Ex-Apple engineer throws 2 real jobs at Grok 4.6 within Cursor…
And he goes to sleep. The next day he wakes up and opens the live results. Zero surveillance. Zero mid-generation revisions. Only pure agent mode running for hours in work
Four different AIs were given the same task.
Draw a slice of lemon integrating it into the human pose. You may have come across delightful videos on social media many times where creativity is demonstrated with this type of food or various ingredients. Results are surprisingly similar for all 3 models
I've been experimenting with a creativity framework.
To test it i give it a base drawing and let it iterate through the framework. here's opus-4.6's attempt. it did much less than opus-5 or fable, and a bit less than chatgpt-5.6sol
If you use Fable in Claude Code, "let Fable do everything" is
too good to waste. While saving the limit, If you want to bring out the performance, this is the configuration 👇 ・Set the model to "Fable 5" ・Set Reasoning to “Max” ・Create a dynamic workflow The division of roles is simple. ・Fable → orchestrator who directs the whole thing…
You have reached the end of the archive
All of Claude Opus 5