Post

Indie devs are in a tough spot: don't use AI, and watch everyone else ship games x5 faster than you or use it and be damned. I don't think there are many coders left that are not using AI agents for coding right now, it's just that effective (been coding games for 28 years, professionally for 20)

absolute bullshit thing to say dude. it literally does not do that

Like, proof required for the "ship 5x faster than you" thing pls.

True, and it doesn't scale as much as people think it does but I'm definitely way, way faster on my own work, and we did tools that would have taken days in hours

oh you're actively feeding the beast huh! actively part of the problem!

Not using it for coding is a luxury, if you can run a business without it more power to you, we have a very finite amount of time/budget

My whole team doesn't use it. It's not that hard.

I'm glad for you, as I said, it's a luxury now. All my colleagues in AAA and mobile are using it extensively

I'm in AAA and I've avoided using it so far

The key is "so far", and belive me, I would 100% go back to a pre-AI world

What stops you from doing so? Can't be the "productivity gains", I have yet to see an example in production where its context window was actually big enough to get a correct answer AND get to that answer before a human does (or could have)

It became actually useful not too long ago, before it was at best ok to speed up boilerplate, opus/sonnet 4.6 can implement entire systems from tech design specifications and even do tools we never had time to do, using agents not a chat window

I would not find that useful. Most of my AAA work has not been implementing entire systems or tools as quickly as possible, it has been looking carefully at the complete context of existing code and situation and making small, safe tweaks that get the designer exactly what they want

I don't see why I'd bother with using a genAI to write the code on larger subsystems and features, either. Talking with designers to decide what code should be written, compile/cook times (including CI), and waiting for code reviews are far bigger bottlenecks than my typing speed IME

Why would I write a tech design specification just to hand to an AI when doing that saves me almost no cognitive load over just writing the code and testing it myself? The code is the easiest part of the whole process!

What are you implementing that takes 2 minutes to decide on and describe, but 2 hours to actually write? I can't think of any such situation I've encountered like... ever

Besides that, I would say that I spend maybe an hour of my 8-hour workday writing code at best. The rest is waiting for stuff to happen, reviewing code, talking to stakeholders, and debugging/testing. On some gnarly bugs it's taken days of careful testing and debugging... to change *one character.*