Episode 14
Unreal Engine and AI Coding: My Hilarious Coding Misadventures
The recent push toward AI video generation felt like a massive shortcut, but I realized I was just playing creative roulette. You hit a button, wait for a result, and hope it matches the vision in your head. When it doesn't, you realize you have zero levers to pull to fix it. That lack of control is exactly why I’ve moved back into the heavy lifting of Unreal Engine 5.7. In an engine, if something is broken, you can actually reach in and fix the logic.
This update covers the messy reality of trying to merge traditional game design with modern AI tools. I’m currently wrestling with "vibe coding" using Claude Code and GeminiCLI to handle the friction of technical implementation. The goal isn't just to make a game; it's to build a world where NPCs aren't just reading from a script. By integrating local LLMs, I’m trying to create characters that can actually hold a conversation, even if they currently spend half their time walking into walls. It’s a transition from being a "generator" back to being a "creator."
Key Discussion Points
- The Illusion of Speed: Why AI video tools offer a fast start but often lead to a dead end because of the "black box" nature of the generation.
- Returning to Unreal Engine 5.7: Reclaiming technical control and why the complexity of a real game engine is a feature, not a bug.
- The "Vibe Coding" Workflow: Using Claude Code and GeminiCLI as high-level pair programmers to bridge the gap between a design idea and working C++ or Blueprint code.
- Local LLM Integration: The technical hurdles of hosting large language models locally to power dynamic, non-linear NPC dialogue systems.
- Solving the NPC Glitches: A look at the "coding shenanigans" involved when your AI characters have great things to say but still can’t figure out basic pathfinding.
- Version Control as Sanity: Why Git is the most important tool in the stack when you’re constantly breaking your project with experimental AI scripts.
Technical Stack & Ecosystem
- Engines: Unreal Engine 5.7 (Primary), Godot (Contextual comparison).
- AI Tooling: Claude Code, GeminiCLI, Local LLM frameworks.
- Concepts: Procedural Generation vs. AI Output, Human-in-the-loop Design, NPC Behavioral Logic.
- Contact for collaborations or feedback: cmonkxxx@gmail.com
Transcript
Hey.
Speaker A:Hello, friends.
Speaker A:Welcome back.
Speaker A:I'm back.
Speaker A:I'm back at home.
Speaker A:No more portable recording setups.
Speaker A:No more walking in the forest.
Speaker A:I'm right here in front of computer, where I guess you like me to be.
Speaker A:I mean, you're tuning in to hear my endeavors, frustrations, ideas, complaints and wishes for my Unreal Engine setup.
Speaker A:I have to say, this identity crisis that I had in recent weeks is kind of gone.
Speaker A:Identity crisis concerned me sticking with Unreal Engine or not.
Speaker A:I was really reaching out everywhere, looking for different ways, faster, easier ways of.
Speaker A:Of creating.
Speaker A:And in the end, I'm happy to say I came back where it all started.
Speaker A:I'm back to unreal.
Speaker A:That is not to say unreal is the best thing always.
Speaker A:It's not the best tool for the job.
Speaker A:All right?
Speaker A:I was looking intensively into AI video generation, only to realize that the.
Speaker A:The little bit of an accessible control that one loses in video generations is actually quite a lot.
Speaker A:Because it's your thing.
Speaker A:You don't want your video to look like AI generated.
Speaker A:And also you don't want to feel like you are generating something.
Speaker A:You want to feel like you're creating something.
Speaker A:So, yeah, I spent, I don't know, too much.
Speaker A:I feel like it's my duty in some ways as a creator to experiment.
Speaker A:So I did experiment with AI tools generating stuff locally and remotely.
Speaker A:I didn't get to see dance yet.
Speaker A:I don't think I will, because my conclusions are kind of the same.
Speaker A:It's still too expensive and too slow and there's not enough control.
Speaker A:And in the end, you get something like.
Speaker A:Like a roulette, you know, like you're playing a bit of a casino game here where you're putting the money in, you're putting the time in, you're waiting.
Speaker A:You're getting something that either works or it doesn't, and you just sink deeper and deeper in that sinkhole.
Speaker A:I don't like that experience.
Speaker A:I like to solve problems if I see them.
Speaker A:Maybe that's the main difference.
Speaker A:You see, when you get something generated, you may need to regenerate it again in order to get desired effect, but you're not really solving anything.
Speaker A:Do you just try again and again and again?
Speaker A:This is what my former boss told me when I told them, guys, you're losing your mind if you think AI will be giving you good designs.
Speaker A:And he said, well, you know.
Speaker A:He shrugged.
Speaker A:He said, well, if design is not right, we'll just generate it again.
Speaker A:You know what?
Speaker A:It's not the way.
Speaker A:And I see that.
Speaker A:I'M writing this.
Speaker A:You just cannot generate, generate, generate until you're done.
Speaker A:It's a stupid process.
Speaker A:It's a very unpredictable process.
Speaker A:You're just wasting time and money anyway.
Speaker A:Yeah, I do feel still a bit of a sting after getting fired because AI apparently does UX design.
Speaker A:But what I want to talk here about is 3D design.
Speaker A:So let me stay on topic.
Speaker A:Yeah.
Speaker A:So I no longer look into AI video generation.
Speaker A:I don't think I will be picking this subject app for the next couple of months at least.
Speaker A:I think it's my duty to revisit and see what's up, what's new, especially after what I'm seeing, what's happening with LLMs in general.
Speaker A:This Gemma 4 model from Google is incredible.
Speaker A:It's a very small LLM model that you can run locally and it's really, really good.
Speaker A:And I want to talk more about this because I've built, I actually built a game, I want to say three times, two, two and a half times, using different tools.
Speaker A:All right.
Speaker A:So I always wanted a game with NPCs that you can talk with.
Speaker A:You can have free, free range conversation.
Speaker A:And the plot of the game kind of generates itself like you have a conversation and then that creates the narrative as you're playing.
Speaker A:You are the architect of the whole game.
Speaker A:Right.
Speaker A:So my idea is that there will be a corner shop where you start the experience, you talk to the friendly clerk, and this clerk is powered by LLM and just story evolves.
Speaker A:You know, story just keeps on happening and adventures keep on happening.
Speaker A:So that's my vision.
Speaker A:All right.
Speaker A:And I vibe coded this using Python and it was like a web app with very rudimentary graphics.
Speaker A:And I realized maybe I don't need any game engine.
Speaker A:I will be just creating those web games, just some simple python games.
Speaker A:But then I quickly felt anxious to just improve the graphics.
Speaker A:Me being a designer, I wanted to improve graphics.
Speaker A:So I, I made these fake 3D models.
Speaker A:These were just isometric flat graphics that I told Claude and Gemini that I want to use them as 3D models.
Speaker A:And it did pretty good job creating boundaries around those pictures to pretend these are 3D objects.
Speaker A:I think it's something we've seen before in gaming.
Speaker A:Right?
Speaker A:It's a common tactic.
Speaker A:You don't do actually model.
Speaker A:You have those billboards, those, those flat graphics that look like 3D, but they're not.
Speaker A:So I did that and the game works.
Speaker A:And you know, a game talks to local LLM Gemma 4 and there are personality traits to that person.
Speaker A:There's memory.
Speaker A:And I don't repeat myself too much here, but I integrated a way to talk with this NPC using Telegram.
Speaker A:So when I get away from the computer, I can continue having conversation with my NPCs on the Telegram and having such a cool idea.
Speaker A:It's probably useless, but it keeps immersion of the game for a very long time or endlessly.
Speaker A:And you can be in your story while being away.
Speaker A:In theory, the NPC could call you.
Speaker A:Because what I also did, I implemented this TTS model that generates voice.
Speaker A:So in theory, if I put my mind to it, the NPC could call you and have conversation with you.
Speaker A:Wouldn't this be fun?
Speaker A:So I did that and then I started to think, okay, what if I use like a very simple game engine, right?
Speaker A:And now I remember I spoke about this right in the previous episode because I am becoming self conscious and I'm trying to remember how to pronounce Godot.
Speaker A:I want to say Godot, but it's Godot and it's a ultra light game engine.
Speaker A:And I've used, I don't know Godot, but I used Vibe coding to just completely Vibe code the game.
Speaker A:And I did it on the bus.
Speaker A:Hey, hold on.
Speaker A:Maybe I didn't record that yet.
Speaker A:I know I was attempting to record the episode and I scrapped it.
Speaker A:Okay, so on my way back from Poland, I was vibe coding this game using Godot.
Speaker A:So I was talking to Gemini cli, I was, you know, exchanging messages with AI and Gemini CLI was interacting with Godot.
Speaker A:And it created this first person perspective game that's kind of perfect.
Speaker A:It does the same thing, you know, it does.
Speaker A:There is local LLM chat, there is telegram functionality, and you can walk around and, you know, explore the world.
Speaker A:And that's kind of like a MVP version of my game.
Speaker A:And I did that without any hiccups, any whatsoever.
Speaker A:And I even asked Gemini to create some rudimentary placeholder models so created like figurines of people, cars, buildings.
Speaker A:Kind of cool.
Speaker A:Kind of cool.
Speaker A:So I arrived home.
Speaker A:I spent a couple of days in front of, in front of computer, obviously trying to catch up.
Speaker A:And I'm happy to announce I felt such a relief, such joy opening Unreal Engine with all the rudimentary graphics.
Speaker A:I had a lot of fun because it's so light and it's so easy.
Speaker A:But when I launched my environment, you might be hearing clicking of my mouse in the background because I'm actually now flying over my environment.
Speaker A:Like the, the graphics, the realism, the light design.
Speaker A:This is, this is part of the story, man, this is.
Speaker A:Yeah, Unreal Engine might be big over complicated and stuff, but you get, you really get value here.
Speaker A:You really get stuff in return.
Speaker A:And I missed it.
Speaker A:You know, I miss it.
Speaker A:I miss metahuman.
Speaker A:Yeah, it's not perfect, but aren't we spoiled here a little bit?
Speaker A:It's incredible system, man.
Speaker A:So what I did was I brought over experiences from my little, how should I put it?
Speaker A:Wild Card chapter with Godot and, and you know, like a raw coding vibe, coding.
Speaker A:And I implemented this cli, Gemini, CLI and Claude workflow with Unreal Engine.
Speaker A:And it kind of works.
Speaker A:I created a similar chat, NPC chat using local LLM.
Speaker A:And it works too.
Speaker A:You know, it works too.
Speaker A:Now what I had to do and it was, it cost me one day of going back and forth and almost giving up.
Speaker A:I had to convert my level into, you know, from, from like blueprint based into C. And only then, only then things started to make sense.
Speaker A:So right now CLI does things for me.
Speaker A:It edits it, it writes Python code and edits.
Speaker A:I think it edits blueprints.
Speaker A:I don't know, I think it might be doing it correctly.
Speaker A:The only difference is I'm stumbling a lot with more complicated features.
Speaker A:And I think I'm myself to blame here because I, I'm throwing two big problems on it, you know, So I accidentally ruined my NPC system.
Speaker A:I've got this mass crowd AI system where NPCs walk around my town.
Speaker A:And it was working pretty well, but there were some obvious naive, fake aspects to it.
Speaker A:Like, you know, all NPCs were working with the same speed.
Speaker A:They were really artificial looking, they were not very convincing.
Speaker A:So I asked Gemini to upgrade this system and I didn't really specify what I need, but it came back with suggestions saying, oh, we can create this state tree and there'll be different personalities, different behaviors, and the system will be more robust.
Speaker A:You can have like thousands of NPCs no problem.
Speaker A:I said, yeah, cool, let's do it.
Speaker A:And ever since, my NPC system is broken, all right.
Speaker A:And I'm trying to recover.
Speaker A:So I stayed up until late last night and I achieved nothing.
Speaker A:And it reminds me the frustration I had with AI video generation, where you just throw a problem on AI and you hope for the best.
Speaker A:Now in this workflow, it's a little bit like that, but it's not quite the same.
Speaker A:I feel like I'm collaborating here because I'm doing things manually also now it's no longer as tedious as it was before, like, I'm no longer looking for a tutorial.
Speaker A:I'm no longer looking those endless YouTube videos that are trying to constantly to sell you something.
Speaker A:I iterate with AI to understand what is this doing?
Speaker A:What do I need to enable?
Speaker A:Can you write something?
Speaker A:Can you solve this?
Speaker A:How do I solve this error?
Speaker A:For example?
Speaker A:That's huge.
Speaker A:You get this weird error report and you just don't know what to do with it.
Speaker A:And now I'm just copying and pasting this to Gemini Cli and I ask to explain and it either says, oh, I will fix it for you, or it tells me what to do.
Speaker A:It's amazing.
Speaker A:And I really want to commit a few more days to this workflow because I really feel like I can.
Speaker A:I just need to optimize it.
Speaker A:I feel like I'm not too far from this being perfect.
Speaker A:And it's just.
Speaker A:It's just incredible to think about it.
Speaker A:I really feel like I have this a coding partner with me and I can do what I was really hoping for to really focus on creating.
Speaker A:And the AI will do the rest.
Speaker A:Now it did the chat feature and I'll be going into that deeper.
Speaker A:It's really just like minimalistic version of it right now.
Speaker A:But it's a great promise.
Speaker A:Okay.
Speaker A:And I can always go back if I get really frustrated.
Speaker A:I can.
Speaker A:I can do sort of parallel project in Godot and I can.
Speaker A:I know how to make games quickly.
Speaker A:They won't look as nice as they look in.
Speaker A:In unreal.
Speaker A:But hey, we're not there yet.
Speaker A:Meaning we are getting there, but we're not there yet.
Speaker A:So.
Speaker A:Damn.
Speaker A:I kind of feel like I said everything I wanted to say and I don't know, maybe I made my point already.
Speaker A:Should I just end this episode?
Speaker A:So let me pause actually and collect my thoughts.
Speaker A:In fact, you know, I'm recording here and I'm trying to problem solve.
Speaker A:I'm really in a loop here.
Speaker A:I have this NPC system with NPCs walking in place, facing different directions.
Speaker A:They're a little bit glitchy.
Speaker A:I know something is wrong with the data.
Speaker A:What is it?
Speaker A:The crowd data set and stage three.
Speaker A:And I just have no idea how to go about it.
Speaker A:Gemini CLI seemed to be doing its best.
Speaker A:And I cannot tell if it's looping.
Speaker A:It's like trying to solve the same problem over and over again the same way.
Speaker A:And I'm just with it in this loop all the time because.
Speaker A:Okay, what I think I need to do is I may need to step back a little bit.
Speaker A:And ask it to.
Speaker A:Yeah, that's what we're gonna do.
Speaker A:We'll do some live vibe coding, we'll jump into the plan mode.
Speaker A:And again, I'm not a developer, right, But I feel like I know basic principles of how to code with AI.
Speaker A:So what we're going to do is we enable plan mode, which will focus more on planning and less on execution.
Speaker A:And I will tell it, no more errors.
Speaker A:NPC still walk in place, face different directions.
Speaker A:We have tried different methods to solve this, but nothing seems to work.
Speaker A:Analyze the problem and solutions you tried thus far.
Speaker A:Just this because I don't know what's going on.
Speaker A:It definitely doesn't know what's going on.
Speaker A:Now if this doesn't work, I will switch models.
Speaker A:I'll try with the most advanced claude, which I still have until I think for another week or so.
Speaker A:I decided not to extend a subscription for that because it's.
Speaker A:I don't know if you're up to date, but CLAUDE is now currently it's the best model out there.
Speaker A:But those companies are now becoming more transparent about how those things, how much those things cost.
Speaker A:And this is where local LLM, like Gemma 4 for example, this is where it makes a difference.
Speaker A:You know, the fact that we can use local LLM for free, Claude being the best, it has limits, like actual limits where you cannot use it too much.
Speaker A:You can easily exhaust your limits if you don't plan accordingly.
Speaker A:And it's a really bad user experience.
Speaker A:It's unfortunate.
Speaker A:The bottom line is those tools are very expensive.
Speaker A:We are a little bit spoiled at this point.
Speaker A:But when a company starts to be transparent how much they actually want to charge you, they can bump the price up, which is something they do sometimes.
Speaker A:But now the tactic is they will incorporate those limits and they will ask you to, to pay more or wait.
Speaker A:Okay, so let's see.
Speaker A:Gemini did some thinking as I was rambling and came up with this report.
Speaker A:So it says based on your logs compilation results and the president working in place behavior, here is a complete analysis of the problem and breakdown of everything we have tried so far.
Speaker A:So we see here.
Speaker A:I can see that it didn't really try to throw one solution all the time.
Speaker A:He was looking to missing fragments.
Speaker A:Fix brute force movement processor.
Speaker A:That sounds good.
Speaker A:The actor sync override.
Speaker A:Yeah, a couple of other things here, let me pause.
Speaker A:I'll read this with comprehension and then let's see.
Speaker A:Yes, I'm reading, I'm pausing, recording.
Speaker A:So don't be confused here.
Speaker A:I just read a bit and it says the problem is fundamentally rooted in.
Speaker A:The problem is fundamentally rooted in how Unreal Engine and Blueprints are interpreting the mass AI setup, their primary culprits that are as follows.
Speaker A:So basically, I feel like I'm dealing with task that's just way too complex to approach it the way I did.
Speaker A:And I feel like this workflow that I'm trying will be more successful if I break it down more into smaller tasks.
Speaker A:And now I kind of told it to build the game, you know, not really, but kind of.
Speaker A:It's just too big of a request.
Speaker A:And now it's.
Speaker A:It's a bit of a mess right now, and I feel like I may need to recreate my mass AI NPC system manually because, yeah, it may be just better use of time.
Speaker A:All right, so here at the end of this message, there were three points titled what to look at next outside of C. So there were some things AI asked me to look at, which is, I feel like this.
Speaker A:This will no longer be a step in the near future.
Speaker A:Like, we see this functionality already where agents can use the computer and they can check things by themselves, but in this workflow, it's not possible.
Speaker A:So it asked me to check a couple of things, and I did.
Speaker A:And you know what?
Speaker A:I appreciate that because I stay in touch with the flow, right?
Speaker A:I get to.
Speaker A:I get to know where things are and I get to understand how things work.
Speaker A:It's not entirely a black box for me.
Speaker A:So it gave me a hint to check a few things, and I did.
Speaker A:And I took a screenshot of each section and I uploaded screenshots just to confirm how things look, because it can understand information from screenshot, which is absolutely great.
Speaker A:I've been doing this back and forth loop asking if this is configured correctly, and I used to show screenshot to demonstrate what I did.
Speaker A:Okay?
Speaker A:So now as I was rambling, it's analyzed three screenshots I gave it, and it said, perfect.
Speaker A:These screenshots reveal exactly why the Walking Place issue is happening, and it entirely explains why the C brute force fixes didn't work.
Speaker A:All right, so you see, this is kind of feedback and interaction I would never get if I was just throwing command at it.
Speaker A:Right?
Speaker A:So let's see.
Speaker A:Here are the specific settings in the editor that you need to change.
Speaker A:Fix the root motion setting.
Speaker A:Screenshot 1.
Speaker A:Okay, let me pause here.
Speaker A:Okay, I went through those suggestions and you see what happened.
Speaker A:Now let's recap.
Speaker A:It gave me things to do, right?
Speaker A:I did it.
Speaker A:I documented with screenshots.
Speaker A:I showed it the screenshots of what I did.
Speaker A:And it found subtle little things which we all know unreal gives a lot of opportunities to miss something.
Speaker A:Not to toggle a toggle switch or not to add some kind of a trait or a tag or whatever.
Speaker A:So it picked up on couple of things plus something.
Speaker A:What will actually solve the problem?
Speaker A:I mean, let's see.
Speaker A:But apparently animation was rooted in place, which is.
Speaker A:I don't know.
Speaker A:I mean, I know why I didn't think of it last night.
Speaker A:I was tired.
Speaker A:I finished at like 1:20am you know, working so late feels good, but it's not very effective, is it?
Speaker A:Okay, you see, So I did that, and my NPCs are still working in place.
Speaker A:I'll try to recompile.
Speaker A:So it gave me those ideas what to do, and I did it, and it still doesn't work.
Speaker A:You see.
Speaker A:I really hope this will be a nice, nice arch for this story.
Speaker A:Arch for this episode.
Speaker A:My coding agent was mistaken.
Speaker A:Okay, let's pause again the recording.
Speaker A:And I will feed this back into AI let's see what happens.
Speaker A:I told it that this didn't change anything.
Speaker A:And it's thinking again now.
Speaker A:Interesting thing about AI is it's never giving up.
Speaker A:Right.
Speaker A:It will start lying to you, which I. I don't know.
Speaker A:Maybe I encountered it, but I couldn't tell.
Speaker A:But it just keeps on coming up with solutions, and it's really fascinating to see.
Speaker A:Again, that was sort of like the most complex way to approach this problem, and it didn't bring results I wanted.
Speaker A:So in this live demo of wipe coding in Unreal Engine, I will probably ask it to summarize where we are, and I'll take it to Claude, and I will document that on this episode as well.
Speaker A:All right.
Speaker A:Another change of plans.
Speaker A:So I was about to write to Gemini, okay, give me the handover.
Speaker A:I will move it to Claude.
Speaker A:And it said, this behavior is a symptom of character movement component conflict.
Speaker A:Here's exactly what is happening.
Speaker A:Okay.
Speaker A:It sounds really convinced.
Speaker A:And then it says, you know, it says what?
Speaker A:It's happening, which is really interesting.
Speaker A:And I don't think it's hallucinating.
Speaker A:And then there is the final fix.
Speaker A:Okay, and what is.
Speaker A:Basically, it asked me to edit crowd character blueprint, which I now did.
Speaker A:I compiled it.
Speaker A:I will save the changes because I don't want to go through it again.
Speaker A:And it doesn't work.
Speaker A:It does it.
Speaker A:Oh, God.
Speaker A:Now this is what's confusing.
Speaker A:It's.
Speaker A:It's really good at looking at things.
Speaker A:And I Feel like it's maybe not seeing the full picture, and it's sort of focusing on just, you know, sort of washing dishes on the Titanic.
Speaker A:You know, as the expression is, I think.
Speaker A:Washing dishes on the Titanic.
Speaker A:Yeah.
Speaker A:So I feel like we fixed collisions, and now my NPCs are waist down on the ground.
Speaker A:So, yeah, no, collisions mean they.
Speaker A:They're not sinking in the ground.
Speaker A:Okay, let's pause and try again.
Speaker A:All right, I complained, and we'll see what's happening.
Speaker A:And I feel like maybe I'm not very productive right now, but I feel like I'm learning a lot because, you know, in the previous explanation, what he told me is, for example, stop AI controller from interfering, go there, click on that, change this.
Speaker A:And then it explains, Now, AI is the brain having standard AI controller.
Speaker A:Pauses, the pawn will continue to hit the brakes.
Speaker A:Since it doesn't have a standard nav mesh path, it gives me those explanations of what's happening and how do you control it?
Speaker A:And I feel like I am learning things.
Speaker A:Maybe I'm not solving the problem, but I feel like I'm learning.
Speaker A:And maybe Gemini is not hallucinating, but it's trying to find those solutions.
Speaker A:It's kind of nuts.
Speaker A:I feel like I'm being played also because it said it's a final fix, and it wasn't.
Speaker A:And I said, well, thanks for Nothing.
Speaker A:Once again, NPCs are working in place, but now they are submerged under the ground and face different directions and walk in place.
Speaker A:Now what?
Speaker A:So I'm.
Speaker A:I'm becoming less technical now in those conversations because frustration grows.
Speaker A:I have still maybe, like, maximum two hours of free time.
Speaker A:I don't want to spend it like this.
Speaker A:That's the thing.
Speaker A:I really don't want.
Speaker A:I really don't know when to give up, because I feel like AI gives me solutions all the time, and I feel like I'm just step away from fixing this.
Speaker A:Meanwhile, I think what's happening is Gemini might be forgetting what it did, and we're just closing loops, walking circles, and Gemini doesn't realize it.
Speaker A:So ultimately, I think, yeah, I need to.
Speaker A:Oh, give me a question which is new.
Speaker A:The reason NPCs are waist down, submerged, and walking in place is that masscrowd disables internal movement for spawned metahumans expecting their component to move, usually via NAV mesh.
Speaker A:Okay.
Speaker A:Because they have no AI controlled NAV mesh, they stay still.
Speaker A:Okay, you see, that makes sense.
Speaker A:I'm assuming it's true, but it kind of makes sense.
Speaker A:Okay, so it's asking Me how I would like to handle this.
Speaker A:Pure kinematic C movement recommended ignore character movement component.
Speaker A:We will write C to read the zone graph, calculate this exact movement step and force the actor to move with a plus 90z offset so they stay on the ground.
Speaker A:And then the second solution is standard air controller navmesh.
Speaker A:We stop using C to force movement.
Speaker A:Instead we will add navmesh to your level.
Speaker A:Well, yeah, that's kind of, I think the way.
Speaker A:Yeah.
Speaker A:So, okay, we'll try that and then I'll take it to Claude and I think there'll be last pausing of this process.
Speaker A:I.
Speaker A:You know, I. I didn't plan vibe coding on this episode, but I guess I couldn't decide if I want to record or if I want to continue working on this.
Speaker A:But it just gives me an idea.
Speaker A:Maybe I will create a video stream one of those days.
Speaker A:And I will be vibe coding live with Unreal Engine.
Speaker A:We'll do an experiment like that.
Speaker A:I can give you exact date when we could do it.
Speaker A:We would do it on Wednesday or Thursday.
Speaker A:One of those days this week we can do it.
Speaker A:If I solve this MPC issue, we can take another task.
Speaker A:And I feel like I know what I want this to be.
Speaker A:And we will work on something small with AI.
Speaker A:We are all learning here.
Speaker A:And I hope you're enjoying the ride.
Speaker A:I know I do.
Speaker A:All right.
Speaker A:At this point, I feel like it's irrelevant if I find solution or not.
Speaker A:I feel like the journey is the destination.
Speaker A:You know.
Speaker A:Okay, I will.
Speaker A:I'll pause again.
Speaker A:I will take this stuff.
Speaker A:I'll take this mess to Claude.
Speaker A:I will explain that I desperately need the solution.
Speaker A:And I'll see if Claude does it.
Speaker A:If it continues to.
Speaker A:To stall me.
Speaker A:I will just rebuild it myself manually.
Speaker A:Because I did this, I know how to do.
Speaker A:All right.
Speaker A:Half an hour later, I'm still here.
Speaker A:I'm talking to Claude.
Speaker A:Claude seemed to be equally engaged and creative and very convinced that it's.
Speaker A:It knows what it's doing.
Speaker A:Now, I don't want to say that this tool is useful just because my experiment is failing, because it does.
Speaker A:I just feel like there is.
Speaker A:There's some kind of fundamental problem in a feature that I'm trying to fix, to implement.
Speaker A:And I just gonna say, just because this was a failure doesn't mean that the workflow doesn't work.
Speaker A:So what I'm gonna do is I'm gonna rebuild this NPC feature on my own manually.
Speaker A:Gonna start from scratch, basically, and.
Speaker A:And I will be working towards perfection of this Vibe coding AI supported coding in C based Unreal Engine project and I will do it in more structured way.
Speaker A:Also something I didn't do consistently is the use of git.
Speaker A:Git is a system of backup control.
Speaker A:Basically it's it's useful exactly in a situation like this where I made a mess and I need to go back to state where things were fine, I have git versioning enabled but I didn't record.
Speaker A:I didn't make a record in time where things were optimal from the NPC perspective.
Speaker A:So effectively I don't have that git record that I can rewind to, but it's something I will be using more consistently and yeah, my bad, my bad.
Speaker A:But I messed up in many ways in this project already so I'm going to solve it.
Speaker A:I'm going to end this episode.
Speaker A:Unfortunately, we have no closure.
Speaker A:But maybe you're listening to this and you're boiling out of frustration because there's something you want to tell me and feel free to do so.
Speaker A:You can email me@simonxxmail.com and you'll find this address also in the episode description.
Speaker A:I always ask for people to email me and get in touch.
Speaker A:This is definitely something I want this podcast to help me achieve and to connect with people in a meaningful way.
Speaker A:If there's something you want to say, ask, comment on, or critique, let me know.
Speaker A:I'll greatly appreciate it.
Speaker A:Now I will need to end this episode because there are other things that concern my life that I need to address.
Speaker A:And yeah, I hope this was useful.
Speaker A:Take care and stay productive.
Speaker A:Bye bye.
