Claude Fable 5.2: How I’d Make Money With It Using 12 Prompts

The most interesting thing about Fable 5.2 isn’t getting a better answer to a question. It’s handing it an entire job.

Audit a business. Research a hundred niches. Rebuild the spreadsheet somebody has been running their company on for 12 years. Design a small hardware product, right down to the firmware and enclosure.

That’s what gets me excited. When you can give an AI a substantial assignment and let it keep working, the opportunity shifts from saving a few minutes to selling a finished result.

I’ve been experimenting across four areas: making money, productivity, content, and hardware. Here are the 12 prompts I’d start with, what I’d use them for, and what happened when I tried several of them.

Table of Contents

• Making Money: Sell the Result, Not the Hours
• Productivity: Better Context, Better Delegation, Better Pushback
• Video and Content: Build Distribution, Not Just More Assets
• Hardware: A One-Person Company Starts to Look More Plausible
• The Bigger Opportunity Is Taking Responsibility for an Outcome

Making Money: Sell the Result, Not the Hours

The big shift is from asking for advice to assigning a deliverable. Fable 5.2’s ability to work through long tasks, potentially over 10 to 12 hours, makes it worth trying jobs you might otherwise hand to a consultant for a week.

That doesn’t mean every assignment will run unattended or come back ready to ship. It means you can think bigger about what you delegate. Define the result, ask for evidence, and make the review process part of the job.

1. Audit Your Entire Business

Most people ask AI to review their homepage. I’d give it the whole business.

Review my website, pricing, onboarding, customer emails, and five leading competitors. Continue until the audit is complete, and maintain a running log of your work. Prioritize the 10 improvements most likely to increase revenue, with supporting evidence for each recommendation.

I ran this on LCA, my design and design engineering agency. What impressed me wasn’t just the volume of feedback. It was how direct the recommendations were and how clearly the output ranked them.

I’ve tried versions of this across businesses we’ve been incubating, and the feedback has consistently felt to the point. That’s useful because a giant pile of suggestions isn’t a strategy. A ranked set of changes, backed by evidence, gives you somewhere to start.

The running log matters, too. It gives you a way to inspect the work instead of accepting a polished conclusion without understanding how it got there.

2. Complete a Freelance Job End to End

Here’s the business model I think a lot of people will build around: find a repeatable freelance job, create a reliable AI-assisted workflow, and sell the outcome at a fixed price.

Use this job posting as the brief and complete the deliverables as a freelancer would. Identify which parts you’re confident about, which need human review, and what fixed price you would suggest charging for the finished service.

I didn’t have a posting ready, so I added another instruction: find a suitable job you think you can complete that could make me money.

It selected an Upwork posting for an equity analyst report on TDMX, listed at a $450 fixed price, and began producing the research. The report was still in progress when I checked it, so this wasn’t proof of a completed client engagement or earned revenue. But it made the opportunity concrete.

Research and creative work are obvious places to explore this. The important part is building a good loop around the model: clear inputs, a defined output, and a human check where judgment or accuracy matters.

Don’t build the offer around how long the AI takes. Build it around the job the customer needs done.

3. Research 100 Niches at Once

Better subagent coordination opens up a different kind of research workflow. Instead of investigating one market at a time, you can ask for parallel research across many markets and then compare the findings.

Assign a separate subagent to each of these 100 niches. For every niche, identify the five biggest customer complaints, the solutions people currently pay for, and how difficult those customers are to reach. Consolidate the findings into a ranked comparison and recommend the five strongest niches to build for.

The three questions are doing a lot of work:

• What hurts? Complaints reveal problems people already recognize.
• What do they buy? Existing spending helps distinguish an interesting problem from a commercial opportunity.
• Can you reach them? A market is less attractive if you have no practical way to find its customers.

I started this research during my experiments, but it was still running when I checked back. I wouldn’t pretend the full hundred-niche sweep had already been validated.

The exciting part is the workflow: parallel investigation followed by a single synthesis. You’re not just collecting more information. You’re trying to make a better decision about where to build.

4. Rebuild Ancient Software

There is so much old software and technical debt sitting inside ordinary businesses. Sometimes the “software” is a spreadsheet that one person understands and everybody else is afraid to touch.

That’s a service opportunity.

This spreadsheet has been used to operate my client’s business for 12 years. Determine how it works and what business processes it supports. Recreate those workflows in a simple web app, migrate the existing data, and produce a one-page guide for the staff.

Notice what the assignment includes. It isn’t just “make an app.” It includes understanding the existing system, replacing its useful functions, moving the data, and helping the team use the replacement.

Historically, somebody might have billed $50 or $100 an hour to do this. The opportunity now is to package the complete migration as an output.

The code is only one part of the value. The customer wants their business to keep working without depending on an ancient spreadsheet.

5. Check a Deal Before Buying a Boring Business

One of the biggest obstacles to AI in finance and legal work has been confident guessing. A model that can acknowledge uncertainty is more useful than one that invents an answer to every question.

That’s the behavior I want to build into a deal-checking workflow.

Review this seller’s profit and loss statement alongside the business listing. Identify inconsistencies and anything that does not reconcile. Label each financial figure as verified, unverified, or suspicious, and draft the 15 questions I should ask before making an offer.

The labels are the point. “Verified” and “sounds plausible” are not the same thing.

I’m excited about a model being more willing to say it doesn’t know, but that is not a reason to stop checking its work. This is a way to organize diligence and surface questions, not a substitute for professional financial or legal judgment.

Use it to make the next conversation sharper. Where is the evidence missing? Which numbers conflict? What should you ask the seller to explain?

Productivity: Better Context, Better Delegation, Better Pushback

Productivity isn’t only about doing tasks faster. It’s also about getting output that reflects how you think, operating the tools you already use, and finding weaknesses in your plans before they become expensive.

6. Use the “Grill Me” Prompt

This is the one I think almost everyone should try.

Ask me 20 questions about how I make business decisions, presenting one question at a time. Use my answers to create a reusable skill file that captures my decision-making preferences and can guide your future work.

Most people give AI almost no context about their judgment, then wonder why the answers feel generic.

An interview reverses that. The model has to learn how you think before it starts recommending what you should do. Asking one question at a time also gives you room to answer properly rather than racing through a questionnaire.

The useful deliverable is the skill file. Keep it available in the setup where you want those preferences applied, rather than assuming every future conversation will automatically have that context.

7. Operate Apps That Never Had an API

A lot of businesses still depend on old applications that don’t offer a convenient integration. Stronger computer operation makes those systems more interesting targets for automation.

Open our existing invoicing application and export the previous month’s invoices. Reconcile them against the bank statement and identify outstanding payments. Prepare the findings, but get my approval before sending any messages.

That’s an entire workflow across an old app, an export, and a bank statement. It’s much more useful than asking for general advice about accounts receivable.

The final instruction is essential: ask before sending anything. Let the agent do the preparation while keeping approval over external actions with you.

In my experiments, computer operation felt noticeably stronger. That expands what you can delegate, especially when the software itself hasn’t changed in years.

8. Red Team Your Plan

Sometimes the most valuable thing AI can do is disagree with you.

Review my plan for the coming quarter from the perspective of an investor looking for reasons to decline. Make the strongest case against it, then identify the single piece of evidence or change that would reverse your decision.

I gave it LCA’s vision deck, including some of my goals, and asked what it would change. I was genuinely surprised by the quality of the feedback.

I’d also asked a well-known advisor for input. In some respects, the model’s feedback was even better. That’s my experience with that specific exercise, not a claim that AI replaces good advisors.

What makes the prompt work is the second half. Criticism alone is easy. Asking what would change its mind pushes the response toward a condition you can actually test or address.

Video and Content: Build Distribution, Not Just More Assets

I’ve been talking a lot about marketing engineering: using AI agents to do meaningful marketing work, not just generate a few captions.

Video is a particularly interesting example. I’ve seen friends produce short documentary-style pieces and animated explainers from very small amounts of prompting. That gets my brain going immediately.

If you can produce useful videos for a few dollars each, potentially in the $1 to $5 range depending on the workflow, what niche distribution could you build?

I think about it a little like real estate, but with far less capital required to get started. A useful library of content can attract attention over time. That attention can support advertising revenue and give you an audience for future products.

The opportunity isn’t simply “make cheap videos.” It’s build a focused channel that serves a specific audience.

9. Turn Raw Footage Into a Short-Form Edit

Take this four-minute recording and remove pauses, mistakes, and repeated takes. Select the strongest 45-second segment, add captions and restrained motion graphics, and export a vertical version. Provide three alternative hooks for the opening.

This is a practical, recurring job. You have the source material, you know the output format, and you can compare a few hooks rather than accepting the first opening it generates.

I’m using this kind of workflow myself. The goal isn’t decoration for its own sake. It’s finding the strongest idea and presenting it clearly.

10. Create Niche Explainer Videos

Produce a 60-second animated explanation of this topic for this specific audience. Develop the script, create the animation, encode and render the finished video, and generate five versions with different opening lines.

The niche matters as much as the animation. An explainer made for everybody is a much weaker distribution strategy than a useful explanation made for a clearly defined audience.

The multiple openings give you options without requiring five completely different productions. Keep the core explanation, then test different ways of introducing the idea.

Depending on the environment, these workflows can involve tools such as Blender or DaVinci Resolve. The useful capability is not just describing an edit or animation, but operating the tools needed to produce it.

11. Turn One Episode Into Every Useful Format

Use this long-form episode as the source for a PDF playbook, five posts in my voice, a newsletter, and a free tool someone can use in two minutes. Flag any factual claim you cannot verify.

This is where one substantial piece of content becomes a whole set of useful assets.

A playbook should help someone apply the ideas. A newsletter should develop a clear argument. A small tool should help somebody get a result quickly. They don’t all need to be the same summary in different packaging.

And keep the verification instruction. Repurposing an unsupported claim into five formats doesn’t make it more reliable. It just spreads the problem.

Hardware: A One-Person Company Starts to Look More Plausible

This is the area that really fires me up.

We’re used to asking AI to build software. But what happens when the assignment is a physical device for one specific customer with one specific need?

12. Design a Device for One Niche

Design a compact device that solves one problem for one niche. Provide a parts list with current prices, a circuit design, firmware, and a printable enclosure. Estimate the cost of producing 100 units.

That prompt moves beyond an attractive product concept. It asks for the components that would let you begin evaluating whether the idea could become a real product:

• A priced bill of materials.
• The circuit design and component connections.
• Firmware to control the device.
• An enclosure suitable for 3D printing.
• A production-cost estimate for a small batch.

You could then explore the commercial side, including a Shopify storefront. But first, you need a product concept that holds together.

The Example: A Minimalist Phone With a Real Keyboard

I tried a device inspired by a BlackBerry-style phone, without the endless apps. Something for people who don’t want to spend their day doomscrolling.

The generated concept was called Plain One: a pocket 4G calls-and-texts handheld with a physical thumb keyboard and a monochrome display.

Its proposed feature set was deliberately narrow:

• Calls and text messages.
• Contacts and notes.
• A clock and alarm.
• A 45-key thumb keyboard.
• A 2.7-inch black-and-white screen readable in sunlight.
• No browser, social feed, or app store.

I loved the product logic. The screen and the lack of feeds weren’t compromises to apologize for. They were part of the anti-doomscrolling proposition.

The design document went into parts, power rails, pin mapping, modem connections, display connectors, and the keyboard matrix. It also produced a visual concept for what the device could look like.

That’s what I mean by vibe manufacturing: using natural language to get much further into a physical product idea than you could with a sketch and a vague brief.

But there’s an important boundary. A generated design document is not the same as a manufactured, tested, working phone. I wouldn’t trust every part price, connection, or firmware detail without checking it. AI can still make mistakes.

The exciting thing is how much of the initial design work you can explore quickly. Work that once required a substantial upfront budget can become a concrete starting point for prototyping and technical review.

The Bigger Opportunity Is Taking Responsibility for an Outcome

The common thread across these 12 prompts is not clever wording. It’s the size and clarity of the assignment.

Don’t just ask for business advice. Ask for a ranked audit with evidence. Don’t just ask for an app. Ask for a replacement workflow, migrated data, and a staff guide. Don’t just ask for a product idea. Ask for the parts, circuit, firmware, enclosure, and small-batch economics.

Then keep the boundaries clear: what has been verified, what remains uncertain, what needs a human check, and what requires your approval before action.

That’s how I’d approach making money with Fable 5.2. Find a job people already care about, define what “finished” means, build a repeatable workflow, and sell the result.

There are hundreds more use cases to discover. That’s what makes this such an exciting time to build.