Before AI, indie game dev was about spending 5 years to make a cute little game. Most of the time the game earned $300, but once or twice it made a billion.
Now indies can use AI to go toe to toe with AAA. We'll start seeing those projects come out in 2027.
Why do people think everyone should be using Claude?
I don't & I am extremely productive like building a Vulkan C++ & Metal game engine in months from scratch without any chatbots coding my work seems like an impressive feat considering I didn't know Vulkan/Metal before.
Claude Code really spent $102 and 28 minutes building a steam train that looks like a Minecraft furnace on wheels π
Meanwhile GPT-6 Astra built an entire cinematic railway scene for $40 in 11 minutes
Same prompt, 4 AI workflows
π₯ Blender MCP β $67 / 20 min
π₯ GPT-6 Astra β $40 / 11 min
π₯ Kimi K3.1 β $12 / 3 min
π Claude 3.5 Sonnet β $102 / 28 min
Claude was 8.5x more expensive and 9.3x slower than Kimi in this comparison
But hereβs the plot twist
Blender MCP actually delivered the most cinematic result, with thick steam, realistic metal textures and dramatic lighting
Meanwhile Claude Code apparently decided that wheels and a chimney were enough to qualify as a train
imagine paying $102 for a steam locomotive and getting a gray refrigerator on railway tracks a year ago β οΈ
You don't need to ban recursive self improvement (RSI) because it's non-existent.
Artificial intelligence (AI) left to build itself would simply go in model collapse due to its statistical nature.
RSI is not possible with statistical models.
Reality already banned RSI.
The compiler didn't even follow the ODR or const qualifications. In other words they set out to make a C compiler and made something tha can make a runnable binary out of the Linux kernel instead.
Anthropic spent $20,000 to find out if 16 Claude agents could build a C compiler from scratch, and the result challenges how we think about AI coding
Researcher Nicholas Carlini ran nearly 2,000 Claude Code sessions, producing roughly 100,000 lines of Rust.
The compiler could build Linux 6.9 for x86, ARM and RISC-V, although it remained an experimental system with significant limitations.
The real breakthrough wasn't writing more code.
It was coordinating agents across a shared codebase without losing control of the project.
For developers, the lesson is clear: scaling AI coding requires task isolation, reliable tests and disciplined integration, not just smarter models.
I have written 720 pages of the Game Engine Black Book: Quake. I am almost done with Quake. Next I want to cover WinQuake, GLQuake, VQuake, and of course QuakeWorld. KDP limit is 840p. I may have to split it in two tomes if I go over the limit.
MATTHEW MERCER IS GANONDORF?!
.....
WHAT THE HELL IS THE MATTER WITH ALL OF YOU?!?!
ARE YOU ALL HIGH OR HAVE YOU GONE AND LOST YOUR LAST 3 BRAIN CELLS?!?!
THIS IS PERFECT GOSH DARN CASTING!!!
WHY IS THERE STUPID HATE FOR NOW?!
the sad thing is that Sword of Hearth and Home isn't even that powerful (the MTG card that art came from) anymore. In fact the swords kinda scaled poorly in general.
Remember when artifacts all had a massive drawback or multiple drawbacks that offset their insane power-levels?
I do.
I often will drop one big artifact into a campaign and the players will be afraid to touch the damn thing.
Heck, they had to do a massive quest for a noble efreet last campaign because the ranger was the epicenter of a natural disaster every single month. Hurricanes, earthquakes, floods, volcanic eruptions⦠all centered on him! They wished the artifact away!
the Vibe coder curve just ends at 9. the Vibe code lacks the skills to debug any thing. (heck i've been programming in C++ for nearly 10 years now and I lack the ability to debug code I write by hand sometimes)