You have 2 days left with Claude Fable. Here's 5 must-do use cases before it's gone.

Happy Saturday! This is a rare weekend edition, but I came back from my walk at 7am this morning and have spent the past 2 hours putting this guide together because I do not want you to miss the opportunity to use the best model we’ve had access to to date.

Claude Fable 5 leaves our plans on Sunday 12 July, so I wanted to give you 5 use cases worth doing with Fable 5, a guide to prompt it so it doesn’t hand you slop, and the why the model leaving isn’t actually the end of the world (despite what people are telling you). Enjoy, and feel free to share this with someone who needs to use Fable 5 before it disappears.

Share

There’s a very specific genre of content taking over my feed this week. A countdown timer. A thumbnail in all caps. Someone telling me the most powerful AI in the world retires on Sunday, and I’ve got 48 hours to build a trading bot or I’ll lose thousands. You’ve probably seen the same ones.

But don’t freak out just yet, because the panic is doing you absolutely zero favours. Claude Fable 5, the most capable model Anthropic has released, sits on your paid plan until around 12 July. After that it doesn’t vanish, it just moves to pay-per-use, roughly $10 per million words in and $50 out, about double what Opus costs and extremely unrealistic for most people. So for a few more days the best model in the world is included in a subscription you already pay for, and then it isn’t.

This guide is here to help you make the most of it, but as you’ll see, most of it will still apply long after Fable 5 is gone.

Fable is the world’s best AI model to date.

“Fable 5’s capabilities exceed those of any model we’ve ever made generally available. It is state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas. The longer and more complex the task, the larger Fable 5’s lead over our other models.”

Anthropic

Stripe let it at 50 million lines of code and it completed a migration in a day that a full team had scoped at two months. That’s the level we’re talking about.

But the part the countdown videos leave out is that this is the new reality of things, not a one-off event you’re about to miss. Just yesterday, OpenAI released 5.6 Sol, their answer to Fable 5. And it’s currently being rolled out to all paid plans.

Which changes what you’re here to do. Instead of just focusing on “grab as much Fable as you can before Sunday,” it’s probably well worth your time to learn how to get real work out of the new era of models, on the jobs that deserve them, and keeping what it makes so the next best model picks up where this one left off. If the past is any pattern, we might even get access restored to Fable 5 down the line.

If you’re reading this, you probably want to build something that’s actually useful, not just a video game or vibe-coded app (as cool as some of them are).

In my experience and research, this comes down the five main use cases. Each one has a prompt underneath to get you started, written to give Fable an outcome rather than a step list. This is intentional, as Fable works best on outcome-based prompts (as you’ll see below)

1. Let it build or improve your skills

The most leveraged thing you can do with the smartest model you’ll ever have is get it to build the assets your everyday model will reuse forever. Brain-dump what you want to build and why, point it at your strategy, and let it write you working v1 skill files you can refine later.

To ask Fable to build you new skills:

Build me working first versions of the skills I've been meaning to make, so
a lower-tier model can run these jobs the way I would.
Read my context and strategy files first. Here's my brain-dump of what I want
each skill to do and why it matters to me: [paste brain-dump].
Interview me to gather any information you need, then build a usable v1 of each one that I can improve later. At the end, explain what you made, flag the things you're least sure about and why, and give me recommended next steps.

To ask Fable to optimise your existing skills:

You are a retiring principal engineer leaving my skill library in the best possible state before you go. Two phases.
PHASE 1 — Audit and rank. List every skill. For each, score how much rewriting it would improve my actual output (High / Medium / Low), with one line of reasoning. Give me the ranked list before writing anything.
PHASE 2 — Rewrite the top 3 to 5 only. For each one you upgrade, write the complete new SKILL.md as a drop-in replacement, and give me a short changelog: what changed and why it produces better output. Encode judgment and worked examples, not just instructions. Preserve my voice rules exactly. Save each upgraded file into a folder with a clear v-next filename; do not overwrite the originals.
Finally, name at most 2 gaps where a net-new skill would meaningfully help, with a one-paragraph spec each. Do not build them yet unless one is trivially obvious. Don't pad the library for its own sake.

2. Let it audit and optimise your setup and how you use AI

Fable has exceptional reasoning abilities, making it the perfect model to assess your setup and notice any gaps or improvements. Have it read your context files, folder structure and skills (for cowork / code), or your project setup and skills (for chat), as well as your chat history, and ask for an audit. Here’s the prompt that Opus made for me to use with Fable:

Act as a senior systems auditor doing a whole-system review of my AI Operating System. I want one thing: a ranked findings report, highest-leverage first. Do NOT edit any file. Propose only.
For each finding give me: the file(s) involved, the problem in one line, what it costs me if left as-is, and a specific proposed fix.
Cover at minimum:
- Contradictions or drift between files (where two files disagree, or a file contradicts the locked decisions in context/strategy.md)
- Opportunities (areas to strengthen setup to improve results, or add new areas or use cases that are currently not in there)
- Duplication (the same fact living in more than one place, which my canon rule forbids)
- Gaps (context a cheaper model would need for good output but can't find anywhere)
- Over-engineering: anything more complex than it needs to be. This is my failure mode. Be aggressive here. Tell me what to delete or simplify.
- Friction: where the current structure makes a routine task slower than it should be
- Staleness: anything date-stamped and now out of date, or missing a date
Rank every finding by leverage (how much better my weekly output gets per unit of effort to fix it). Put a clear "fix these five first" section at the top. Keep the whole thing skimmable. Australian English, no em-dashes, no emojis.

If you’re using Chat, or don’t have a folder structure set up, just give Claude the prompt above and ask it to adapt it to your situation:

I want to use Fable to audit and optimise my AI system, including context, skills and my day-to-day use. Review the prompt below which is optimised for a full operating system and adapt it to my situation so I can run this audit for my level. Ask me any questions you need.
[Insert the prompt above].

3. Let it plan a strategic project for Opus and Sonnet to execute

Fable should not be doing basic execution or writing for you. Instead, use for the thinking and strategic planning, ask it to produce a detailed plan, then use Opus or Sonnet to follow that plan later.

You are a strategist and a prompt engineer. Your job has two stages: first interview me, then write me a single, ready-to-paste prompt for Claude Fable to work on a strategic project. That prompt must produce both a genuine strategic recommendation AND an execution plan that a cheaper, everyday model can carry out later without re-doing the thinking.
STAGE 1 — INTERVIEW ME.
Before you write anything, ask me a focused set of questions (no more than eight) to understand:
- The decision or project. What am I actually trying to figure out or achieve?
- My situation. What do I do, what's the business or context, and what stage am I at?
- My constraints. Time per week, money, team or solo, hard boundaries or things that are off the table.
- My resources and assets. What have I already got going for me (audience, skills, cash, existing offers, whatever's relevant)?
- What a good outcome looks like. How will I know this worked? Any target numbers or dates?
- History. What have I already tried, decided, or ruled out, and why?
- My instinct. What am I currently leaning towards, and what's my gut worry?
- Anything you sense is missing.
Ask, then stop and wait for my answers. Do not write the prompt yet. If my answers are thin or unclear, ask one short round of follow-ups before proceeding. Do not ask me things for which you already have sufficient context.
STAGE 2 — WRITE THE PROMPT.
Once I've answered, write one self-contained prompt I can paste straight into Fable. Do not solve my problem yourself; your output is the prompt. The prompt you write must instruct that model to:
1. Act as a sharp strategist and honest thinking partner, not a yes-machine. Reason from first principles. Challenge weak assumptions. Give options with tradeoffs where the call is mine, and a clear recommendation even if it contradicts my instinct. Name the risks and the assumptions the recommendation depends on.
2. Deliver the strategic thinking as a durable decision document
3. Then produce an EXECUTION PLAN built so a cheaper, everyday model can run it later with no further strategic reasoning required.
Embed all the context I gave you directly into the prompt so it stands alone. Write it in plain, copy-paste-ready text with no special formatting, so I can paste it anywhere. Match my tone and language conventions if I gave you any. After the prompt, add how to use it.

4. Let it rebuild an app you currently pay for

You don’t need to clone every feature of a tool, just the main things you actually use it for. Fable is strong enough to get a working version done end to end.

Open up a chat with Opus and use this prompt:

Research how [tool/idea] works: core features, architecture, what a minimal version needs. Write PRD.md for a version that [your constraints]. Include: goals, non-goals, stack, milestones, and success criteria I can verify myself.

Then, get open Fable in Claude Code and use this prompt:

Goal: Build the app described in PRD.md.
Success: [3 verifiable criteria].
Work autonomously. After each milestone, update STATE.md with
what passed, what failed, and any rule worth remembering.

I used this to create a self-hosted version of WhisprFlow. Yeah, it’s really that good.

5. Let it find your own use case

Your Claude knows you better than I ever could. To save your Fable usage, open a chat with Claude Opus in a space that has your context, and let it do the research and the thinking about what you should build. THis is the exact prompt I used:

Help me find the highest-value thing to build with Claude Fable before I lose easy access to it.
First, read my context. Then, research what people are building with Fable right now that's giving real value. Suggest 10 use cases that would matter most for someone in my exact situation, ranked by impact: what saves me the most time or unlocks the most value.
Ask me any questions you need. At the end, offer to generate prompts for any of these, and if I say yes, make the prompts optimised for Fable 5 specifically

You should not be using Fable in the same way that you use Opus or Sonnet. One, because Fable is way more expensive and used wrong you’ll burn through your usage limits before you’ve even done anything useful. And two, because it thinks differently.

The biggest piece of advice I can give you is to give it the outcome, not the steps. That’s what all the prompts above are optimised for.

With a cheaper model you spell out every step because it needs the guidance. Fable follows instructions so closely that a step-by-step list turns into a cage, it’ll do exactly what you told it even when what you told it was slightly wrong. So instead, describe what a good result looks like and what you’re trying to achieve, then leave it room to get there. If you’ve spent a year learning to write more detailed prompts, this feels backwards, but do it anyway.

Three smaller habits sit under that.

First, let it run, because it can work on a single goal for hours. Hand over the whole job, go and work on something else, then come back when it needs you.

Second, make it prove its work, because on long tasks a powerful model will sometimes tell you it did something it didn’t quite do. It isn’t lying, it’s over-reporting, and the fix is one line at the end: “show me the evidence, don’t just tell me it’s done.”

And third, use it on either medium or high effort. You do not need to use Extra-High or Max, and if you’re on the pro plan, it will consume way too much of your rate limits in a very short space of time.

1. 10 new skills. I brain dumped what I was thinking of building and the goals for them, told table to read my strategy and review, and then build out v1 of the skills in a format that i could then improve on later.

2. A custom event software. We run in-person events about once a quarter at work, and i wanted to find a way to increase the value of the event long past just the 2-3 days themselves. I created a PRD with Opus, then gave that to Fable to build out. Again, it’s not perfect, but the most important thing was getting v1 done with Fable - I can refine with Opus

3. The system audit described above

4. Building a local, custom WhisprFlow tool. Still a work in progress.

For some more inspiration, here are some creators who have posted a lot about this:

@jens.heitmann
@mr.paidsocial
@charliehills
Harper Carroll AI (Instagram reel here)

• Pick the use case that stands out to you

• Grab the prompt as a starting point

• Customise it to your goals (or ask Claude to help you do so)

• Give it to Fable and let it run

New here? Every week I share anti-hype guides to build your business and personal brand with AI. Join the newsletter below, or become a paid member for direct responses to your challenges.