This New Codex Feature Changes AI Coding Workflows

Added: 2026-05-13

[00:00:00]CHGPT have just released goals inside of codeex. Basically, it allows you now to pursue a single goal in code for like 15 hours, which is nuts. And all you have to do is literally give it slashgo and it will go away and just keep working for ages and ages and ages until it gets the result. So, in today's video, what we're going to show you is how you can actually get get it set up and running, what you can actually do with it, how it really works. Simply put, of course, um some real ways you can actually get started today. Um and then a kind of simple note on does this keep going on forever and never ever. Well, let's see.

[00:00:38]So, first things first, go. It works for 15 hours. It's like codeex driving forward to a finish line. He's a little image that was made by Chad Gupt image generator, which I'm super impressed by.

[00:00:50]Basically Sam Olman driving codeex to the goal. That's kind of the gist. Now, you cannot use it inside of the app. You can only use it in the CLI, right? But how do you actually set this up? And obviously, of course, it should be coming here soon. How do you set this up? Well, inside of the config tommo, you want to set goals to true inside of features. Now, you can either go here or an easier way to do it is simply open up your codeex app if you have it. Go to codeex settings and then in here you go to configuration on the left and then open config tommo here. And then it'll open up the config tunnel which is this and you'll look and go ah what's going on here? Now if you don't have features you just simply add it and then set goals is true. I added this entire block. So literally just verbatim copy and paste that in and save it. It's basically just your configurations.

[00:01:44]Quick pause. If you've been watching me build AI tools and my SAS, you'll know that relying on prompting alone can be incredibly slow. Instead, you need to understand the code it's writing, or at least the concepts that it's using. And that's where today's sponsor comes in, Scribba. Now, Scribba isn't just another site where you have videos and you open up VS Code or some cloud environment.

[00:02:06]Instead, you can code directly inside the video. Whilst your instructor is coding along right on the video, you can at any moment pause the video, start typing inside the video, and then it will branch off. You can start running this, adding environment variables like your own API keys and stuff. Even download a zip of the code, and then jump back into live preview. So, if you've ever watched a tutorial before and thought, "What the heck is this guy doing?" you can just pause it now and just start using his actual code right there and then. Plus, what they teach is super relevant right now. They have a claude course and they have an entire AI engineering course covering vector databases, MCP servers as well as a bunch of front end, backend and full stack courses as well. That's why the website is so fast, so responsive and so good. Now, if you want to try out this interactive video/code editor, you can sign up completely for free and start on some of the courses. So, check out the link in the description below. It will get you 20% off Scribba Pro. And if you like it, grab the annual plan and you'll get 50% off on top of that. So link below. Let's get back to the video. And then you'll have to start fresh a new terminal here um with codeex and then it'll pick it up. And the first time you open it up, it will have this under development features enabled goals. So you'll know that it's actually there.

[00:03:29]And you can do slash goals. And when you run one, you'll see it's pursuing the goal and the length of time. Now, this new one that I'm running is only 8 minutes so far, but they can run for hours. Like, I've seen guys on Twitter and YouTube that's like 10 15 hours. So, what can we actually do with it? Well, there's kind of four ways in which we can leverage this. Number one is goal/go. So, this starts the goal. You just give it some text. So, for example, here for me, we have given it some goal.

[00:03:59]It's hard to see here because it's gray.

[00:04:01]It's easier to see here. kind of improve the homepage for blah blah blah, right?

[00:04:06]Which actually a really bad goal, but it told me to do it. So, you know, we'll blame uh chat GPT for that one. But in any case, you have your goal. You can also pause your goal. So, slash goal pause and you can slashgo resume. So, to pause it, start it. Uh also, you can uh slashgo clear. Now, this will stop the goal entirely. So, if we come over here, for example, this is just a little demo for me to show. just ask it to give you some goal suggestions. But here, if we make this easier to read something like this, we can do slashgo pause. Oops, cancel. That was not pause. That was pod. But slash go set or view the goal.

[00:04:44]And you can do pause and then go pause as you can see here. Now, what we can do is we can start it again with slashgo resume. And then it'll go and start the goal again. And we can also ask, you know, ask a question kind of on the side and send this in. and then it will get added into the context here of like it will actually answer it which is good so it doesn't lock down the session not fully and that's why I say it's non-blocking here you can integrate other questions whilst it's working away so there's kind of two ways in which to think of how you work for how you can actually use this so I notice there's two main ways you use AI for everyone basically which is like one shot well definfined this is what we want to do and the other thing is like you're just kind of experimenting. Now, I'd seen in a guy's channel, Ry, shout out to Ray, what's his name? I'll bring him over just to shout him out a little bit.

[00:05:36]Oops. This guy here, Ray Amjad, and he talks about, you know, how goals can be used really effectively for this if you just chat up front with the AI to get you a well- definfined goal. So, that's kind of like you can do this to then get some output to then put it into this.

[00:05:54]But if you're like me, you're not really, you know, you're more trying to just let AI be the employees and you're kind of like the investor, [laughter] I guess. Um, is these are the two different ways to use it and goals works better here. So, you're like, I want this thing to happen. So, like I want my site to rank on Google, enhance the SEO or something, right? Whereas try this here would be like I want customers. And then it like you should try this and you're like, I don't know, maybe maybe blah blah blah. So with this one you have this whole conversation you get the I want and you br bring it in here or you start with the I want ideally something well defined and the best way to well well defined stuff is to use metrics or specifics so ideally you know like reduce this by x% change this to make it whatever you know I'm being very specific I know but you get the gist and now you'll be thinking well isn't this just like prompt isn't this just what it should do anyway well no the kind of old way is that it would actually stall you know like some part of the way there it would just kind of give up or it just wouldn't work properly. And to fix this, Claude used these Ralph Wiggum loops.

[00:07:02]Well, now the goal should work here where it actually gets to the end. It shouldn't break. It should just keep going and going and going. Hence why it can last for hours and hours. And also interestingly when tokens run out you end up there's a there's a guy income stream software and he was talking about how his was Why is that not finding anybody? There we go. He was talking about how for him it ran for like 15 hours which is crazy. Um, but I then commented and I think it might actually have just been that you run out of tokens and then it would restart again, which to be fair on its own is actually really good because then it means you can just run cheap subscriptions and you never have to worry about trying to go back and say continue, which I would used to have to do before I upgraded. It would just run. So before we look at these good examples, you may have a kind of quick question of like, hm, but look, if it's going to go all the way up, use all the context. Because if we come over here and see it's going to use all the context, won't it then have to slash compact itself? And isn't compaction bad? Well, I mean, if you don't know what that is, just first of all, basically, your conversation can only be so long and when it gets to this like point here, either you have to go and start a new chat or you can make it compact it and just take the essential bits and then it will bring it into a brand new kind of chat kind of. this here inside of clawude this compaction everyone always said it was awful but inside of codeex I've been using it in codeex more because you can't inside the codeex you can't just slashcle like you used to be able to do so see in claude you would run a slashcle and then that would clear your entire context and then I just paste in like the last section from before I cleared it but in codeex you don't have this slash clear you have to go and make then instead a brand new chat which I find it's a bit hard to keep track of.

[00:08:57]Like if I come to the application here, you can see that I have all of these chats and I'm like which is which? I don't know what's going on ever, which is a bit hard. So then instead of that, I just let it autoco compact and it seems to work really well and it should work well like if it does because it should just take the essence of this or the main points and then it should just take those and carry them over. So good compaction should actually get, you know, better across time of of just keeping the concise information. Anyway, side note, so it should be fine is basically the long long and short of it because it may have to do this type of thing, you know, like a million times because it's going on for hours and hours and hours. Now let's practical examples you before we go away today. So I was talking to chat GPT to find out what it says is good, what's bad outcomes wise, sorry, prompts wise, outcomes are good examples. Things like get this app demo ready for a YouTube recording. It says fix all build lint and runtime errors. So fix all here being the key one. Uh implement the save projects feature end to end. So make sure that it's fully implemented.

[00:10:03]Refactor this repo. You get the gist.

[00:10:05]Blah blah blah. So essentially it's like it says here in those cases normal codeex might do step one and stop. Goal will keep it going until it finishes.

[00:10:15]Now the other side what you should not use it for are tasks. So here it says like rename the button, explain the error, add one CSS, write reax. I don't know about you but for vibe cod and I don't do any of these. Sometimes I'll do this one but not the others. Um for these saying just prompt normally. So essentially it's meaning if it's just a little thing then you know here you don't have to use the goal. So also don't use go when the task needs a lot of uh taste decisions from you. For example redesign my whole SAS until it looks premium. And that's kind of what I was trying to get at with this here which is if you know what you're trying to actually do that I want this do that.

[00:10:52]But if you don't if you're like open-ended then try this. So there you go. Go and use goals. Let me know how you got on in the comments below. I'll be using it going forward because I want to just let it go and do its thing. Uh, but a weird side note that I noticed before I go is basically that here I wonder because chat GPT is definitely slower than Claude. I wonder if how long things take is now actually becoming an indicator or a proxy for people to think that's how good it is. It can go for that long. But I wonder if they're actually slowing it down on the other side to reinforce that metric just for the time being because basically then it means they don't have to process stuff as fast. So then they don't have to have as good of processors and they can save processing power etc. But you as a user, you'll be like, "Oh, it ran for 25 minutes or three hours. It's amazing."

[00:11:46]You know, I don't know. Let me know in the comments what you think below and I'll see you in the next video.

[00:11:52]Bye-bye.