i'm going to teach you how to run Fable 5 on autopilot, using my own library of loops and goals... 25 workflows, each with a prompt and the exact tool it plugs into
the method follows karpathy's recipe for agents: "here's an objective, here's a metric, here's your boundaries of what you can and cannot do. And go"
every workflow below is exactly that... an objective, boundaries, and the proof it's done
Fable 5 is the model built for this: it holds one job for hours without losing the plot, where other models wander off after minutes
remember these two Claude Code commands:
> /loop - a job that repeats on a schedule until it's done
> /goal - a finish line you write once, and the model works alone until it's crossed
one warning before you schedule anything: a model that never gets tired never stops on its own, and Fable is the most expensive model on the market
run it without a budget and a stop rule and the bill will find you... every workflow below carries both
the ready-to-run versions of these loops, and the training on turning them into income, live in the real time AI ops community: weeklyaiops.com
# what's a loop?
a loop is a job your agent repeats on a schedule: every morning, or every time something new lands
five parts, always:
> a schedule - when it wakes up
> ONE change per round - it fixes the single most important thing it found, never everything at once
> the same check every time - so this week's score can be compared against last week's
> a state file - a text file where it writes what it did and what's queued next
> a stop - a hard cap on rounds and a rule for what "done" and "blocked" look like
it's the same shape frontier researchers run their own experiments on: change one thing, test it, keep it only if it improved, write it down, repeat
the state file is the part almost everyone skips, and it's the part that makes every run smarter than the last
Generated by Thread Navigator
Press ⌘ + S to quick-export