GPT-5.6 Sol with High thinking and Fast mode lets me crush work without killing my weekly limit. Fable 5 is gone in half a day. Sol lasts a week, and Sam keeps giving me frequent resets.
I use Fable 5 for brutal engineering. Codex gets my planning and almost everything else.
Codex/ChatGPT desktop has replaced Claude Code as my daily driver. The AI coding crowd is grading them on the wrong test when it focuses on terminal performance alone.
Browser use, computer use, voice mode, and the mobile app make Codex better for how I actually work.
I manage Codex sessions from my phone and generate insanely good images in the app without another MCP, separate tool, or context switch.
That workflow makes me 10x more productive. A benchmark win means nothing when the product adds friction everywhere outside the terminal.
I used to think I needed years of coding before I could make anything move.
My process now is blunt: describe the next piece, build it, test it, explain what broke and try again.
What should I tackle: dump runs, couch loading, truck upgrades, employees or a house cleanout?
I knew fuck-all about making video games.
Now Iβm driving a junk-removal pickup through a town, towing a trailer I made in Blender and blasting songs I made in Suno. The whole thing runs in Unity, and I built it with AI.
Junk Empire: Landfill Legends. π§΅
Git history says the first commit, HUD, saves, jobs, customer route, towable trailer, traffic, Bobβs Hardware and Android build menu all landed on August 22.
I still had to direct it, test it and fix the broken parts. AI got my first game to a playable build in one day.