How to Work With Fable

Delegation patterns for long tasks: give it the whole problem, set checkpoints, and know what to verify.

Working with Fable 5 well means changing one habit: stop pre-chewing the work. With earlier models, people broke big projects into tiny prompts. With Fable 5, that habit throws away its biggest strength.

The better pattern: give it the whole messy problem. The folder of old drafts, the contradictory feedback emails, the half-finished outline, the goal you are trying to hit. Fable 5 is built for long-duration autonomous work across millions of tokens. The mess is the input.

A good long-task brief has three parts:

Everything you have, including the embarrassing half-done parts.

What done looks like: the finished output, who it is for, how they will use it.

How you want to be involved: where it should pause and show you its work.

Set checkpoints. When Claude works autonomously for a long stretch, your first look at the work should not be the final product. Build pauses into the brief: after the plan, after the first draft, before the final version. At each checkpoint you confirm direction or correct course. Checkpoints are how you stay the decision-maker while Claude does the labor.

Verify at the end. Check three things:

Coverage. Did it use all the inputs you gave it?

Decisions. Ask Claude to list the significant judgment calls it made so you can review them.

Facts. Spot-check the names, numbers, and claims that would embarrass you if wrong.

You delegate the work. You never delegate the accountability.

The Fable 5 delegation pattern: hand over the whole messy problem instead of tiny pre-chewed prompts, define what done looks like, and set checkpoints (after the plan, after the first draft, before the final version). When long work finishes, verify coverage, review the judgment calls it made, and spot-check the facts that matter. Next: how safety routing, access, and pricing work.