Vanessa Chang·@thinkwithv

a beginner's guide

Prompting Fable

Fable is Anthropic's smartest model, and it needs a different skill than the one you've been practicing. A lot of old prompting advice is backwards now — and the fix is mostly deletion.

Stop mapping the route. Name the destination.

tap the button — watch a prompt get shorter →

the same prompt, before and after

You are a world-class senior analyst with 20 years of experience.

Think step by step and show all your reasoning.

Step 1: pull the churn data. Step 2: list anomalies. Step 3: rank by severity. Step 4: summarize.

Format your answer as exactly: Summary (3 bullets), then a table, then a conclusion.

Look at last quarter's churn and tell me what's actually going on.

Every line above except one is doing the model's job for it.

Why everything flipped

Older models needed more instruction: step-by-step lists, output checklists, role-play, and tricks like “think step by step.” All of that filled in what the model couldn't work out on its own.

Fable can work it out on its own. Anthropic's guide says prompts built for older models now say too much — the extra help gets in the way. One clear sentence can replace a list of ten rules. Your job is simpler now: say what you want, why it matters, how you'll know it's good, and what's off-limits. Fable fills in the steps.

The ten tips

tap any to expand — the navy lines are ready to copy and paste

1

Bring your hardest problem, not your easiest

Easy tasks waste it. Give it the hard stuff.

Fable gets more useful as the task gets harder. People who only tried it on easy tasks came away thinking it was worse than it is. Give it the thing you've been putting off because it felt too big.

2

Say why, not just what

It does better when it knows what you're trying to do.

When it knows the goal, it makes smarter calls. Anthropic suggests this pattern:

I'm working on [larger task] for [who it's for]. They need [what the output enables]. With that in mind: [request]

3

Describe the destination, not the route

Say what a good result looks like. Skip the step-by-step.

The step-by-step you'd write for a weaker model makes Fable's work worse. It finds a better path than you'd map out.

4

Define “done”

Tell it how you'll judge the result.

One sentence about what “good” means beats three paragraphs of formatting rules:

Good enough that I could hand it to my accountant without explaining anything.

5

Say whether you want a check or a fix

Smart models often act when you only wanted their opinion.

If you're just thinking out loud, say so. One line sets the line:

The deliverable is your assessment. Report findings and stop. Don't fix anything until I ask.

6

Let it plan the work first

Big, messy request? Let it sort out the plan.

For big tasks, ask for the plan first:

Plan this first, ask me any questions, then do it.

7

Demand receipts

Make it show proof, not guesses.

In Anthropic's tests, this line almost wiped out made-up progress reports on long jobs:

Before reporting progress, audit each claim against actual evidence. Only report work you can point to evidence for; if something isn't verified, say so.

8

Build in a fresh-eyes check

Make it grade its own work like a tough stranger.

Have it re-check its work before it calls the job done:

When you think you're done, verify your work against the success criteria as if you were a skeptical reviewer seeing it for the first time.

9

Ask for its pick, not every option

Don't make it list choices it won't even pick.

Left alone, Fable lays out every option. Point it at an answer:

If you're weighing a choice, give a recommendation, not an exhaustive survey.

10

Don't ask it to show its hidden reasoning

“Think step by step” is now a trap.

Asking to see its private thinking can trip a safety rule. When that happens, a backup model (Opus 4.8) answers instead. Ask for the answer and the proof behind it, not a play-by-play of how it got there.

What didn't change

Being clear, giving context, and showing examples still win. The one thing that flipped is the turn-by-turn instructions weaker models needed. The basics didn't change.

  • Long waits are normal. Hard tasks can take several minutes per turn. The model is working, not stuck.
  • A “responded by Opus 4.8” notice means you hit a protected area. The system is just being careful — you didn't do anything wrong.

take it with you

Get the cheat sheet + the skill

One page: the inversion at a glance, plus every copy-paste line — pin it next to your keyboard. Bundled with it: the Fable Prompt Builder skill — drop it into Claude, describe your task, and it assembles the Fable-ready prompt for you.