Why I stopped using Codex and started using T3 Code instead.

0:00 Hi, if you have been a follower of my channel for a while, you know that I'm a big Codex fan. Codex being the agentic coding agent from OpenAI and their Codex app, where I write most of my code nowadays. So why in the world are we looking at a thumbnail here saying that I switched away from Codex into something called T3 Code. It's a annoying name to say, so that's definitely not one of the reasons why I've switched.

0:34 But why did I switch and what is T3 Code? Well T3 Code is an open source coding agent or not really a harness as we normally call it because it's just building on top of all the others. It doesn't have any tools on its own. It simply just calls all the others APIs behind the scenes. So in order to use T3 Code, you need to still install a Claude Code or a Codex, but you don't use their interface then. You just use their intelligence.

1:10 It's built by some web developers who really liked Codex, but was very annoyed that it wasn't open source. It had OpenAI open source to the Codex app. This product would probably not exist. But behind the scenes, the GUI is just like any of the other coding apps that are now the visual. There's the CLIs, of course. There's the integrations into Visual Studio Code versions.

1:45 And then there's the apps. So there's Claude Code Desktop, there's Codex app, there's OpenCode, there's GitHub Copilot and so on. And they all have the same structure built up. They have on the left hand side sidebar with projects. They have the chat and then they often have right sidebar with an integrated browser, a diff editor and so on and so forth. And this one has the same just on the surface, but Let me go into six different reasons why I use this instead of just having the Codex, because this is essentially just an extra step in order to use Codex.

2:32 In that I need to install two things instead of one. But I have six reasons why I personally switched. So let's go through them. The first one is the sidebar. So I'm gonna get rid of this over here. We're not looking at that. We're only looking at the sidebar because it is amazing compared to any of the others and I'll explain why. But I've just put in Codex version of it over here just for reference. And as you can see each entry into the sidebar, you can see one here and one here is much taller. And you might say then you don't really have room for the same amount, but the things that are about this is that you shouldn't really see this as a project sidebar where you have different projects with different chats and so on.

3:28 You should see this more as an inbox. And once you get that mindset into it and if you like zero inbox, because that's essentially the same concept as well, where you actually get rid of all what you have in your inbox, then you get the mental picture. Because whatever is up here is what something I need to react to. It's something that is done coding. The AI is done with the task and I need to do something about it.

4:04 Then while it's working it puts it down into this working area. This is an option if you don't like it. Some people like it putting down here, other people don't. But if you have like four or five different tasks running at the same time, it's nice to go out of the way and wait, in this case 18 minutes, before we get in there. And this is also one of the things you can on the fly see very much details about what's going on in each chat you have with your AI.

4:39 But getting them out of the way, and this is normally collapsed, I just opened it so we could see it. So it's just out of the way, out of mind, because there's nothing to do there yet until it's actually done. Another thing is you can snooze things just like you can do in Gmail. You simply right click and say I don't get to work on this until Monday next week, for example.

5:09 And then the most intelligent part of this is the settle system. So in all the other ones you do an archive of a thread and while that works it puts it into getting it on archive. You need to go into settings and find it. This has a middle step called settle, which I essentially end everything on. Where it's just a single click it settles, it goes down here, and if I quickly need to get it back I can go in here and reopen again.

5:43 And the cool part is that if you have something that runs on pull requests and that pull requests are merged then it will auto settle. Again, this is a setting if you don't like a thing like that. But it can auto settle this saying we do the work it's still lingering on here because it's not done. But once we we merge the pull request in GitHub it will auto settle it and it will automatically end up down here without you needing to do anything.

6:22 So even if you do get in the task in another area, meaning GitHub, it will automatically also get rid of it in this list. And that can sound confusing and strange and it takes a little, it takes a date in order to get used to this, but once you do this really feels like a nightmare that you are in charge of needing to archive and finding the right one and they're already always in the wrong order and stuff like that.

6:57 So this is the main parts. Then it's nice that you can on the fly see the branch that you're working on instead of needing to click it and figure it out. It's nice that you can see that it's working and how long it has been working. It's nice to see that this is done working now. It's nice to see what provider it is. We haven't talked about that, but God, we can use multiple providers in my case. It's Codex behind the scenes and we can see that with a small little icon here.

7:28 Also, if there is a pull request, it will show the pull request number here as well. Either in green if it's open or in purple if it's merged already. So this is by far the most important part of why I switched. And again, once you have tried this for me a couple of weeks now, I can never go back. I don't think I can go back. Perhaps Codex will at some point make something similar.

8:01 And then there could be some reasons, but right now without this I can't. I feel much less productive. Second part is the per project settings. So you have projects like the others, but there's all these small niceties in for example, you can set up icons. onto a project. You can go in and set that a model is specifically linked to a specific project. So I can say my youtube should run GPT-6.1-Sol Low, but my finances should run in high and so on. So each one of these can be configured to your own settings.

8:48 And in the same manner you can also do that for the workspace part, so you can set it to this project or I always use worktrees, this project I don't use worktrees. worktrees are so important in some of my projects nowadays. While not as it's not important, then I can just set it to just use main or not even git if it's not under git control. So the settings shouldn't be underestimated how important it is to start a new thread and it's just the default like you like it every single time you start one. The third one is the mobile experience.

9:36 There's apps for tablets and mobiles and while ChatGPT have the same it's always linked to one account. And in my case, I both have a work account and I have a private account. So I need to choose on the mobile if I wanted to work with my mobile, my work Codex or my private Codex. In T3 Code you can do it across multiple accounts and across multiple machines which is amazing. You can even get the desktop versions to talk with each other as well if you like to. In my case, I don't because I don't want to mix home and work here.

10:21 But it doesn't care if it's multiple. It doesn't care if it's from multiple providers from multiple machines Linux, Windows, Mac. It just works and everything is here. You can do almost exactly the same directly in the mobile experience. It's just so much better than what ChatGPT have put out. Then there's stats and limits. It's built into the system that you can actually see how many tokens you use either in real tokens or in API costs despite not still using a sub, of course.

11:04 You can see your limits for your five hour limits, your weekly limits and if you have any resets you can see them from here and you can see them in various intervals. So it's nice to see how which models I use the most and so on and so forth. Fifth thing is the GitHub integration. So as you can see over here, I've pulled out a previous settled one and it shows the pull request number. It is purple, so I know it's merged.

11:42 But if I press that number I get an inline experience of the pull request which I to the point like so much to the point where I actually don't go to GitHub anymore. So I can see everything I need from here. I can press here if I want to actually go to GitHub, but I can also do the merge directly from here. And again when I merge this automatically goes down into settle.

12:13 The sixth thing I like is the multiple providers. So right now I'm using Codex primarily and don't really have any subs for any of the others. But a while ago I tried Claude again and Claude's code desktop is a mess and I don't want to be in a CLI. So this was the only real tool that was ever even considered for me using Claude. And again behind the scenes it's just using the CLI just like Codex is. So just get a better experience with the extra experiences.

12:53 But you can do that for cursor, you can do it for Grok. I did it for OpenCode a while while I tested some of the open source models and there's integrations for Antigravity as well. So this is also nice and inside these you can actually set which models should be highlighted to you, what should be the default and such. And it also automatically checks the CLIs and updates them without you needing to go and do those updates yourself. So it's always up to date. There's a few small things I miss from Codex and things that could be better. Of course there's always things and I could go in and do some of them myself if I made a fork of the code but I hope some of these will come along.

13:46 The first one that I miss the most is the side chats in Codex where you can write slash side and you get a little chat about the chat where you can ask questions about what is going on, what is happening here, do all the metadata about it. It's not there yet, I know it's on their to-do list but that will be really nice if they put that in. Another one is the schedules. There's no scheduling feature right now. I don't really use the schedules too much in Codex so it's not a big deal breaker and I just have Codex on the side during the few schedules I do have. But it could be nice if it was managed directly inside T3 Code or had an even better schedule system. But because I use Hermes Agent I don't really need to use the schedules in that regard other than for a few things code related. There's also no MCP apps visualizations. So if you use MCP servers with MCP apps with visuals, I will make a video about that fairly soon.

15:01 Those unfortunately don't isn't supported yet. They will not simply not show. That's just a matter of priorities and MCP apps are nice but they're not end of the world that I don't have them here. I also can't do the management of plugins. I use a lot of plugins in Codex and of course because it's just a wrapper on top of Codex it can use the plugins but it doesn't have any management of these plugins. So I still need to go to Codex if I want to add a different MCP server or set up this extra plugin and so on. But it can totally fine call them just by using the SDK. One of the annoying things is the sub-agent visualizations. It can certainly do sub-agents. You can spin up sub-agents and so on but you can't really go in and see other beyond a name of a sub-agent what it's actually doing.

16:11 I've heard that's something about Codex encrypting their sub-agents so they can only decrypt them by themselves. I think that's the reason why we don't see them. So we can only see a list of them. We can't see what they're actually doing. A little annoying but sub-agent is normally just a thing that works in the background anyway so I live with it. Finally a better diff viewer.

16:45 None of these coding agents have a good diff viewer in my opinion. This one is slightly worse than Codex versions. It's a little more dark and a little more subtle in the colors I feel like. Diff viewers is a whole religion among developers of course. I really like the visual studio version but right now there's no way to customize the diff viewer with the colors and everything. That's a little annoying so I don't use it but had it had a decent diff viewer I would really really like it. But that is the six small things. Again no deal breakers at all but as you saw there were six major things especially the sidebar. Give yourself the chance to try that different way of working. It won't click immediately after one hour. You need to give it a couple of days at least. But this is why I switched. This is why whenever I do the coding agent videos you see something from T3 Code and not from Codex anymore.

18:02 So with that thank you for attending this session and see you on the next one.