I was really excited about this hackathon because it was an offline event and completely focused on prompting and building a web game using AI.
Bu...
For further actions, you may consider blocking this person and/or reporting abuse
This is a genuinely useful post because it documents exactly the gap between "AI generates code" and "AI generates correct code under pressure."
The hackathon setting makes it even more telling. When you're vibe coding for a demo, the stakes are low — if it breaks, you regenerate. But the habits you build there carry into real projects where the cost of a subtle bug is way higher.
What I keep coming back to: the bottleneck was never generation speed. It was always verification. You can get AI to write a game loop in 30 seconds, but figuring out whether that game loop actually handles edge cases correctly takes the same amount of human attention it always did. Maybe more, because the code looks plausible enough that you trust it.
The METR research on AI-assisted coding showed developers perceived they were 24% faster while actually being 19% slower. The gap comes from exactly what you described — the time spent debugging and re-prompting eats the generation speed advantage.
Not saying vibe coding is useless. But knowing when to switch from "generate fast" to "verify carefully" is the real skill. Sounds like the hackathon taught you that under pressure, which is the best way to learn it.
I'm really glad you found my article useful.
Yes, you are absolutely right about the verification and you have made a valid point of game loop actually handling the edge cases. This got me so much frustrated as I wanted to add different levels to the game and make it more difficult for the user to play but while doing so AI started doing things on it's own. The original game logic for loosing was kinda lost while building levels in the game. So, yeah I agree with you it doesn't make us faster it actually makes us slower cause the execution can divert into any possible way and I have to teach AI to do how I want it.
Yes, I learned a lot about AI Studio, Antigravity and Prompting while building the game. If you are willing to learn even a small ant can teach us amazing lessons.
What's was the bug you found most annoying while debugging?
This is such a real take! It’s easy to think AI tools make building a game simple, but the 'pressure test' of a hackathon always reveals the gap between prompting and actually debugging. What was the most unexpected bug the AI introduced for you?
I'm pleased that you liked it. Yeah, pressure test really shows the capabilities of AI in real life like how would a developer react to a 2am call for a production bug but for me the gap was larger.
The most unexpected bug the AI has introduced to is when I prompt it to do some changes in a long conversation it tends to redo the tasks I prompted it to remove. Like, I'm waiting for an output but all I get is new output and previous outputs. It really creates a mess in the code and also in AI studio it can add files but can't remove them, I prompted it many times as it was an unnecessary file but it couldn't delete the files.
What's an unexpected bug you faced while vibe-coding or any interesting story you wanna share?
That's such a relatable experience! 😅 The 2am production bug analogy is spot on — pressure testing really does expose the gaps.
And yes, that issue with AI redoing old tasks in long conversations is painfully familiar! It's like the model loses context of what 'remove' actually means and just keeps adding layers. Definitely creates chaos in the codebase.
The AI Studio file limitation is weird too — being able to add but not delete? That's such a basic need. Hope they fix that soon.
As for unexpected bugs I've faced — one time the AI kept importing the same library 5 times in one file no matter how many times I told it to stop 😂 Took me longer to clean up than to write it myself!
Would love to hear more about your hackathon experience — what were you building?"
Yes, it has happened to me as well. I told it to remove the library but with newer instructions it kept adding it.
It was a wonderful experience every hackathon comes with it's pros and cons and as for the building part I have provided the link in the submission attempts. Though, it is still in prototypal phase but I included it to share what I have built on. Here's the link for you.
Glitch Hunt
What's your hackathon experiences or have you taken part in a Dev Challenge how was it?
The hallucination-under-pressure pattern is consistent — models get more confident-sounding exactly when context degrades, which is the worst time to trust them. The planning phase working well but execution breaking down is exactly the gap most vibe-coding workflows haven't solved yet.
Yes, exactly what I faced while working with vibe coding workflows. I think they lack in the execution of the context because if I say do Task A then it will do Task A + B resulting in a more chaotic output.
What's your favorite vibe coding tool you use?
Hi @konark_13, I'm building a code editor, Stellify, that's all about preventing AI (and human developers, remember them?) from going off track in the way you've described. The main things it does so far:
Uses a set stack, PHP (Laravel on the server), my own JS framework (with adaptors for Vue, React, Svelte) on the client. When AI starts a task, the first call it makes is to a get_project tool that gives it info about the stack, the libraries that are available (plus versions) and key references (see the next bullet).
Stores code as granular JSON in a database with persistent references and other metadata. This allows AI to be surgical when it comes to making updates and keep the context window to a minimum. When you're ready to export/ deploy the JSON is re-assembled back to source files.
In terms of prompts, I have an intermediate layer that analyses your prompt for keywords and attaches only relevant instructions. The API also provides, code analysis and testing endpoints that AI can use (via MCP) to keep it on track.
The next (hopefully last) piece of the puzzle, is assisting with planning prompts to optimise for clarity, not fully sure how best to do this given the vast array requirements that are possible. In theory I like the idea of capturing prompts and running a "prefetch" to AI that breaks apart your prompt and makes it optimal before passing he refactored prompt to AI. I'm currently building lots of projects to get a feel for what's right here as it's hard to reason about (at least for me).
I'd be interested to hear your thoughts!
Hi Matthew, Stellify seems like a wonderful idea and an idea that most of the people to help be a better developer. The tech stack that you have used is also good and the workflow that you have mentioned seems good. I can tell more about it to you after I have used the product and all the best for your launching on product hunt.
For optimizing the prompts what I have learned is you can use the classic way of tokenization, embedding and vector searches this will help you analyze it and get better context. But your idea of running a "prefetch" to AI that breaks apart prompts and make it optimal will be more costly as most of the AI models runs on tokens and with tokens there's a limit for it. For optimizing more tokens will be generated and that will cost you and the user for using LLMs behind the scene. If you can generate your own code to do what LLMs does then the cost can reduce and you can also train the model specifically on just optimizing.
The idea and execution is amazing. Keep up the good work.
Hi Konark, I appreciate the kind words! You make a valid point about the increase in tokens/ costs. What I've ended up doing as a result of lots of testing (essentially building apps on repeat and analysing the outcome) is injecting defensive code into my api endpoints. I'm really happy with this approach, it's much more effective than trying to engineer the prompt either by prefetching or simply adding more and more instructions to guide AI.
Here's a YT video I put together showing Claude building a Feedback application.
Although I'm not a vibe-coder, I'm a developer. It actually gives couple of interesting facts to bear in mind as I will attend a hackathon in April. Good to know, thanks for the article !
I'm really glad you liked the article. Thank you so much for your support. Give your best in the hackathon maybe we will get to see an amazing build by you and shared in form of an article. Looking forward for your experience.
Hopefully, that hackathon will be my official farewell with web-dev and smart-contracts and focusing only on cryptography and blockchain architecture (What I actually do currently). You can read my first article where I introduce my self also feel free to connect on signal or telegram.
For sure, I would read that. I am sure you will lead the way of cryptography and blockchain on Dev.to. I would love to connect and learn more about blockchain from you. It is the most fascinating thing I like in Web3.
This AI hype is a skill degraiding phase for ppl invested in it. Programmers who don't engage with it will have an advantage over time. Don't print, use the brush.
Great writeup. Your lesson #1 really resonated — "plan before you vibe code."
I hit the same wall but earlier in the process. Before I even start prompting, I kept finding out the thing I wanted to build already existed. Claude would happily help me code a food delivery tool for 6 hours, then I'd search GitHub and find 12 mature competitors.
The hallucination problem you describe during execution is real, but there's an even earlier hallucination most people miss: when you ask "is this idea original?" and the AI says "yes, go build it!" without actually searching anything.
I ended up building an MCP tool that does that search automatically — scans GitHub, HN, npm, PyPI before you write a single line. Would've saved me a lot of the "reset and start over" cycles you described between rounds.
Curious — for your Glitch Hunt game, did you check if similar web-based duck hunt remakes existed before building? Not criticizing at all, just wondering if knowing the landscape upfront would've changed your approach or scope.
Thank you so much. I'm really glad you found my article helpful.
Yeah, "the original idea" and "go build it" are the hallucination for us to keep us in delusion that what we wanna build never existed before. Even if it searches anything it could even then tell you to build it with slighter modifications.
Wow, so cool making that MCP tool. Do share the link if it is publicly available might help me and other developers to research the before even starting planning.
Yes, I had researched but couldn't found a proper one and due to time constraints I didn't have time to do proper research. If it had already existed then I would have taken the inspiration from it and made my own since the theme provided wanted us to make one. Even if the web-based duck hunt did existed. Either I would have gone for another game that wasn't build or I would have added AI in the game. If you have played the original duck hunt the dog laughs at you when you misses the shot and mocks you. Those taunts I wanted to add in my game as well I used Gemini in the backend to generate taunts based on the user behavior but before it was even launched I finished my quota so dropped and choose random texts for taunting the users.
It is so true
I'm really glad you liked it. You are also providing such amazing articles. Keep up the good work.
Ah, thank you!
I'm still trying to wrap my head around the idea of building a game with AI tools. Konark, your experience with hallucinations and deployment issues is a great reminder that these tools are still so new and finicky. I love that you're documenting what you've learned, especially the importance of verification and human judgment, it's a crucial balance to strike when working with AI.
Thank you so much Aryan for your kind words. Making game was basically the theme provided to us for the hackathon so no choice left.
Yes, I love to share my learning and learn from other what's their approaches towards building with AI is. What's your take on it and what are the problems you faced while using AI tools?
That's very cool of you to share these learnings, they are very valuable indeed.
I have also experienced almost the same problems as you, and a fresh start has been the answer most of the times. While others include my ideas being too abstract or complex to vibe code and I've usually tackled this one by breaking down things into smaller problems, nothing fancy just the fundamentals of SDLC done right.
Thank you so much Aryan for finding my lessons valuable. I share my learning so that everyone can learn and avoid the mistakes I did and also share their experiences for me to learn from.
I myself felt my ideas are a bit complex to build using vibe code. I maybe wrong but I feel the more complex you make the website the more hallucinations the model goes in. It keeps building my frustration because I can't fully be a vibe coder I need to debug as well. But yeah your approach of breaking into smaller problems and using SDLC as a building block is the most common step that everyone should take.
Thank you for mentioning peer feedback. This is one thing I wish I had more of when I'm building.
Of-course, Peer feedback is also important and valuable. If you are building you can share your app/website on this platform cause people here are very supportive and helping. So, I'm sure the feedback you get here and can make your product 100 times better.
This is sow true!
ITHINKSO