I do think that if someone built a new OS today, it would probably be on mobile or server platforms. Server is more likely for the fluid compute scenarios. Maybe I try it next weekend... 🙈

There's tons of complexity - hardware interfaces, memory management, scheduler interactions. Even with perfect AI, you'd need deep systems knowledge to ask the right questions...

It's still got some ways to go, but definitely better than 2 years ago. Those prompting away still need to understand how & why certain things should done in certain way (eg: for security). We still don't solve the main problem though - understanding what the problem we're solving ;)

Would love to hear from players: would you be comfortable with a lightweight "proof of human" step if it meant better rewards?

#GameDev #Web3Gaming #DigitalIdentity

The tech has definitely moved on a lot from when I last looked at this in depth. It's not perfect, but better than what we have now? The biggest hurdle is adoption of the orbs. It only recently opened in London.

Unless needed for regs, normal KYC is overkill for gaming. But is there a middle ground? Worldcoin's approach could let players verify they're human once, then use that credential across games without sharing personal data. Win-win?

For example, I’ve been playing a lot of V Rising lately. It has great player progression and feel throughout. Unlike Valheim, there wasn't endless mining runs just to get gear. Boss fights feel right. VFX and audio on point. Underrated game.

#VRising #GameDev #IndieGame

What's been your experience with AI coding? Are you still throwing entire features at it, or have you found a better approach? Drop your thoughts below - I'd love to hear what's working for you.

#AI #DevTools #SoftwareEngineering

As with normal software development, most of the work is in analysis and setup, rather than coding. If you provide vague requirements, you will receive a vague program.

Test and iterate: Take each output, ensure it works, and then request the next piece. "Now add the storage module with these specific requirements". Build on solid foundations. Use git so you can revert as needed.

Start with project layout: I always begin with structure. Set up folders & dependencies, and the basic skeleton. Then move to individual features one by one. It's less for the AI to do later, saving you time and money.

Generate specific, detailed prompts: Instead of saying "build me a CLI app," I use AI to help me create the actual prompt I'll use. I'll ask: "Help me write a detailed prompt for building a Rust CLI application with these features..." Example:

Development guidelines: We create a document that defines the best practises for our development language(s), security considerations we need to take into account and other things like performance optimisation. Example:

Act as an engineer: Once we have all our feature specifications, we can ask AI to help us define two things. Our development guidelines & implementation instructions.

Use project memory: Some tools like Claude have a concept of Knowledge/Memory that persists across different chat windows. Save your feature specifications in that area, ready for the next step.

Act as a product owner: Use AI to help define your features into a specification that you would normally hand to an engineer. AI can help you, too, if you need it.

What I found was a powerful, but misunderstood tool. You can easily waste hours having it generate something wrong, or sometimes incomplete, or not work at all. Here's my lessons from the past few months:

Interesting! Related, have you played with DevContainers yet?