6 ms·

I have a good story to share that I came across recently. Around 2 years ago I had to clean up a mess because someone who doesn't really know what they're doin

I have a good story to share that I came across recently. Around 2 years ago I had to clean up a mess because someone who doesn't really know what they're doing designed an instancing system for a game. They heavily used AI to design every part of it and it was awful. Data corruption, performance problems, lost items, race conditions everything you can think of was an issue. It took me 2 weeks just to get it to an "acceptable" level and it was still awful as the whole design was simply flawed. Fast forward to today: different company, same person, SAME issues with an AI that is 'allegedly' much better than it was. This time I only heard about these issues and wasn't the one who had to deal with it so I just had a really good laugh. AI is only as good as the person using it, that's why we have such vast range of what people "claim" AI can do and why everyone has way different opinions of it.

onlyrealcuzzo 5mo ago
> They heavily used AI to design every part of it and it was awful. Data corruption, performance problems, lost items, race conditions everything you can think of was an issue. Presumably that's better than no game at all. > It took me 2 weeks just to get it to an "acceptable" level and it was still awful as the whole design was simply flawed. Doing 2 weeks of fixing might have been hell, but this sounds like it was overall still a great deal for the company. You're not really selling the "AI is useless story". It might be, but your anecdote seems like just another case of AI being worth it, though obviously flawed.
- himata4113 5mo ago
  No the game failed because of these non stop issues, it lost all hype and the "fixed" version couldn't sustain itself anymore as it took around a month to fix the remaining issues. At the 2nd company it wasted thousands of dollars of advertisement because the server could not withstand the load and obviously data loss issues tained the image forever and will likely end up the same way. Also please don't take this as "AI is useless". I use AI and I use it a lot. It's great and I love it. However, without a good understanding of architecture and general development structure you end up with things that can't scale and fail.
  - bitexploder 5mo ago
    What particularly gets me is if you use AI with a bit of engineering rigor, especially around design and testing my experience is the latest models are great to work with. They can structure performance and stability tests, implement 90%. Humans have to do the hardest and critical 10% of the design. The current tools are good enough to do virtually all of the implementation now if your quality gates are right and your design is good enough, but you absolutely have to design the right things for your scale and reliability needs or very bad things are in store.
  - darkerside 5mo ago
    If it were really groundbreaking, I imagine it wouldn't have burned out after a little missed hype. See No Man's Sky. The other way to look at this is, thank goodness we didn't waste months or years on a failed game concept. Instead we got to market and validated (or invalidated) the concept fast.
    - himata4113 5mo ago
      it's a validated concept with a popular youtuber behind it, tried and true method that worked for years all brought down by game being non functional.
  - onlyrealcuzzo 5mo ago
    > No the game failed because of these non stop issues You should've led with that then. The company is likely to disagree and think it failed for a number of reasons, that only being one of them, and still depending on the cost may be very happy with their decision. For one, even if it was a complete dumpster fire disaster, that is at least potentially a learning opportunity. Whether they saw it as one is a different thing entirely. If they think they can make games for 1/10th the engineering price, they are likely going to try until proven otherwise. It's harder to convince them, no, it can't be done, just trust me, bro, I know from experience of never even trying.
    - himata4113 5mo ago
- surgical_fire 5mo ago
  > Presumably that's better than no game at all. Wrong. A bad game can absolutely tank a studio. Shipping a game that has awful reviews will absolutely affect negatively your sales for future games.
  - himata4113 5mo ago
    Luckily this was minecraft so stakes aren't as high, but it definitely tainted the reputation of the company behind it and youtubers will likely never work with them again.
- zaphirplane 5mo ago
  Why does it “sound” like a good thing for the company? Unless it’s a mega establishment product people move on and don’t stick with buggy crashing products This wouldn’t be acceptable for a car safety, well I could like a whole bunch but you should get the idea
otar 5mo ago
> AI is only as good as the person using it, that's why we have such vast range of what people "claim" AI can do and why everyone has way different opinions of it. Banger statement.
- jjav 5mo ago
  > > AI is only as good as the person using it, that's why we have such vast range of what people "claim" AI can do and why everyone has way different opinions of it. > Banger statement. True, but it's even a bit more removed than that. For a version 0.9 prototype, AI can produce pretty awesome-feeling output even in the hands of the most incompetent. So it looks like a miracle that does it all, with zero effort. It's later down the road that it starts hitting walls that need a competent architect in charge.
  - daheza 5mo ago
    We’ve got this C Level guy at my work, who’s in charge of the AI offerings at our company. Every few weeks he does a presentation to the entire company where he pulls up this vibe coded locally hosted html file that is supposed to be this huge new workforce solution. It’s so scary to see him present this as if it’s working and not just a locally hosted mockup at best. Sales and Marketing do not see this as just a design presentation instead seeing something they can actively sell. I don’t know when this will backfire on us, when I brought it up to our VP he said “well maybe he’s selling the idea and a few years down the road after the contracts are signed they will build it”. I think this is just how corporate America is going to work now.
- sph 5mo ago
  Shades of that famous quote by Kurt von Hammerstein-Equord. What is terrifying is that those that claim 100x increased productivity are those with the least wisdom to tell good code from bad. They are the ones inundating the world with utter slop.
rco8786 5mo ago
> AI is only as good as the person using it Nailed it. That said, it couldn't have possibly been that bad if you got it to "acceptable" in 2 weeks.
- himata4113 5mo ago
  I used AI too and acceptable means "you can play it without losing your progress".
bunderbunder 5mo ago
> AI is only as good as the person using it I think it might be even worse than that. It seems to be a multiplier for the Dunning-Kruger effect. Possibly because being trained to exhibit positive demeanor means that it will always tell you you're the best, no matter what.
- himata4113 5mo ago
  Not sure why I saw this getting downvoted. This is exactly what is happening to that person, they think they are way more capable then they really are that's why they're taking on hard challenges and high complexity tasks when they're not ready for it.
  - bunderbunder 5mo ago
    And I do think it's important to be aware of this. When I've got a good idea, the agent says, "Good idea!" When I've got a bad idea, the coding agent also says, "Good idea!" And I'm most inclined to just go with what when I'm outside my realm of expertise, because that's the time when the coding agent promises to save me the most time. Because developing a proper understanding of things like the problem space and tech stack is often the most time-consuming part of the job.
hmokiguess 5mo ago
What is an instancing system?
- duckmysick 5mo ago
  To better scale the game infrastructure, instead of letting every player on the same big server, they are put in different smaller servers (instances). Players can interact with other players on the same instance (see them, do quests together, trade, etc.). Sort of like AWS instances. Normally instances are random within the same region, but usually there's a system in place so you can join the same one as your friend.
  - hmokiguess 5mo ago
    Ah gotcha, I never heard that called that way. I'm familiar with matchmaking systems and VM provisioning, but never heard the term "instancing" haha. Thanks for answering me!
amarant 5mo ago
You've hit the nail on the head as we say where I'm from. This is also why I think the "boycott AI" movement is misguided. AI doesn't produce slop: unskilled AI operators do. Heck just the other day I saw a headline about a Nobel literature laureate apparently using AI, with some "expert" confidently claiming the winning novel was 100% generated. AI output quality ranges from slop to Nobel price worthy, depending on who uses it. Which seems to support the notion that it's a tool, much like any other.
epolanski 5mo ago
This reminds me of antirez saying that he made extensive use of AI to implement arrays in Redis. He spent 3 weeks (which for him likely means 60+ hours a week) only iterating on the design with different models and not writing a single line of code.
mizzao 5mo ago
2 years ago AI coding agents had barely emerged yet, right? Couldn't have gotten good results even if a good engineer was forced to use it...
- himata4113 5mo ago
  People in the know already had coding agents, they just needed a lot more back and forward as they weren't trained to be "agentic" and they were mostly cli based... then claude code was born because anthropic saw the potencial. The person in question used warp.dev which happened to bundle AI into so they stumbled into it more than anything.

onlyrealcuzzo 5mo ago

> They heavily used AI to design every part of it and it was awful. Data corruption, performance problems, lost items, race conditions everything you can think of was an issue. Presumably that's better than no game at all. > It took me 2 weeks just to get it to an "acceptable" level and it was still awful as the whole design was simply flawed. Doing 2 weeks of fixing might have been hell, but this sounds like it was overall still a great deal for the company. You're not really selling the "AI is useless story". It might be, but your anecdote seems like just another case of AI being worth it, though obviously flawed.

himata4113 5mo ago
No the game failed because of these non stop issues, it lost all hype and the "fixed" version couldn't sustain itself anymore as it took around a month to fix the remaining issues. At the 2nd company it wasted thousands of dollars of advertisement because the server could not withstand the load and obviously data loss issues tained the image forever and will likely end up the same way. Also please don't take this as "AI is useless". I use AI and I use it a lot. It's great and I love it. However, without a good understanding of architecture and general development structure you end up with things that can't scale and fail.
- bitexploder 5mo ago
  What particularly gets me is if you use AI with a bit of engineering rigor, especially around design and testing my experience is the latest models are great to work with. They can structure performance and stability tests, implement 90%. Humans have to do the hardest and critical 10% of the design. The current tools are good enough to do virtually all of the implementation now if your quality gates are right and your design is good enough, but you absolutely have to design the right things for your scale and reliability needs or very bad things are in store.
- darkerside 5mo ago
  If it were really groundbreaking, I imagine it wouldn't have burned out after a little missed hype. See No Man's Sky. The other way to look at this is, thank goodness we didn't waste months or years on a failed game concept. Instead we got to market and validated (or invalidated) the concept fast.
  - himata4113 5mo ago
    it's a validated concept with a popular youtuber behind it, tried and true method that worked for years all brought down by game being non functional.
- onlyrealcuzzo 5mo ago
  > No the game failed because of these non stop issues You should've led with that then. The company is likely to disagree and think it failed for a number of reasons, that only being one of them, and still depending on the cost may be very happy with their decision. For one, even if it was a complete dumpster fire disaster, that is at least potentially a learning opportunity. Whether they saw it as one is a different thing entirely. If they think they can make games for 1/10th the engineering price, they are likely going to try until proven otherwise. It's harder to convince them, no, it can't be done, just trust me, bro, I know from experience of never even trying.
  - himata4113 5mo ago
surgical_fire 5mo ago
> Presumably that's better than no game at all. Wrong. A bad game can absolutely tank a studio. Shipping a game that has awful reviews will absolutely affect negatively your sales for future games.
- himata4113 5mo ago
  Luckily this was minecraft so stakes aren't as high, but it definitely tainted the reputation of the company behind it and youtubers will likely never work with them again.
zaphirplane 5mo ago
Why does it “sound” like a good thing for the company? Unless it’s a mega establishment product people move on and don’t stick with buggy crashing products This wouldn’t be acceptable for a car safety, well I could like a whole bunch but you should get the idea

otar 5mo ago

> AI is only as good as the person using it, that's why we have such vast range of what people "claim" AI can do and why everyone has way different opinions of it. Banger statement.

jjav 5mo ago
> > AI is only as good as the person using it, that's why we have such vast range of what people "claim" AI can do and why everyone has way different opinions of it. > Banger statement. True, but it's even a bit more removed than that. For a version 0.9 prototype, AI can produce pretty awesome-feeling output even in the hands of the most incompetent. So it looks like a miracle that does it all, with zero effort. It's later down the road that it starts hitting walls that need a competent architect in charge.
- daheza 5mo ago
  We’ve got this C Level guy at my work, who’s in charge of the AI offerings at our company. Every few weeks he does a presentation to the entire company where he pulls up this vibe coded locally hosted html file that is supposed to be this huge new workforce solution. It’s so scary to see him present this as if it’s working and not just a locally hosted mockup at best. Sales and Marketing do not see this as just a design presentation instead seeing something they can actively sell. I don’t know when this will backfire on us, when I brought it up to our VP he said “well maybe he’s selling the idea and a few years down the road after the contracts are signed they will build it”. I think this is just how corporate America is going to work now.
sph 5mo ago
Shades of that famous quote by Kurt von Hammerstein-Equord. What is terrifying is that those that claim 100x increased productivity are those with the least wisdom to tell good code from bad. They are the ones inundating the world with utter slop.

rco8786 5mo ago

> AI is only as good as the person using it Nailed it. That said, it couldn't have possibly been that bad if you got it to "acceptable" in 2 weeks.

himata4113 5mo ago
I used AI too and acceptable means "you can play it without losing your progress".

bunderbunder 5mo ago

> AI is only as good as the person using it I think it might be even worse than that. It seems to be a multiplier for the Dunning-Kruger effect. Possibly because being trained to exhibit positive demeanor means that it will always tell you you're the best, no matter what.

himata4113 5mo ago
Not sure why I saw this getting downvoted. This is exactly what is happening to that person, they think they are way more capable then they really are that's why they're taking on hard challenges and high complexity tasks when they're not ready for it.
- bunderbunder 5mo ago
  And I do think it's important to be aware of this. When I've got a good idea, the agent says, "Good idea!" When I've got a bad idea, the coding agent also says, "Good idea!" And I'm most inclined to just go with what when I'm outside my realm of expertise, because that's the time when the coding agent promises to save me the most time. Because developing a proper understanding of things like the problem space and tech stack is often the most time-consuming part of the job.

hmokiguess 5mo ago

What is an instancing system?

duckmysick 5mo ago
To better scale the game infrastructure, instead of letting every player on the same big server, they are put in different smaller servers (instances). Players can interact with other players on the same instance (see them, do quests together, trade, etc.). Sort of like AWS instances. Normally instances are random within the same region, but usually there's a system in place so you can join the same one as your friend.
- hmokiguess 5mo ago
  Ah gotcha, I never heard that called that way. I'm familiar with matchmaking systems and VM provisioning, but never heard the term "instancing" haha. Thanks for answering me!

amarant 5mo ago

You've hit the nail on the head as we say where I'm from. This is also why I think the "boycott AI" movement is misguided. AI doesn't produce slop: unskilled AI operators do. Heck just the other day I saw a headline about a Nobel literature laureate apparently using AI, with some "expert" confidently claiming the winning novel was 100% generated. AI output quality ranges from slop to Nobel price worthy, depending on who uses it. Which seems to support the notion that it's a tool, much like any other.

epolanski 5mo ago

This reminds me of antirez saying that he made extensive use of AI to implement arrays in Redis. He spent 3 weeks (which for him likely means 60+ hours a week) only iterating on the design with different models and not writing a single line of code.

mizzao 5mo ago

2 years ago AI coding agents had barely emerged yet, right? Couldn't have gotten good results even if a good engineer was forced to use it...

himata4113 5mo ago
People in the know already had coding agents, they just needed a lot more back and forward as they weren't trained to be "agentic" and they were mostly cli based... then claude code was born because anthropic saw the potencial. The person in question used warp.dev which happened to bundle AI into so they stumbled into it more than anything.