I was wrong about Vibe Coding & need to retire

0:00 Multiple CEOs said that AI is going to replace most, if not all, coders by 2026,

0:05 which as a coder of 20 years, I think is an indisputable brain rot take.

0:10 So, I made a video to prove that only

0:11 someone with AI psychosis can believe this by asking

0:14 ChatGPT to make an app that would be

0:16 a challenge for even the most seasoned developer,

0:18 just to get a sense of how good it actually is.

0:20 And of course, it failed miserably.

0:23 All right, let's run it.

0:24 Yo, I can't I can't do this anymore.

0:27 I can't do this anymore.

0:28 But, after I did that test, you all said that I did it wrong.

0:31 And some of you even got really offended.

0:34 So, I'm going to do it the way that you guys say that I should do it because

0:37 either I'm wrong or some of you also need

0:39 to check yourself into a psychosis clinic with the CEOs.

0:43 Okay, so I got multiple comments telling me the right way to vibe

0:45 code [music] is to use cursor with the latest Claude model, which real fast,

0:49 vibe coding for those of you new here is basically a tech bro's wet

0:52 dream of describing your app idea to an AI and it brings it to life.

0:56 Doesn't matter.

0:57 On the toilet, blackout drunk, whatever.

0:59 What takes teams of hundreds of talented people to do,

1:02 vibe coding will get the job done apparently on its own.

1:06 Right.

1:08 So, here I have the Cursor free trial with the Claude model ready to rock.

1:11 And in my last vibe coding attempt,

1:12 I tried to vibe code a fascinating yet challenging evolution simulator.

1:16 But, the result was very embarrassing.

1:18 I feel bad for all you vibe coders out there

1:20 because now I'm going to embarrass y'all's alleged vibe coding king.

1:23 I'm sorry in advance.

1:24 But, before we get into that, we need to lay down some ground rules.

1:27 Rule number one, I can't touch not even a single line of code.

1:31 I can't taste it, smell it, hear it, nada.

1:34 I practically need not even know that it exists.

1:37 If AI is as good as y'all want me to believe, it shouldn't need my help.

1:40 And [music] that's it for the ground rules.

1:41 Let's get started.

1:42 All right, let's try and figure this out.

1:43 And the worst part about new software is learning it.

1:45 I'm looking to build an evolution sim where there's

1:48 going to be a sub sim and a main sim.

1:50 All right, let's run down what we're trying to build.

1:52 My evil sim is broken down into two parts.

1:54 There's a main sim [music] and the sub sim.

1:56 The sub sim simulates an organism's microbiology

1:58 and decides things like when they move,

2:00 eat, defecate, reproduce, [music] you get the idea.

2:03 While the main sim is where all these actions take place,

2:06 with every action being the result

2:08 of something simulated in the sub [music] sim.

2:10 For example, this organism has a bunch

2:11 of energy converters that converts radiation into potential energy,

2:15 making it nearly immune to radiation [music] mutations.

2:18 And hopefully by this point,

2:19 you can see just how tight this sim would be in action.

2:21 Let's give this a shot.

2:22 I think I can drop off image in here.

2:24 I start off by showing my concept design to the AI and I was

2:27 quite surprised to know how quickly it picked up on what the goal was.

2:30 So, I know the AI will be tapping

2:32 y'all any moment here when I add more complexity.

2:34 I'm just so skeptical.

2:35 There's no way that it can do

2:36 all this and still maintain understandability, right?

2:39 So far, so good.

2:40 This is looking good, but there's some changes that I need to make.

2:43 We need to change the behavior of the elements slightly.

2:46 Now, this actually impressed me.

2:48 I yapped here for a good minute or two trying to provide

2:50 as much detail as I could and wow, look at that.

2:53 They summarized everything I just said in a couple sentences.

2:55 I'm excited.

2:56 I'm excited to see what we got.

2:57 Oh my god.

2:58 No, no, no, no.

2:59 And yep, sure enough, we have our first expected regression where the AI added

3:03 my feature request at the cost of removing other features.

3:06 Not good.

3:07 This is not good.

3:07 This is the beginning of the end, I feel.

3:09 Barely just got started getting to know him.

3:10 But, it was not the beginning of the end.

3:12 It was actually just a slight misunderstanding because I continued to explain

3:16 to the AI my design idea and what it got wrong.

3:19 And surprisingly, it just kept correcting the mistakes like a giga.

3:22 But then, it was time to increase the difficulty of my coding request.

3:26 I wanted to give each receptor a binding element

3:29 and check for collisions with that specific binding element.

3:31 This is a pretty complicated step.

3:33 I expect it to kind of fall apart

3:35 and not understand exactly what I'm asking for here.

3:37 But, if that's not the case,

3:38 if it actually does understand what I'm asking for and is able to deliver,

3:42 I'm going to be incredibly impressed and you

3:45 will get some concessions from me here.

3:47 Uh maybe it's impressive, but I don't know what their binding elements are.

3:52 So, and after being able to work with the AI

3:54 a bit and verify the sim is working as intended enough times,

3:57 my tune on vibe coding was starting to shift a little bit.

4:00 Which by the way, marks the perfect opportunity to let y'all

4:03 in on a bit of a Jabril's YouTube behind the scenes.

4:05 As a YouTube creator that makes videos around coding tech,

4:08 I get a lot of offers to promote vibe coding products.

4:11 But, I'll be honest, up until this experience,

4:13 I thought it was all largely just snake oil.

4:15 But, now that it looks like I'm a newly vibe code converter,

4:18 I want to thank today's sponsor, Vibe Code.

4:20 Imagine you can simply describe your app idea and it would be made.

4:23 What app would you make?

4:25 Well, Vibe Code is a mobile app that lets you build

4:27 App Store ready apps without writing a single line of code.

4:31 Imagine like a Canva or a CapCut, but for creating apps with the power of AI.

4:35 And it works like this.

4:36 Let's say I want to create a YouTube assistant app.

4:39 I work with the AI agent to generate assets for my app.

4:41 That's images, sound effects, music, etc.

4:44 Then I tell the agent where and how to use these assets.

4:47 And if you're looking to make something more complex,

4:49 you can enable Vibe Code Cloud to add user authentication,

4:53 databases, and storage.

4:55 And get this, you can ship to the App Store with one tap.

4:58 And soon, Vibe Code will be adding support for payment processing.

5:02 So, the only limit on what you can make will be your ideas.

5:05 And look, the truth is we live in a different era now.

5:07 And if you want to make your first app

5:09 without having to write a single line of code, you can try Vibe Code.

5:12 I got to say, yo, I'm having a lot of fun right now.

5:15 I'm pleasantly surprised.

5:16 In fact, the AI was doing so good that the only issue I was really

5:19 facing was not being able to initially

5:21 verify features I requested were working as intended.

5:23 But, after a short sidebar adding in the features I needed to verify it,

5:27 it was nearly always working as intended.

5:29 And not going to lie, it started to leave me speechless.

5:32 Wow, wow, wow.

5:33 Wow.

5:34 Yo, I'm so impressed.

5:36 Wow.

5:37 This is so fun.

5:40 Honestly, I'm going to be real with you.

5:41 I did not think that I was going to get this far.

5:43 And this is about as far as I've designed the sim cuz why waste

5:46 the time if I knew AI wasn't going to be able to do it.

5:48 But, yo, my bluff has been called.

5:50 But, now I need to sidebar with the AI to kind

5:52 of get a better idea on how to get to the next step.

5:55 Sidebar time.

5:56 Then, we start to work on some of the bigger picture design.

5:59 Like, how do all the receptors interact with the sim?

6:01 Do the elements really serve any purpose?

6:04 And the AI was not only there to code it up often on its first try,

6:07 but also to provide ideas.

6:09 I can't believe this, man.

6:10 So many Ws.

6:11 I feel incredibly confident that we're going to be

6:13 able to see this through to the end.

6:14 [music] So weird to say.

6:15 Are we really at this point?

6:17 And it was at this point I was fully sold on vibe coding,

6:20 paying lots of attention to what it can do and how it works.

6:23 Dog, it even auto updates the read me.

6:28 Yo, [laughter] we are cooked, yo.

6:30 This is the hardest part of coding, is updating your damn read me.

6:33 If AI can do this, stop liver.

6:34 I'm sorry.

6:35 I owe a lot of people in the comments some apologies.

6:38 Then, I reluctantly had to end our vibe and go to bed for dumb human reason.

6:43 All right, I am both awake and excited to continue vibe coding.

6:48 I cannot believe I'm saying that sentence.

6:49 I got morning voice, can you hear it?

6:51 I'm so excited I am.

6:51 I just jumped right out of bed,

6:53 washed my face, and now I'm back at the computer.

6:54 But, before you jump back into it, let's take a second to pause here and answer,

6:57 out of all the things to benchmark an AI on, why an evolution simulator?

7:01 Well, one, me love a good evo sim.

7:04 And two, I believe evolution sims are the [music] final frontier of coding.

7:08 They demand all the coding principles to be firing at once.

7:11 Maths, design, scalability, patience.

7:14 If there is any AI out there that can successfully impress given this task,

7:18 I mean, what better way to know if we're cooked or not, right?

7:20 Speaking of which, back to morning voice Jabril.

7:22 Very nice, yo.

7:23 I'm really impressed.

7:25 Not only is it working practically flawlessly, man,

7:28 like I haven't had nowhere near the experience I had in the last video.

7:31 Every request I made removes so many other requests.

7:34 It's a pretty looking app, too.

7:35 I got to say, man, I did not know that we were here already.

7:38 But, I always felt like things were a little too good to be true and I

7:42 think I was right because we finally started

7:44 to run into our first patch of errors.

7:45 Wow.

7:46 First error.

7:47 This is not good.

7:48 Hopefully, this is not a sign of things to come.

7:50 This error started to show up a bit more often at this point.

7:53 It seems like the AI was forgetting how to correctly format the Python script.

7:56 But, I didn't really care because the AI's

7:58 track record at this point was near spotless.

8:01 I would describe something to [music] it

8:02 and it would always figure it out eventually,

8:04 often time on the first try.

8:06 So, if I ran into any errors,

8:07 it didn't make me want to play in traffic like the first

8:09 time I did this experiment cuz I trusted that it would deliver.

8:12 And this was honestly the flow for the rest of the project.

8:15 the feature I wanted to the AI and in just a couple of takes,

8:18 it would implement it flawlessly.

8:19 And then sometimes, we'd have to sidebar to add visual features for me

8:22 to be able to verify if the sim is working as intended.

8:25 But, there was a lot of impressive observations along the way.

8:28 Like when I just flat out relied on the AI

8:30 to add the main sim in the way that it envisioned it.

8:32 This is like the ultimate test for me.

8:34 If it does not break here and actually delivers on something, flabbergasted.

8:38 I'm done.

8:38 We're cooked.

8:39 I completely concede that I I I was wrong and I owe

8:42 a lot of apologies cuz this is some serious great stuff right here.

8:45 I'm going to get some water in the meantime.

8:47 It finished.

8:48 I am so nervous.

8:49 And yes, I'm about to eat yogurt before I code.

8:51 What are you going to do about it?

8:52 All right, moment of truth, ladies and gentlemen.

8:53 Let's see what happens.

8:54 Run it.

8:55 This is really interesting.

8:56 How can I be so sure that they're actually calculating the sub sims?

8:59 I think it's a good start, pretty,

9:01 but I have no way to confirm that it's doing anything that it's intended to do.

9:05 So, we side barred to add the connective features to make sure it worked.

9:08 And I just repeated this process having a good

9:10 old time see my vision come to life.

9:12 And I'm happy to tell you that it did work.

9:14 Now, its improv vision was a little off

9:16 from what I initially imagined for the main sim.

9:18 But, as the AI explained to me,

9:20 that's mostly because it was more focused on performance.

9:22 But, yo, the AI absolutely killed it on this project.

9:25 There are so many things that the AI did that had my jaw on the floor.

9:28 Like when it made a whole interactive app for the sim

9:30 and I didn't even ask for that level of polish.

9:33 I can click an organism, too?

9:34 You're joking.

9:35 This is such a glaze fest.

9:37 I can't believe this.

9:38 I owe some apologies.

9:39 You are absolutely joking right now.

9:41 This is absolute GG's.

9:43 I'm just like in my head calculating how long it would take

9:46 me to add all this stuff and it would be hours, yo.

9:48 It would be actual hours.

9:49 But, this can do it in matter of seconds.

9:52 I can't believe this.

9:53 Or that it was really good at adding polish,

9:55 which allowed me to get carried away and vibe all over the code.

9:58 This is so fun.

9:59 I don't want it to end, to be honest.

10:01 And here is the final product.

10:02 Now, the polish on this baddie is crazy.

10:05 I can click on any organism to learn more about it.

10:07 We can see its internals here, and it's all being simulated simultaneously.

10:11 I have some button controls over here in the bottom right.

10:14 You can see change the speed, speed it up, and you jump to the end.

10:17 Amazing.

10:17 I have a few different graph options to choose from down here,

10:20 population over time, elements over time, offspring.

10:23 Get this, I can even scrub through the timeline?

10:27 Geez.

10:27 And let me remind you, I didn't touch a single line of code to make this.

10:32 And now that the evolution simulator by Dreams is done,

10:34 I'm just letting it run in the background to see

10:36 if I can create life with some emergent behaviors.

10:38 But I've ran it for over 20 million frames.

10:41 It was like a 20 gig file, and no luck.

10:44 So, if that does happen,

10:45 I'm going to have to report back to you in another video.

10:47 But hey, [music] look, Earth took billions of years to create life.

10:50 You can't really expect my app to do it in a day, could you?

10:53 And finally, back to the question at hand.

10:56 Do I think that y'all need to check

10:57 yourselves into a psychosis clinic with the CEOs?

11:00 Nah.

11:01 In fact, I take back everything I said about them.

11:03 After this experiment, I can see what you guys see, and I think I'm wrong.

11:08 Well, kind of.

11:09 I'm not convinced on the claim that this is

11:11 going to replace all jobs, especially by 2026.

11:14 It still needs a [music] good level of hand-holding,

11:16 but it can do some really impressive stuff.

11:18 But I'm curious to know what the vibe coding experience

11:19 is like for someone who doesn't know how to code.

11:21 But regardless, the possibility of AI replacing a bunch

11:23 of jobs is much closer than I initially anticipated.

11:26 And I want to leave this video on a bit of a personal note,

11:28 what this experience has taught me.

11:30 I imagine you're a sailor, and across your 40-year career,

11:32 [music] you've always relied on physical maps to get chart the seas.

11:36 But then one day, something called Google Maps comes along,

11:40 and you have your thing.

11:41 You know how to navigate physical maps, you know how to navigate it well.

11:44 You can't be bothered with learning [music] user interface,

11:46 swipes, and hold the press, and why bother?

11:49 Now, the ocean hasn't changed at all.

11:51 So, in theory, you could chart the seas like that forever.

11:53 [music] However, the technology to the chart the seas has moved on.

11:56 And charting the seas doesn't have to be as [music] painful as you make it.

11:59 But you'll never know just how much pain you're

12:01 actually in when you refuse to update your tools.

12:04 That's kind of how I feel at the moment.

12:05 Heard about these agents for years,

12:06 [music] never wanted to touch it, and look where it got me.

12:08 I'm amazed by news that's probably 2 years [music] old at this point.

12:11 I hope you're all well.

12:12 Shout out to all the patrons on screen, and I'll see you on the next one.