if I could make this loop, and it made a if I could make this loop, and it made a if I could make this loop, and it made a loop that makes sub loops dynamically. loop that makes sub loops dynamically. loop that makes sub loops dynamically. This isn't a hard-coded, every time I This isn't a hard-coded, every time I This isn't a hard-coded, every time I make a change I spin up one reviewer make a change I spin up one reviewer make a change I spin up one reviewer that reviews it and then they go back that reviews it and then they go back that reviews it and then they go back and forth. This is a dynamic workflow and forth. This is a dynamic workflow and forth. This is a dynamic workflow that was created based on the specific that was created based on the specific that was created based on the specific needs of this specific problem I was needs of this specific problem I was needs of this specific problem I was solving. My loops created loops, and solving. My loops created loops, and solving. My loops created loops, and they did a great job at it. This was they did a great job at it. This was they did a great job at it. This was real code that landed and sadly I real code that landed and sadly I real code that landed and sadly I couldn't have Fable come in and review couldn't have Fable come in and review couldn't have Fable come in and review it because this was after the ban. But it because this was after the ban. But it because this was after the ban. But the idea of your agents being able to the idea of your agents being able to the idea of your agents being able to orchestrate dynamic work in a way that orchestrate dynamic work in a way that orchestrate dynamic work in a way that is specifically tailored to the problem is specifically tailored to the problem is specifically tailored to the problem is so cool. Throughout most of my is so cool. Throughout most of my is so cool. Throughout most of my career, when I worked at real companies, career, when I worked at real companies, career, when I worked at real companies, we would follow some form of the we would follow some form of the we would follow some form of the traditional Agile sprint loop. Where we traditional Agile sprint loop. Where we traditional Agile sprint loop. Where we would put tickets inside of our backlog, would put tickets inside of our backlog, would put tickets inside of our backlog, and then once every week or two weeks, and then once every week or two weeks, and then once every week or two weeks, the start of the week, we would pull up

the start of the week, we would pull up the start of the week, we would pull up the backlog and decide what was worth the backlog and decide what was worth the backlog and decide what was worth working on and how long we thought it working on and how long we thought it working on and how long we thought it would take, and then try to make sure would take, and then try to make sure would take, and then try to make sure work that's blocking other work was work that's blocking other work was work that's blocking other work was prioritized accordingly, that everybody prioritized accordingly, that everybody prioritized accordingly, that everybody had unblocked work to do. But the actual had unblocked work to do. But the actual had unblocked work to do. But the actual flow of all of this was pretty static. flow of all of this was pretty static. flow of all of this was pretty static. It was the classic agile waterfall-y It was the classic agile waterfall-y It was the classic agile waterfall-y structure. And we kind of had to force structure. And we kind of had to force structure. And we kind of had to force our work to fit that shape. The most our work to fit that shape. The most our work to fit that shape. The most productive teams were the ones that productive teams were the ones that productive teams were the ones that would build their own alternative shape would build their own alternative shape would build their own alternative shape around the problems they were trying to around the problems they were trying to around the problems they were trying to solve. That is what makes this so cool. solve. That is what makes this so cool. solve. That is what makes this so cool. The shape of the loop, the shape of the The shape of the loop, the shape of the The shape of the loop, the shape of the structure, the shape of how work happens structure, the shape of how work happens structure, the shape of how work happens can be dynamically generated based on can be dynamically generated based on can be dynamically generated based on the shape of the work that you're doing. the shape of the work that you're doing. the shape of the work that you're doing. And you can use this for all sorts of And you can use this for all sorts of And you can use this for all sorts of crazy stuff. You can use this to monitor crazy stuff. You can use this to monitor crazy stuff. You can use this to monitor pull requests that need to be merged. pull requests that need to be merged. pull requests that need to be merged. You can use this on a schedule to every You can use this on a schedule to every You can use this on a schedule to every morning start your day with feedback on morning start your day with feedback on morning start your day with feedback on what PR's are worth merging and what what PR's are worth merging and what what PR's are worth merging and what ones are worth forgetting about. I use ones are worth forgetting about. I use ones are worth forgetting about. I use this type of thinking to find the best this type of thinking to find the best this type of thinking to find the best solution for a 5G hotspot. And since I solution for a 5G hotspot. And since I solution for a 5G hotspot. And since I had a loop checking what the best deals had a loop checking what the best deals had a loop checking what the best deals were, I got early information about the were, I got early information about the were, I got early information about the new Verizon plan they just put out new Verizon plan they just put out new Verizon plan they just put out because my loop pointed it out to me because my loop pointed it out to me because my loop pointed it out to me randomly on Discord. It's so cool. And randomly on Discord. It's so cool. And randomly on Discord. It's so cool. And again, to my earlier point, I wrote a again, to my earlier point, I wrote a again, to my earlier point, I wrote a handful of prompts in this thread. I handful of prompts in this thread. I handful of prompts in this thread. I wrote most of the prompts. Actually, no wrote most of the prompts. Actually, no wrote most of the prompts. Actually, no I didn't cuz it got in that schedule I didn't cuz it got in that schedule I didn't cuz it got in that schedule after. But up until the schedule after. But up until the schedule after. But up until the schedule started, I wrote all the prompts and I started, I wrote all the prompts and I started, I wrote all the prompts and I read the responses. I said, "Yeah, that read the responses. I said, "Yeah, that read the responses. I said, "Yeah, that sounds good. Let's see what happens."

sounds good. Let's see what happens." sounds good. Let's see what happens." And then I did see what happened. And And then I did see what happened. And And then I did see what happened. And what happened was kind of [ __ ] what happened was kind of [ __ ] what happened was kind of [ __ ] awesome. So, what I would highly awesome. So, what I would highly awesome. So, what I would highly recommend you do here, the info you take recommend you do here, the info you take recommend you do here, the info you take from here, is to think about the work from here, is to think about the work from here, is to think about the work you do before, during, and after you you do before, during, and after you you do before, during, and after you prompt your agent. When your agent prompt your agent. When your agent prompt your agent. When your agent completes its task, pay attention to completes its task, pay attention to completes its task, pay attention to what you do next. For me, what would what you do next. For me, what would what you do next. For me, what would happen is I would tell the agent to happen is I would tell the agent to happen is I would tell the agent to build the thing. And then once it built build the thing. And then once it built build the thing. And then once it built it, I would run the thing and go see if it, I would run the thing and go see if it, I would run the thing and go see if it worked. And if it did, I would commit it worked. And if it did, I would commit it worked. And if it did, I would commit the thing and then push the thing and the thing and then push the thing and the thing and then push the thing and then make a pull request on GitHub for then make a pull request on GitHub for then make a pull request on GitHub for the thing. I would then wait for my code the thing. I would then wait for my code the thing. I would then wait for my code review agents to get feedback. I would review agents to get feedback. I would review agents to get feedback. I would address that feedback. I would then ask address that feedback. I would then ask address that feedback. I would then ask my team for feedback. I would address my team for feedback. I would address my team for feedback. I would address that feedback. And then I would merge that feedback. And then I would merge that feedback. And then I would merge it. Start from where you started there. it. Start from where you started there. it. Start from where you started there. The first thing I did after the changes The first thing I did after the changes The first thing I did after the changes were completed was run a dev server. were completed was run a dev server. were completed was run a dev server. Tell the agent to do that. I then Tell the agent to do that. I then Tell the agent to do that. I then checked if the work worked. Tell the checked if the work worked. Tell the checked if the work worked. Tell the agent to do that, too. Computer use has agent to do that, too. Computer use has agent to do that, too. Computer use has gotten really good. After I verified the gotten really good. After I verified the gotten really good. After I verified the work, I would then commit. Tell the work, I would then commit. Tell the work, I would then commit. Tell the agent to do that once it's verified agent to do that once it's verified agent to do that once it's verified things are correct. Tell it to push up things are correct. Tell it to push up things are correct. Tell it to push up the code and file a PR once it's ready.

the code and file a PR once it's ready. the code and file a PR once it's ready. Then I would go get those code review Then I would go get those code review Then I would go get those code review comments and copy-paste them into the comments and copy-paste them into the comments and copy-paste them into the agent to fix. Tell the agent to do that agent to fix. Tell the agent to do that agent to fix. Tell the agent to do that itself, too. Maybe tell the agent to itself, too. Maybe tell the agent to itself, too. Maybe tell the agent to spin up other threads to do its own spin up other threads to do its own spin up other threads to do its own reviews. The other spicier way of reviews. The other spicier way of reviews. The other spicier way of putting this is that we are looking at putting this is that we are looking at putting this is that we are looking at the code too early. If you are reading the code too early. If you are reading the code too early. If you are reading the code your agent put out before the code your agent put out before the code your agent put out before another agent read it and gave feedback another agent read it and gave feedback another agent read it and gave feedback on it, you're wasting your own time. on it, you're wasting your own time. on it, you're wasting your own time. That's time that the agent could have That's time that the agent could have That's time that the agent could have spent instead that you could have used spent instead that you could have used spent instead that you could have used to find other work worth doing or to to find other work worth doing or to to find other work worth doing or to relax a little or go spin up a side relax a little or go spin up a side relax a little or go spin up a side project. I don't know what you're going project. I don't know what you're going project. I don't know what you're going to do with your free time, but I have to do with your free time, but I have to do with your free time, but I have had far too many instances where I read had far too many instances where I read had far too many instances where I read an agent code and was like, "That's an agent code and was like, "That's an agent code and was like, "That's obviously wrong." and then told it to go obviously wrong." and then told it to go obviously wrong." and then told it to go fix it. They can figure that [ __ ] out fix it. They can figure that [ __ ] out fix it. They can figure that [ __ ] out themselves, too. And now, when the human themselves, too. And now, when the human themselves, too. And now, when the human comes in, all the [ __ ] is gone and comes in, all the [ __ ] is gone and comes in, all the [ __ ] is gone and you can focus on the hard stuff. It's so you can focus on the hard stuff. It's so you can focus on the hard stuff. It's so much more fun. Try to find where you much more fun. Try to find where you much more fun. Try to find where you have to be involved and see what it have to be involved and see what it have to be involved and see what it takes to prompt yourself out of it. I'm takes to prompt yourself out of it. I'm takes to prompt yourself out of it. I'm not saying you need a bunch of custom not saying you need a bunch of custom not saying you need a bunch of custom skills. I have almost none here. I'm not skills. I have almost none here. I'm not skills. I have almost none here. I'm not saying that you need to build fancy saying that you need to build fancy saying that you need to build fancy plugins or install a bunch of [ __ ] I'm plugins or install a bunch of [ __ ] I'm plugins or install a bunch of [ __ ] I'm just using stock Codex. I'm not even just using stock Codex. I'm not even just using stock Codex. I'm not even using T3 code for this. I do hope to get using T3 code for this. I do hope to get using T3 code for this. I do hope to get these features added to T3 code soon cuz these features added to T3 code soon cuz these features added to T3 code soon cuz they're really cool, but I'm just using they're really cool, but I'm just using they're really cool, but I'm just using stock Codex with a normal account here.

stock Codex with a normal account here. stock Codex with a normal account here. There is one catch, though. There is one catch, though. There is one catch, though. Cost. Cost. Cost. You will burn many more tokens when you You will burn many more tokens when you You will burn many more tokens when you run things in loops like this. And if run things in loops like this. And if run things in loops like this. And if it's going down the wrong path, it might it's going down the wrong path, it might it's going down the wrong path, it might go down that wrong path for longer to go down that wrong path for longer to go down that wrong path for longer to burn more tokens and potentially cost burn more tokens and potentially cost burn more tokens and potentially cost you more money. If you're paying API you more money. If you're paying API you more money. If you're paying API prices, you probably shouldn't be doing prices, you probably shouldn't be doing prices, you probably shouldn't be doing loops yet. That said, you might be loops yet. That said, you might be loops yet. That said, you might be surprised how far you can go with them. surprised how far you can go with them. surprised how far you can go with them. Remember that loop I mentioned earlier Remember that loop I mentioned earlier Remember that loop I mentioned earlier that I was using Opus and Claude code that I was using Opus and Claude code that I was using Opus and Claude code for where it's watching the PR and for where it's watching the PR and for where it's watching the PR and updating it constantly? Not only is it updating it constantly? Not only is it updating it constantly? Not only is it doing that, I've noticed that every time doing that, I've noticed that every time doing that, I've noticed that every time it gets feedback, it spins up a workflow it gets feedback, it spins up a workflow it gets feedback, it spins up a workflow with eight steps or more to address all with eight steps or more to address all with eight steps or more to address all of it. I had one agent spend under 10 of it. I had one agent spend under 10 of it. I had one agent spend under 10 minutes leaving feedback, and based on minutes leaving feedback, and based on minutes leaving feedback, and based on that feedback, the Opus workflow ran for that feedback, the Opus workflow ran for that feedback, the Opus workflow ran for eight hours and did over 3 million eight hours and did over 3 million eight hours and did over 3 million tokens down to address like three small tokens down to address like three small tokens down to address like three small comments. It was brutal. It was absurd. comments. It was brutal. It was absurd. comments. It was brutal. It was absurd. If I was a blocked during that time, it If I was a blocked during that time, it If I was a blocked during that time, it would have been very rough. And would have been very rough. And would have been very rough. And honestly, I was kind of blocked at that honestly, I was kind of blocked at that honestly, I was kind of blocked at that point cuz this is a big overhaul and I point cuz this is a big overhaul and I point cuz this is a big overhaul and I want this in before doing other changes want this in before doing other changes want this in before doing other changes cuz I'm unfucking the the TypeScript cuz I'm unfucking the the TypeScript cuz I'm unfucking the the TypeScript that looks like Python that GPT 5.5 that looks like Python that GPT 5.5 that looks like Python that GPT 5.5 wrote. Cuz as great as the model is at wrote. Cuz as great as the model is at wrote. Cuz as great as the model is at writing code that functions, it does not writing code that functions, it does not writing code that functions, it does not write code I like looking at. And write code I like looking at. And write code I like looking at. And product models write better looking product models write better looking product models write better looking code. I wanted to do this with Fable.

code. I wanted to do this with Fable. code. I wanted to do this with Fable. Fable was taken. So instead, I burned a Fable was taken. So instead, I burned a Fable was taken. So instead, I burned a shitload of Opus tokens. shitload of Opus tokens. shitload of Opus tokens. This thread that's like breaking my SSH This thread that's like breaking my SSH This thread that's like breaking my SSH and Claude code. I can't even scroll up and Claude code. I can't even scroll up and Claude code. I can't even scroll up far enough to get to my first prompt cuz far enough to get to my first prompt cuz far enough to get to my first prompt cuz this thread is just so much [ __ ] going this thread is just so much [ __ ] going this thread is just so much [ __ ] going on, very little of which has involved me on, very little of which has involved me on, very little of which has involved me at all. at all. at all. So how was my usage? I do have two So how was my usage? I do have two So how was my usage? I do have two Claude code accounts right now, so I'm Claude code accounts right now, so I'm Claude code accounts right now, so I'm sure this burned through it really sure this burned through it really sure this burned through it really aggressively, right? aggressively, right? aggressively, right? Well, this combined with everything else Well, this combined with everything else Well, this combined with everything else I've been working on for the last few I've been working on for the last few I've been working on for the last few days using Opus days using Opus days using Opus still has me at only 29% of my weekly still has me at only 29% of my weekly still has me at only 29% of my weekly limit which expires in eight hours. I limit which expires in eight hours. I limit which expires in eight hours. I was maxing out my limits with Fable. And was maxing out my limits with Fable. And was maxing out my limits with Fable. And with Opus in a loop like this, I'm not with Opus in a loop like this, I'm not with Opus in a loop like this, I'm not even close even close even close to getting my limits. And I've had like to getting my limits. And I've had like to getting my limits. And I've had like five of these types of loops running in five of these types of loops running in five of these types of loops running in that time. Really big piles of changes that time. Really big piles of changes that time. Really big piles of changes happening. happening. happening. And it doesn't [ __ ] matter. It's not And it doesn't [ __ ] matter. It's not And it doesn't [ __ ] matter. It's not getting close to my limits. I am on the getting close to my limits. I am on the getting close to my limits. I am on the $200 plan. I will also say that I ran a $200 plan. I will also say that I ran a $200 plan. I will also say that I ran a workflow using the new Claude code with workflow using the new Claude code with workflow using the new Claude code with Opus 4.8 when it came out on the $100 Opus 4.8 when it came out on the $100 Opus 4.8 when it came out on the $100 plan and I hit the 5-hour limit plan and I hit the 5-hour limit plan and I hit the 5-hour limit instantaneously. I have never come close instantaneously. I have never come close instantaneously. I have never come close to the 5-hour limit with Opus and loops.

to the 5-hour limit with Opus and loops. to the 5-hour limit with Opus and loops. And I'm also not coming close to the And I'm also not coming close to the And I'm also not coming close to the weekly limit with it either on that $200 weekly limit with it either on that $200 weekly limit with it either on that $200 plan. So, if you're already on a $200 plan. So, if you're already on a $200 plan. So, if you're already on a $200 plan or you're willing to be on one and plan or you're willing to be on one and plan or you're willing to be on one and you find that your usage is not getting you find that your usage is not getting you find that your usage is not getting like lethal, like you're not getting like lethal, like you're not getting like lethal, like you're not getting close to maxing out, start looping more. close to maxing out, start looping more. close to maxing out, start looping more. And since you can't use these plans that And since you can't use these plans that And since you can't use these plans that normal companies usually because of the normal companies usually because of the normal companies usually because of the differences and restrictions and how differences and restrictions and how differences and restrictions and how you're supposed to use an enterprise you're supposed to use an enterprise you're supposed to use an enterprise plan at API prices, go use this for plan at API prices, go use this for plan at API prices, go use this for crazy [ __ ] that you don't think should crazy [ __ ] that you don't think should crazy [ __ ] that you don't think should be possible. I would also recommend be possible. I would also recommend be possible. I would also recommend experimenting with the tools that are experimenting with the tools that are experimenting with the tools that are included with our harnesses now. A lot included with our harnesses now. A lot included with our harnesses now. A lot of them are pretty powerful. Codex's of them are pretty powerful. Codex's of them are pretty powerful. Codex's ability to spin up new threads is ability to spin up new threads is ability to spin up new threads is really, really cool. Both Codex and really, really cool. Both Codex and really, really cool. Both Codex and Claude code have a {slash} goal Claude code have a {slash} goal Claude code have a {slash} goal primitive, which allows you to get one primitive, which allows you to get one primitive, which allows you to get one thread going forever on a task where it thread going forever on a task where it thread going forever on a task where it keeps double-checking at the end of a keeps double-checking at the end of a keeps double-checking at the end of a turn, did you finish the work? If no, turn, did you finish the work? If no, turn, did you finish the work? If no, okay, keep going. That type of like okay, keep going. That type of like okay, keep going. That type of like linear, never-ending loop is different linear, never-ending loop is different linear, never-ending loop is different from a dynamic workflow like I showed from a dynamic workflow like I showed from a dynamic workflow like I showed earlier where it creates dynamic work earlier where it creates dynamic work earlier where it creates dynamic work based on a pre-planned goal versus a based on a pre-planned goal versus a based on a pre-planned goal versus a traditional {slash} goal where it just traditional {slash} goal where it just traditional {slash} goal where it just keeps plugging along on that one thread keeps plugging along on that one thread keeps plugging along on that one thread until it completes. I have a goal until it completes. I have a goal until it completes. I have a goal running right now that's over 12 hours running right now that's over 12 hours running right now that's over 12 hours in that's trying to rewrite Hermes agent in that's trying to rewrite Hermes agent in that's trying to rewrite Hermes agent in Rust so that I can run it in isolates in Rust so that I can run it in isolates in Rust so that I can run it in isolates that are much smaller and use less that are much smaller and use less that are much smaller and use less resources cuz my Hermes agent uses over resources cuz my Hermes agent uses over resources cuz my Hermes agent uses over a gig of RAM. It is getting close. It'll a gig of RAM. It is getting close. It'll a gig of RAM. It is getting close. It'll probably work. It probably won't be probably work. It probably won't be probably work. It probably won't be production ready. It probably won't be production ready. It probably won't be production ready. It probably won't be something I want to put out there and something I want to put out there and something I want to put out there and sell or anything, but it's a fun use of sell or anything, but it's a fun use of sell or anything, but it's a fun use of my spare tokens and it's really my spare tokens and it's really my spare tokens and it's really interesting to see what types of interesting to see what types of interesting to see what types of problems can be solved when you throw problems can be solved when you throw problems can be solved when you throw these crazy rate limits at them. The

these crazy rate limits at them. The these crazy rate limits at them. The point I'm trying to make here is that point I'm trying to make here is that point I'm trying to make here is that you should be treating these limits like you should be treating these limits like you should be treating these limits like challenges. If you're on the expensive challenges. If you're on the expensive challenges. If you're on the expensive plan, you should be trying to get close plan, you should be trying to get close plan, you should be trying to get close to maxing it out because that's just to maxing it out because that's just to maxing it out because that's just money you're losing if you're not. The money you're losing if you're not. The money you're losing if you're not. The 70% I'm not going to hit in my weekly 70% I'm not going to hit in my weekly 70% I'm not going to hit in my weekly limit here at 8 hours is thousands of limit here at 8 hours is thousands of limit here at 8 hours is thousands of dollars of inference that I paid for dollars of inference that I paid for dollars of inference that I paid for that I could have done that I didn't do. that I could have done that I didn't do. that I could have done that I didn't do. But again, I need to be realistic with But again, I need to be realistic with But again, I need to be realistic with you guys. you guys. you guys. In all of May on this computer, I did In all of May on this computer, I did In all of May on this computer, I did about $1,900 about $1,900 about $1,900 of inference. I didn't pay that of inference. I didn't pay that of inference. I didn't pay that obviously cuz I'm using the obviously cuz I'm using the obviously cuz I'm using the subscriptions with Claude Code and subscriptions with Claude Code and subscriptions with Claude Code and Codex. But this month, June, which we're Codex. But this month, June, which we're Codex. But this month, June, which we're only 17 days into, I'm at nearly $6,000 only 17 days into, I'm at nearly $6,000 only 17 days into, I'm at nearly $6,000 of usage. of usage. of usage. But that's just this computer. But that's just this computer. But that's just this computer. As I mentioned earlier, I'm using As I mentioned earlier, I'm using As I mentioned earlier, I'm using multiple computers. multiple computers. multiple computers. My Mac mini has another $2,600 My Mac mini has another $2,600 My Mac mini has another $2,600 of inference on it. I'm at 10 grand for of inference on it. I'm at 10 grand for of inference on it. I'm at 10 grand for the month across all of my machines. And the month across all of my machines. And the month across all of my machines. And that's on three of those $200 plans, two that's on three of those $200 plans, two that's on three of those $200 plans, two Claude Code, one Codex, and I haven't Claude Code, one Codex, and I haven't Claude Code, one Codex, and I haven't used the second Claude Code account used the second Claude Code account used the second Claude Code account since Fable was taken from us. That's a since Fable was taken from us. That's a since Fable was taken from us. That's a shitload of value that I'm getting given shitload of value that I'm getting given shitload of value that I'm getting given for relatively cheaply. To spend $600 for relatively cheaply. To spend $600 for relatively cheaply. To spend $600 and get back 10 grand of inference, that and get back 10 grand of inference, that and get back 10 grand of inference, that means you can do a lot. And if you're means you can do a lot. And if you're means you can do a lot. And if you're not pushing loops to their limits, not pushing loops to their limits, not pushing loops to their limits, you're not using that as much as you you're not using that as much as you you're not using that as much as you could be. I've been having way more fun could be. I've been having way more fun could be. I've been having way more fun with loops than I expected to, and I'm with loops than I expected to, and I'm with loops than I expected to, and I'm curious if you guys will as well. Take a curious if you guys will as well. Take a curious if you guys will as well. Take a look at what you do when you're done look at what you do when you're done look at what you do when you're done prompting, see what additional steps you prompting, see what additional steps you prompting, see what additional steps you take, and ask the model, can you do

take, and ask the model, can you do take, and ask the model, can you do this? You might be surprised at what this? You might be surprised at what this? You might be surprised at what it's capable of. I know for a fact that it's capable of. I know for a fact that it's capable of. I know for a fact that I was very surprised myself. What I'm I was very surprised myself. What I'm I was very surprised myself. What I'm trying to say here is that loops are trying to say here is that loops are trying to say here is that loops are cool not because the technology or the cool not because the technology or the cool not because the technology or the mindset's really cool, but the idea of mindset's really cool, but the idea of mindset's really cool, but the idea of letting agents do more is unbelievably letting agents do more is unbelievably letting agents do more is unbelievably powerful. If you take anything from this powerful. If you take anything from this powerful. If you take anything from this video, it really should be that. Ask video, it really should be that. Ask video, it really should be that. Ask your agent to do the next step and see your agent to do the next step and see your agent to do the next step and see if it impresses you. I know it impressed if it impresses you. I know it impressed if it impresses you. I know it impressed me. me. me. Let me know how it goes, and until next Let me know how it goes, and until next Let me know how it goes, and until next time, time, time, peace, nerds.