Still building apps, playing songs etc. apps: https://teapaw.org/t.co/QMKzQaRyiW
Norwich, UK
Joined August 2007
- Tweets5.9K
- Following2.2K
- Followers1.1K
- Likes4K
The weird thing about making games with AI is it becomes a fun kind of meta game where you play the game once removed and the actually fun part is inventing the game to be played.
It does not mean the game itself will ever be fun to play.
You sit. You plan for three hours. You make decision after decision after decision. You change your mind. You change it back. You spike something out then realise it doesn't work. You do some extra research on something. You draw a diagram. You make some changes and do it all again. Loop. After a while you land on a direction you like.
Finally, you say, "Go."
An army springs to life. You get up to put the kettle on.
Turns out you can just build a DAW, a bunch of instruments and effects, then start generating shitty "jungle" music.
In a web browser.
I know, I know, I'm sorry...
Claude can play the drums. Check this drum track with an epic solo inspired by John Bonham.
It's ridiculous to build a DAW in the browser. It's pretty fun though.
By now we all know you can build a passable DAW with some brainslop into your phone from the Morrison's carpark.
It's still wild when it happens, and includes a demo track called "First Light".
I saw the video of the guy comparing Gemini 2.5 Pagoda 3D modelling to Gemini 4 using Blender modelling.
Meanwhile in the software factory...
You're going to be able to get back in the flow state while AI writes the code at the speed you can speak requirements.
Think of what we're going to do with that ability...
The future is going to be nuts. Just gotta survive til then somehow.
I used gpt-oss-120b on Groq at some stupid token rate a while ago and it was pretty clear then that once inference was fast with good enough models all the issues we have with context switching would disappear.
That moment is getting really close.
Just need it to be affordable.
Made a cuppa for the 63 year old tree surgeon doing some work in my street today.
He asked what I did for work. I told him I was a software developer.
He asked how I was coping with all this AI stuff.
I said I wished I was a tree surgeon.
I want local models too, but I also like running things in parallel.
You can get a beefy (expensive) machine to run one big open weight model.
But even if it runs at a reasonable tokens per second speed you're going to be single threaded.
At least you won't be complaining about context switching.
Are people still thinking they're going to have endless work cleaning up AI slop projects or has the penny dropped now?
I'd love to see the chart of usage going through the roof when @thsottiaux confirms there'll be a reset imminently.
Had a weekend off for once, so I haven't even tried Astra yet.
Time to see what all the fuss is about...
