I'm done with terminals transcript

This is a video I never wanted to make, but the more I think about it, the more I feel like I kind of have to because if I'm being realistic with you guys, I've always been a terminal first person. I know, I know I'm the vibe coding soy boy JavaScript guy that's always plugging things in VS Code and invested in a VS Code fork, but I am not that at heart. I do like being able to scroll around code and click and select things with a trackpad or a mouse, but generally for my day-to-day work, I was always in a terminal.

I've been a big terminal nerd since before I even wrote code. I was using them to host my Minecraft servers, to SSH into computers all over the world and do all sorts of fun, crazy things. I had a deeply modified GNU screen config that I put a lot of time into before I started moving over to T-Mox because I'm old enough that GNU screen was still the standard when I started. I have been using terminals for the majority of my life.

Since I was 13, I've spent most of my time running, working, auditing, and doing my day-to-day stuff inside of my terminal. And that's just not the case anymore. I had a moment a few days ago where I was getting a bunch of work done and I started commanding to try and find my terminal and I couldn't find it because I had done a reboot and not opened up my terminal because I didn't need it for like eight plus hours. And that's when it hit me.

I'm not doing my work in a terminal anymore. And I really, really need to talk about why. Before I can do that, we're gonna have to take a quick break for today's sponsor. When I'm working on a new project and it's time to go live, the first thing I do every single time is I put it up on GitHub. But right after that, I use today's sponsor, Blacksmith, because as soon as I want real CI, I don't want to deal with GitHub actions anymore.

If all I got from Blacksmith was the better speed, it would be worth it. If all I got from Blacksmith was the cheaper price, it would be worth it. If all I got from Blacksmith was the four times faster cache downloads, it would still be worth it. But that's not all you're getting. You also get the best observability I've ever seen into what your CI is actually doing. What jobs are passing? What jobs are failing? Where are they coming from?

Where's time being spent on projects like T3 Code, this is so helpful. Trying to figure out why our CI is slow and fixing it has never been easier. It's fast by default, but you can get even faster when you can do things like SSH into the box and see what's up or use an actual native MacR runner for your Mac builds like we have to for our iOS and Mac apps. It's honestly gotten hard doing those time comparisons because I haven't chosen to use GitHub actions in months.

Save stress, time, and money on your actions at soy.link/blacksmith. I already know how this one's going to go. There's going to be endless bias accusations of, "Oh, Theo's just here to plug T3 code. He's [ __ ] on terminals so that he can plug his own thing." Oh, do you mean this open- source thing that I spent a shitload of money and time building because I so deeply believe the terminal is the wrong place? If you think T3 code is the reason I'm doing this, your reasoning is worse than Sonnet 4s is because this is obviously a thing I built because of my beliefs, not I'm lying about my beliefs because I built this thing.

That obviously doesn't make sense. I wish you guys would just use your brains a little more before leaving those shitty comments. I'm sure there's already a hundred of them plus. Scroll through the Obsidian ones and I'm sure you'll find plenty about how I'm just plugging my own [ __ ] But in order to skirt those accusations for now, I will switch to Chat GBT. So, here we are. An app I haven't used in a while. I just had to update it cuz I haven't been in it as much, but it still is, in my opinion, one of the best starting points for a developer who is moving into coding more with AI.

While the Codex app was the big thing that pushed me over on this belief that terminals were just not the right place to be coding anymore, it was not the start of my journey here. The place that I started thinking about this most is honestly kind of embarrassing and I even feel bad putting it in here. I I feel sick even typing it. Oh god, just can Can I delete it now? Do I do I have to leave this word on the screen?

I'd really like to just hide this. I'm just going to get rid of that. Okay. Really though, when I first tried out anti-gravity, I was mostly unimpressed. But there was one particular piece that impressed me. There was a hotkey that opened up the agent manager view that let you see what the agents were doing across different instances of anti-gravity. So if you have like three windows open for three different projects, you could have this one window on top that showed you what was going on with agents in each of those other projects.

It was a left sidebar with projects that had threads in each project that you could see from one external view. Huh, sounds familiar. Interesting. As much as I don't want to admit it, the anti-gravity agent manager was one of the first times I experienced that. And if anti-gravity wasn't such a [ __ ] show, I probably would have actually used it a bit. I even said as much in my coverage of anti-gravity when it dropped.

But what's much more exciting to me is the new agent manager workflow. I think this is a great idea and there's a lot of little pieces that I'm hyped about. So, anti-gravity was the first step in my descent as I realized that the terminal was not the right place to do this anymore. Funny enough, just a few weeks prior, I was at the cursor office complaining to them that all their new integrations with the cursor v2 agent view made it too hard for me to use my terminal because they made it way too annoying to like get the path that a work tree was in.

So I could cd into it in my terminal and then run things. Because a flow I often had is I would use cursor. I would open up something in the sidebar, tell it to go do something, get a result, want to go check it, and then have to find some way to CD into that path for the work tree in order to run it, install things, and do stuff. And the built-in terminal was no, not happening, especially at that point in time. So, I ended up bullying them enough to give me the one-click copy buttons that I needed in order to keep using my terminal for the actual doing things beyond sending prompts.

And I got to the point where I would use cursor to send a prompt and then my terminal to do everything from reviewing the diffs, reading git history, investigating changes, actually running the code and doing things. And I was split like 50/50 between I would be in the cursor app for prompting and that would be my terminal for everything else, especially git stuff. I did all of my git through the stock bog standard git commands.

Occasionally I would add like oneliner helpers to do things like commit and push with one line stuff like that. But for the most part I just used stock git cli. I need to get this anti-gravity thing off the screen. It's going to hurt me. But then the models got better and importantly the codeex app came out. And when I started using this, my beliefs that I had from anti-gravity saying that this is horrible, but it's also kind of the future got me really excited.

And the more I played with it, the more impressed I was. The ability to just break your work up into different projects and clearly see what's going on in them. to have multiple things going at the same time all in their own work trees. The ability to quickly open up a terminal and type things in here if you need for whatever stuff that you're doing. I got to the point where I was still looking at my git history through the terminal inside of my codeex instance and it was solid.

It was actually quite good. And in a world where I was using my terminal more and more because I had at this point moved to cloud code and was constantly running cloud code in my terminal, opening lots of terminals, using awesome tools like cmx, which I still highly recommend to manage all of my different things going on. I was still a big T-Max guy. I was still a very deep big T-m guy. I only started using tabs in my terminal at the start of this year as like a personal challenge when I set up a new computer to see if I could finally move off T-Mux.

The answer was no spoiler, but I I tried. So, I went through all of that and then the Codeex app dropped and I was like, wait, I can just screenshot something and paste it and see it. I can select text and copy it and paste it and the formatting is not broken. I can scroll and it doesn't lag a whole bunch. And most importantly, by far, I can hop between different things easily by just clicking. I know this sounds silly and dumb, but when you're working on three plus things at a time, it's just a couple too many hot keys in T-Max.

And keeping track of what is where is even more annoying. You have to remember which hotkey goes to which thing. And then you start building systems in your head for these three panels are for these three things. This fourth panel is for this thing. This fifth panel is the flex panel. I use it for whatever. But then you end up with a sixth task and your whole model breaks down. I tried my hardest to figure this out. And again, to their credit, Semox got the closest anything has gotten to figuring this out.

By giving you a sidebar similar to what you get out of something like the Codeex app, but also just giving you a terminal on the other side. It's great and it's still my default terminal. In fact, all the boxes I SSH into, I have in a little folder called SSH, so I can have all of these connected and quickly hop into them at any point. But man, having a real UI with text that was more pleasant to read with much easier to navigate and find things with a category system that made sense, having actual projects that are based on where things are in your computer, it just it made me code way more.

But one of the biggest problems I had with Codeex is that it was closed source. So when it did have regressions or problems or things that I wanted to fix, it just wasn't super realistic. I was able to bug the Codex team a bunch. And thank you to my friends over at OpenAI. You guys have been awesome about taking all of my feedback and addressing it meaningfully and swiftly. But there would still just be random regressions and changes and things like the merge into chat GPT.

Like this is no longer Codex. This is now chat GPT. They let you use the old icon thankfully, but like this is now the chat GPT app. Because the Codex app was so good, it made these tools accessible to people who weren't developers and they decided to give that capability to non-devs by rebranding it and calling it chat GPT. This is one of my biggest issues. If you wanted good open-source solutions, you basically had to use the terminal.

You had the codec, but you also had lots of other awesome things. Shout out to stuff like Open Code and Pi, phenomenal harnesses and terminal tools for coding with AI. But if you wanted a guey like codeex, your options were to use the one locked to the provider that you like. So you can use the cloud code app with your cloud sub or you can use the codeex app with your codec sub. You could hack the codeex app to use other things kind of.

You could kind of do the same for the cloud code app, but you shouldn't because the cloud code app is just not very good. But your only option if you wanted one place for everything was open code, but then you lose all the subsidization because open code isn't using the official harness. You also don't get the features that get added to things like cloud code or codec. If you wanted the best features for the best price with the best models, you had to use claude in the CLI or codeex in the desktop app because both companies kind of stop supporting the other side as much where a lot of new codeex features only happen in the app and a lot of new cloud code features only happen in the CLI.

And I got to a pretty nasty point where if I needed to use my cloud models, I was in the CLI. And if I wanted to use Codex models and GPT models, I was in the guey. And I ended up finding myself defaulting to OpenAI models even in things they were worse at because I just liked having a good app more. And it's all these little things that add up. Like seriously, if you haven't tried one of these guies, I I would love to self-plug, but just go try the codeex app.

It is genuinely really good. It has its edges. It has its annoyances. They added the new activity view, which is inspired by what we've been building in T3 Code. And as cool as it is, it's not quite all the way there. But god damn, if this isn't a massive improvement on what you're already doing, I don't know what to tell you. It's so much easier to keep track of what's going on, where work is happening, what you've been doing, and actually getting stuff done.

I can't tell you how many times I lost track of some terminal session. I had to open up cloud code and tell it to go find the cloud code I was doing work in or the work tree that I started from, or all these other things. I had to put a ton of time into just keeping track of what was going on. And as you can see from my chaotic setup here, I put a lot of time in. I had to reboot this machine earlier today, so I lost all the sessions I had, but I had like 30 terminals in T-Mux on this one Linux box for various cloud code and CEX things that I was doing.

I still have a whole bunch over on my other Linux box here for all these types of crazy tasks that I've been trying just like running goals indefinitely to see what happens. like the TypeScript Rust port that I was briefly working on before I ran out of tokens. Just keeping track of what is what is so annoying. And you can do like renames and things if you want to take the time to, but it's so annoying. And the amount of work I had to put in to try and make an environment that was reasonable just to switch between cloud code and codecs on real work and keep track of what I was doing.

It's just more mental overhead. I'm going to be real with you. It just is. And this is not historically how terminals worked. Historically, terminals have been the option that is less mental overhead and slightly more upfront cost. You had to spend a little time to set it up how you want. And then from that point on, the terminal is mostly just doing the same thing the way you want to. But in a world where the work I'm doing is changing so actively that the projects I'm working on might be brand new, they might be months old, they might be years old, and the amount of work I'm doing in parallel can change a ton as well.

I might be working on one thing. I might be working on 10 things. The rigidity of the systems I built for myself in the terminal was not flexible enough for the work that I was doing today. And I tried my godamn hardest to fix that. I put a ton of effort into making it so I can paste images over cmox over SSH through my T-Max layer here. So I can command V here and paste an image through cmox over SSH into the T-modex instance on this machine remotely.

That [ __ ] was so buggy for a while and now it's relatively stable. So I could finally paste images over SSH, but that's like a thing I had to put time and effort into to do something as basic as screenshot something and then send it to my agents. With a guey, I just do it. And I would say like halfish of my prompts have images because when I have an error or a thing that is annoying me, it's a lot easier to just screenshot, send it, and say fix and it will.

Whether that's a screenshot of your app or just a weird error you didn't expect to get when you ran a command. Sending screenshots to your model is very effective, but it's not very effective when you can't even see what the image is. But my biggest complaint is none of this. It's actually okay. It's kind of part of one of these things. It has very little to do with the terminal and very little to do with the agents.

It's three very specific, very annoying letters. SSH. SSH kind of [ __ ] sucks. It's really useful, really powerful, and I spend a ton of time in SSH, but as soon as my connection isn't perfect, I end up with all sorts of problems. I had a problem a few days ago where I was running Cloud Code over SSH on my local network, which is a Wi-Fi 7 network with gigabit to 10 gig speeds depending on if I'm wired or wireless. And I had my computer plugged in over 10 gig at my desk downstairs. and I was getting sticky keys while I typed trying to do work in cloud code.

It turned out it was bound to the Wi-Fi, not to my Ethernet. But the fact that I even had to care is absurd. And don't get me started on trying to do things on mobile. GH I I tried so hard to make Terus work for me. I paid the full price for it. I support the developer. I think it's awesome and it helped me debug something even this morning. Like I still use these things. They're just not the thing I default to anymore because they're fragile.

And it's incredible how powerful a standard SSH is. That's been around for over 30 years and it works so great to this day. And there are other fun things like I know SSH3 is apparently much better. I know mosh is really cool, but I do not have it in me to deal with all of those things to get marginal improvements, especially if I want these to work on my phone because trying to do real work on my phone via a terminal has me questioning my own sanity and also questioning the reliability of all the other people who were saying that using their terminal over their phone was a totally acceptable way to do quad code.

It just isn't. And thankfully, Codex accepted this fact and they introduced the ability to control the Codex app, now the ChatGpt app, over the Chat GBT app on your phone remotely. And it was so much better. It had its rough edges, but the ability to spin up some work on my computer and then finish the thread on my phone when I'm away, it just kept me off my computer a lot more. And I found myself like walking around doing more chores, leaving my apartment to go run errands, and being able to check in on the stuff that I was doing overtime.

It was really fun and cool and also obnoxiously buggy and most importantly was super restricted to the Codeex desktop app, which meant I could only use it on my Macs. And when you combine that with all of the annoying parts of how codeex is implemented around the security policy stuff on Mac OS where it spins up five plus processes per thread and per sub agent which can just pose your performance. I was getting annoyed and I had a couple things that I wanted and I will list all of the stuff that I wanted to make sure I could do.

The things that were annoying me obviously pasting images was a very important one for me. I also wanted all models with subsidization. So, I wanted to make sure whatever came out here worked with my cloud code subs and my codec subs, so I could get the 8 to 14 grand for 200 bucks instead of having to spend 8 to 14 grand for it. I wanted good remote control. I wanted best-in-class Linux support because, as I've said in many videos, check out my Linux video I did recently, Linux is so much better for agentic work because the file systems just suck a lot less.

And spinning up a work tree takes 2 seconds instead of 30 seconds to 2 minutes. just so much better. And even when I'm going like really heavy on my work on my Linux boxes, it is rare I can spike even like a third of my utilization. Like I have 32 cores on this machine. It's an a Ryzen AI max box. It's the framework desktop. I mentioned that before. I have tried to stress this machine and I basically cannot do it. And a lot of that is just the fact that Linux doesn't have all of the chaos trying to scan every single thing Codeex does.

So, I best-in-class Linux support, good mobile support, good performance, low battery impact, and probably the single biggest part, a good way to manage lots of things at once. This is all the stuff I wanted. The only one of these things terminals offer me are best-in-class Linux support. They don't help in any of these other categories in any way. And now you guys have waited long enough. The commenters can do their thing.

I need to talk about T3 code because Julius and I made T3 code because we shared a lot of these frustrations. We wanted to love Codeex and did for a lot of things. I know Julius still uses it heavily. I've slowly moved off the Codex desktop app. It has its benefits for sure and I'm hoping that we can clone as many of those as possible in the near future. Julius and I tried all the other options that existed at the time and were very unhappy with them.

The couple that were okay, we gave a lot of feedback to and they fixed some things, they failed others and we got annoyed. I got so annoyed about the performance part in particular that I tried building my own native guey for codecs to just try and make something faster. I ended up in hell with scrolling text because Swift UI and AppKit are not great at big scrollable views with text. And it turns out Google put a lot of good work into Chrome for that.

So I begrudgingly threw that project away. Begrudgingly spun it back up on Electron just out of curiosity. decided to architect it in a way that made other things like the remote control a little easier by putting everything on a websocket instead of binding it through the electron RPC. And the result was something surprisingly usable. The result was T3 code. It did not look at all like this at the time. It was sinfully ugly at the time, but it allowed you to use codecs with an open-source guey that was a lot less impactful on your systems performance.

I'll be honest though, for the first few months, I still found myself using the Codeex app more than T3 Code. I would rotate between them for sure, but I spent most of my time in Codeex probably. There would be moments where I went up and down, but I kept finding myself back in codeex, but Julius couldn't take that. Julius continued iterating on T3 code because Julius has something that I don't. Julius hates terminals.

This one was hard and I've argued with him a bunch about this. He exclusively used Git through the VS Code extension. He didn't even He probably knows most of the Git commands in the terminal, but he never uses them. The only terminal he would open was the one in VS Code or cursor to spin up his dev servers. Everything else he did through the editor. He's one of the best devs I've ever met in my life. He's without question the best dev I've ever known that hates terminals.

For me, T3 Code was a way to have an open-source solution that could potentially solve a handful of my problems. For Julius, T3 Code was his ability to escape the terminal. And he took the challenge, and the [ __ ] did it. He made T3 code so reliable, so stable, so well implemented and integrated. He got work trees working perfectly. He got cla working perfectly. He got open code and grock working as well. And even cursor, although the cursor integration needs some work.

Big update coming there soon. He did all of that because he wanted to take advantage of all of the models from all these providers at all of these subsidized rates, but he didn't want to have to use a shitty terminal app and memorize a whole new set of hotkeys for every different one. And T3 Code became his surface for managing all these different agents and all these different solutions with a onestop place where everything behaved the same regardless of which models you were using.

And reminder, this is using your cloud code and your codecs on your machine. And we're not doing some hacks like -p. We were using the official agent SDK blessed by Ampropic. They were going to change the policy. They have temporarily held it. They might indefinitely hold it. Still waiting for updates there. But this all works above board officially supported by both Cloud Code and Codeex to the point where the Codex team is publicly shouting us out as inspiration for features they're adding into Codeex.

We put a lot of work into all of this. Not just building it, but negotiating with the labs to make sure we do things the way they want so that we can continue offering your subscriptions to the platforms you like in our app without getting banned. And then the magic piece comes in. There's a little button here at the bottom. Other apps have this to be clear. Most of them have it as a use it on your machine or use it on the cloud button.

We have something different. When you set up T3 code on a machine, you can now control that machine through T3 code on the app on another computer on the website on app.t3.codes or in the mobile app. Yes, we have a fully supported mobile app for iOS and Android. And yes, it is also fully open source. So, if you want to build it yourself, you can. This here is magical. The thing that has fundamentally changed how I work is the ability to just pick a computer, send off a task, and close my laptop.

Before this, I found myself having to like prioritize my life around when I could run an agent. Because if I had to run an agent and then I had a meeting coming up and that was going to overheat my computer and make it unpleasant to use in a call and spin my fans up so loud that I couldn't use the mics on my laptop, then I would wait until I got to the end of the meeting and then spin it up, hoping I don't have another one coming up.

If I had an appointment or a trip or an event or something I had to go to, I wouldn't spin up my agents because as soon as I lose Wi-Fi, they die. One of the more embarrassing things I must admit here is that I'm notorious for coding in Ubers. I would spend a lot of time driving around to different events and things because I have to go to a lot of different [ __ ] And I would often pick an Uber just because I could sit in the Uber and get some code out.

And when I'm at my office, I am so behind on so many other things. I have to do all of that. So my Ubers became my like peaceful coding time. I have a lot of posts where I talk about the crazy performance improvements and changes I made from Ubers. But something bad happened. I started using claude code when Opus 45 dropped. I This poses me going back to Opus 45 and trying something. But notice how I ended it. I will have to stop momentarily as I am currently coding in an Uber.

My bad data connection was causing me problems. And as soon as I got there, I had to close my laptop and whatever agent I was running died and I had to spin it up later hoping that it would continue. And the more that this happened and the more I was doing tasks that would take hours instead of minutes, the more I just didn't want to code in Ubers anymore because as soon as I have to close my laptop, it's over. And now I have to think about my work in a way where I know I have enough time of laptop open for the task to be completed.

And I was like structuring my day around this. It was bad. Now I pick the box I want this to run on. I send the prompt and I stop thinking about it. I could send something to soul like do a deep audit of this codebase and then command Q T3 code so it's fully dead. There is no T3 code open on this computer anymore. I can then reopen it and the work is still going. Or I can command Q it again, get it hard killed on my machine and go to app.t3.codes and see the exact same work across all of my machines with the ability to continue sending prompts, see what's happening, paste images, scroll through history, and even spin up the terminal if that's what you want to do.

You got to admit that's pretty cool. And all of this is over the websocket connections that are spawned by T3 code when you run it on a machine. We now have a bunch of easier ways to connect as well. If you're already a tail scale person, I have a feeling you'll dig this new feature I recently added. npx t3. I'll do it nightly cuz that's what I've been using lately. T3 at nightly pair-tail scale. This command will give you a pairing code that you can scan on your phone or a URL you can use to connect on the web.

And now you have a full T3 code instance that you can control. You do have to serve it somehow. So T3 nightly serve and that will host it. Or you can use the even easier way which is npxt3 connect which will open up a URL to sign in with GitHub, Google or whatever else. And then it will create a background instance on this machine as long as on Linux Mac support coming soon that allows you to forever connect to this machine and run the agents that are already configured on that same machine.

So, if you have an old laptop somewhere at your house or a PC that hasn't been turned on in a while or already have a Linux machine that you use, set up a Linux on it if it's not already set up. Install Cloud Code Codeex, both or neither. Run this one command, npxt3 connect. And now you can control that machine and all of the agents on it from a much much better interface without having to deal with all of the consequences of things like SSH, of things like terminals, of all of the different annoying pieces that you have to deal with.

But I really, really hope you guys understand that I'm not doing this in some secret plan to make money. I'm doing this one out of frustration. I'm doing this one out of the love of the game. If I wanted to make a bunch of money, I could have just sold this to OpenAI or Ananthropic at this point, but the goal is specifically almost a counterbource software that we use to build software with. When I got into coding, it was mostly a closed source ecosystem.

All the editors people use were closed source. Things like Visual Studio, things like Sublime Text. They had some concept of extensions, but barely. And then GitHub released Atom, which was built on top of Electron. And then Microsoft decided to fight back with VS Code, which was an open-source electron editor that became the industry standard used by over 80% of developers. And suddenly we all were using tools that we could customize.

And now AI is huge. And everyone moved to cursor, which is a closed source fork, or to claude code, which is a closed source terminal, or to the Codex desktop app, which is a closed source desktop app. and our ability to customize our tools started to fade fast, which is particularly ironic because we're also now in an era where it's ever been easier to customize tools as long as you have source code access. And it just deeply frustrated me that the tools I used would randomly get worse and I couldn't fix them and had things I wanted and didn't have the ability for me to add them.

And I wanted to make sure that the next generation didn't have to grow up with a bunch of closed source tools they couldn't customize and agents that are more than capable of customizing. So, I ended up spending a lot of time and money and resource and betting hard on doing this. Not because I think I can make money on it. Sure, there's some potential to do that in the future, but the goal here is to make the best possible open-source solution for coding with AI today.

And I do not believe that that solution is a terminal. I genuinely believe that solution is a graphical interface that prioritizes the people who want to get real work done with systems that help you get that work landed. I have so much more I want to say about T3 code in particular, but that's going to have to wait for another video. For now, I just hope you guys get this takeaway. I don't care which guey you use. Ideally, use a decent one.

So, like if you use the cloud code desktop app, you're not getting the like proper guey experience, but use something like T3 code or codeex or you can deal with the performance issues in the new cursor glass version or you build your own thing or use gene or conductor any of these other guies for managing your agents. I think you'll feel the difference because it really really is significantly better than dealing with a terminal, especially once you have more than one or two things going on.

The ability to just hop around or hold down a hotkey and see which thread maps to which button and I can just commandclick between the different things I'm doing. Being able to rightclick and archive or just click settle, all these little things, especially once you start using images and stuff too, it's just it's so much better. And once I was liberated from the reality that doing this on another computer was too hard and no longer sshing constantly just to make sure the agent was durable.

Once all of that was implemented, it changed how I code. And I went from landing like three or four PRs a week to on my like heavy coding days as many as 20 PRs a day landing. And moving from the terminal to a real graphical interface is a significant part of why I was able to make that shift. Just to make sure this is clear, I love terminals and there's a lot of me that misses using them as much as I used to. But at the same time, they're just not the right shape for the things that we do today.

And it's time to stop pretending they are. The same way that iPhones used to make all the apps look like the real world objects in the skumorphic era because it was important that people saw a calendar and it looked like a calendar because otherwise they wouldn't use it. We're in that era still now with AI dev tools. We're doing things the way we used to because otherwise it's too much at once. I need you to look forward a bit though.

I want you to challenge that belief because the devs who were the wizards at terminals are the same devs who should be wizards with agents. And as hard as it is to admit, cuz I know it was really, really hard for me, the terminal is not the ideal interface for the way that we build today. And if I'm being fully honest with you, the only way the terminal isn't holding you back would be if you're not building big enough.

Push yourself more and try out some gooies. I have a feeling you'll be surprised. And until next time, peace nuts.