Scott Forsyth retweeted
You can't guess a recipe just by looking. Snap any dish and get the ingredients, nutrition facts and every step. Don't wing it. Snap it. whatdish.link/WD2HRKP6
One of my biggest breakthroughs came when I was on vacation. More on that below.
Building your own AI software factory, 10 of 21 🧵
🤖 Made with AI
It's really fun (and productive) to come back from a busy day, a night's sleep, or a vacation to tickets that are fixed and replied to as well as I would have done it, or better.
I've done it a few times since. The more we can trust our systems, the more we can step back.
🤖 Made with AI
Everyone interviews their fridge, right?
We asked the fridge. It had thoughts. 🧊 Snap your fridge and WhatDish shows you what you can cook tonight with what's already there. whatdish.link/DETMPHU8
Scott Forsyth retweeted
How far would you go for a restaurant's secret recipe? Disguise, bribe, sneak into the kitchen… or just snap it with WhatDish. Which plan would you try? whatdish.link/MAJUXZQU
First teaser video for WhatDish.
Early user response is great. People are blown away by how well we can generate a recipe from an image.
Or how it will suggest recipes from a photo of your fridge or pantry.
whatdish.link/PT5CR9C7
WhatDish is officially live! The issue with the Apple submission was my own, but it's been resolved.
It's searchable in the App Store and Play Store by 'WhatDish'.
I have some fun videos to share tomorrow.
Well said. It's better, in the short and long term, to have a long-running ticket that owns the plan and potential iterations.
Just mentioning it is all that's needed. Or, possibly just changing the status in the ticketing system and having it picked up automatically.
Replying to @housecor
I should simply be able to prompt “implement ticket 123”.
If I can’t, the ticket isn’t detailed enough. So I should fix that first.
The biggest jump in my confidence came when my agents got a place to prove their work, not just write it.
Building your own AI software factory, 9 of 21 🧵
It's not perfect yet. Some runs fail, and you still need to double-check the work. There's still a lot of human involvement (more on that in another post). But it's getting steadily better, and even very complex work can come back close to perfect on the first pass.





