Quick answer
The most practical Claude Fable 5.2 business ideas are services where the model does a long, clearly defined job and you sell the finished result at a fixed price. Think business audits, research reports, ranking niches, rebuilding an old spreadsheet into a simple web app, or checking a seller’s accounts before a purchase. In a solo episode, Greg Isenberg shares 12 prompts across four areas: making money, productivity, content and hardware. They are good starting points, not proven income. Fable 5.2 is expensive, it still makes mistakes, and every output needs a human check before a client sees it.
Key takeaways
- Sell outputs, not hours. Greg’s big idea is packaging work Fable 5.2 can do into fixed-price “done for you” services.
- Long, unattended jobs are the headline feature. Greg says it can work for 10 to 12 hours on one task.
- Subagents let one prompt split a big research job, such as scoring 100 niches, into many parallel pieces.
- It is more willing to say “I don’t know”, which Greg sees as opening finance and legal-style checking work.
- The demos were early. Several were still running when the episode ended, and nothing shown was sold or independently tested.
- Start with your own business (an audit, a decision “skill file”, repurposing content) before selling anything to clients.
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- How to Use AI to Check a Small Business’s Accounts Before You Buy It — a careful first-pass check of a seller’s profit and loss figures, plus the questions to ask.
- Grill Me Prompt: Turn Your Business Decisions Into an AI Skill File — let the AI interview you, then save your answers as rules it follows every time.
In this guide
- Watch: the source video
- What makes Fable 5.2 useful for building a business?
- What are the 12 Claude Fable 5.2 business ideas?
- How do you turn a prompt into a fixed-price service?
- What did the live demos actually show?
- Can Fable 5.2 run a content or faceless channel business?
- Can one person really design hardware with it?
- Which ideas should a small business try first?
- What are the risks and limits?
- How to test one Fable 5.2 business idea this week
- Frequently asked questions
- The bottom line
- Related posts
We have already covered what Claude Fable 5.2 is and how it scores, and when it is worth paying for over cheaper Claude models. This article answers a different question: what could you actually build or sell with it?
It is based on a new episode from Greg Isenberg, host of The Startup Ideas Podcast, called “Claude Fable 5.2: How I’d Make Money With It”. Greg shares 12 prompts and runs several of them live. Below, we put them in plain English, separate what he showed from what he suggests, and add the checks you need before charging anyone.
Watch: the source video
What this video covers: Greg Isenberg walks through 12 Claude Fable 5.2 prompts for making money, productivity, video and content, and hardware, and runs several live, including an audit of his design agency, a freelance job the model found on Upwork, a 100-niche research sweep and a minimalist phone design.
Video: Greg Isenberg (@gregisenberg, The Startup Ideas Podcast). No paid sponsor; Greg mentions his own agency LCA (Late Checkout) and his Idea Browser site. This summary, the opinions and the practical advice are our own. Money ideas and results are Greg’s suggestions and demos, not independent tests or promises of income.
What makes Fable 5.2 useful for building a business?
Greg points to four strengths that make Fable 5.2 different from the AI tools most people have used so far. Here is each one, with how much weight to give it.
| Strength Greg describes | Why it matters for business | How solid is it? |
|---|---|---|
| Works for 10 to 12 hours on one job | You can hand over a task a consultant might take a week on | Greg’s description. Long runs also mean a big bill. |
| Runs many subagents well | One prompt can research dozens of options in parallel | Shown starting live, but the 100-niche run had not finished on camera. |
| Admits when it lacks an answer | Makes checking work, such as finance, more realistic | Greg’s experience. He still says “I wouldn’t 100% trust it”. |
| Strong at operating a computer | Can use old apps that have no API | Described, not demonstrated in the episode. |
For benchmark scores and pricing, see our Fable 5.2 review. The short version: it is the strongest Claude model available, and it costs double Claude Opus 5.5 per token, so match the job to the price.
What are the 12 Claude Fable 5.2 business ideas?
Greg groups his prompts into four areas. We have reworded each one and added a quick reality check. His own wording is in the video description.
| # | Idea | What it produces | Our reality check |
|---|---|---|---|
| 1 | Audit your whole business | A ranked list of the 10 changes most likely to make money, with evidence and a running log | Great first use. Check the evidence yourself before acting. |
| 2 | Do a freelance job end to end | Finished work, plus which parts need a human check and a suggested fixed price | Read the platform’s rules on AI use first. |
| 3 | Research 100 niches at once | One table of complaints, current spending and reachability, with five to build for | Treat as a shortlist, then talk to real customers. |
| 4 | Rebuild old software | A simple web app that replaces a client’s long-running spreadsheet, with data moved and a staff guide | High value, but test with real data and keep the old sheet as a backup. |
| 5 | Deal checker for buying a business | Every number labelled verified, unverified or suspicious, plus 15 questions for the seller | A first pass only. Use an accountant before you buy. |
| 6 | “Grill me” | A skill file built from 20 questions about how you make decisions | Cheap, useful and works with any good model. |
| 7 | Run apps with no API | Exports from an old app matched against a bank statement, with unpaid items listed | Keep the “ask me before sending” step. |
| 8 | Red-team your plan | An investor-style argument against your plan, plus what would change its mind | Useful for spotting blind spots. |
| 9 | Short-form editor | Best 45 seconds of a talking-head video, vertical, with captions and three hooks | Needs the right tools connected. See our Claude video editing guide. |
| 10 | Niche explainer videos | A 60-second animated explainer, built in code, with five opening variations | Check facts and platform rules on repetitive content. |
| 11 | One episode, every format | A PDF playbook, five posts in your voice, a newsletter and a small free tool, with unverified claims flagged | A strong fit for building owned reach. |
| 12 | One-person hardware company | A device design with priced parts list, circuit, firmware, printable case and the cost to make 100 | A paper design, not a safe, certified product. |
💡 Tip: Fable 5.2 is not needed for every idea on this list. Prompts 6, 8 and 11 work well on cheaper models. Our guide to choosing the right Claude model explains how to decide.
How do you turn a prompt into a fixed-price service?
Greg’s main money idea is a shift in how work is priced. Jobs like rebuilding a spreadsheet used to be billed by the hour. If an AI does most of the work, he suggests selling the outcome instead: one job, one fixed price.
That can work, but only if the job is narrow and repeatable. A sensible way to package one of these prompts:
- Pick one output, such as “a ranked audit of your website, pricing and onboarding”.
- Run it on your own business first and on one or two friendly test clients, free or at a low price.
- Write down your checking steps. What you verify, what you rewrite and what you never send unchecked.
- Time and cost it honestly, including model usage, your review time and revisions. Our AI agent cost guide helps with this.
- Set the scope in writing: inputs needed, rounds of changes and what is excluded. See how to scope an AI-assisted service without overpromising.
⚠️ Warning: Be open with clients about how the work is produced and what you have checked. Promise the deliverable, never a business result such as “more sales”.
What did the live demos actually show?
This is where it helps to separate evidence from enthusiasm. Here is what was on screen, as far as the episode shows:
- Business audit. Greg ran the audit prompt on LCA, his design agency. He showed a ranked list of changes with evidence and said the results were “to the point”, as they had been on other businesses he incubates. We did not see whether any change has been made or tested.
- Freelance job. Asked to find a job it could do, the model picked an Upwork listing for an equity analyst report at a $450 fixed price and started the work. It was still working when Greg moved on. Nothing was submitted or paid on camera.
- 100 niches. The subagent run was started but had not finished by the end of the episode.
- Red-team. Greg gave it his agency’s vision deck and says the feedback was, in some ways, better than a well-known adviser’s. That is his opinion, not something viewers can check.
- Hardware. The model produced a design document for “Plain One”, a minimalist calls-and-texts phone with a physical keyboard and a monochrome screen. It is a detailed plan, not a built or tested device.
🔎 Unconfirmed / rumour: Greg says he recalls Anthropic stating that Fable 5.2 designed three working devices from a single prompt, and mentions a large jump on a freelance-work benchmark. Check Anthropic’s own launch post for the exact claims before quoting them.
Want to try one today? The business audit (prompt 1) is the safest place to start. Run it on your own website and offers, then check every “evidence” point it gives you before changing anything.
Can Fable 5.2 run a content or faceless channel business?
Greg makes a strong case for using Fable 5.2 to produce videos, saying they can cost roughly $1 to $5 each. He suggests faceless explainer channels in a niche, and compares social accounts to property you build up over time.
The idea that matters most for most readers is his point about owning distribution. But a social channel is still rented reach. The platform sets the rules and can change them. Use any channel you build to move people onto an email list or website you control. Our article on owned reach vs rented reach explains why.
Two practical cautions:
- YouTube and other platforms limit or demonetise mass-produced, repetitive content. Each video needs genuine value, accurate facts and a clear point of view.
- Prompt 11 (“one episode, every format”) is the lower-risk route. It turns content you already made into a playbook, posts and a newsletter, and asks the AI to flag claims it could not verify.
If you want the editing side, see our guide to editing video with Claude.
Can one person really design hardware with it?
Greg calls this “vibe manufacturing”. One prompt asks for a single-purpose device for one niche, with a priced parts list, circuit design, firmware, a 3D-printable case and the cost to make 100. In the demo, it designed a pared-back phone that does only calls, texts, contacts, notes, a clock and settings.
That is a remarkable first draft for an idea. It is not a product you can sell. Anything with a battery, mains power or a radio needs real prototypes, safety testing and, in the UK, the right conformity marking before it goes on sale. A mobile phone also needs network and radio approval.
Our step-by-step guide to using AI to plan a simple electronics gadget covers the safe way to start.
Which ideas should a small business try first?
If you run a small business or create content, you do not need to start a new venture to get value from these prompts. A sensible order:
| Start here | Why | Model to use |
|---|---|---|
| Audit your own business (1) | Low risk, and the output is for you, not a client | Fable 5.2 if the job is big, otherwise Opus |
| Grill me (6) | Gives every future AI task your way of working | Any good model |
| One episode, every format (11) | Turns existing work into owned assets such as a newsletter | Sonnet or Opus is usually enough |
| Red-team your plan (8) | Cheap way to stress-test a launch or offer | Any good model |
| Then: one fixed-price service (2, 3, 4 or 5) | Only after you know your checking process | Fable 5.2 for long, hard jobs |
💡 Tip: Before you buy a business or sign a deal, use our guide on checking a seller’s accounts with AI.
What are the risks and limits?
- Cost. Fable 5.2 is the most expensive Claude model. A 10-hour job can use a lot of tokens, so set a budget and a stop point. Our kill switch guide shows how.
- Mistakes still happen. Greg himself says to assume some hallucination. Check numbers, sources and code before anything reaches a client.
- Platform rules. Freelance sites, ad platforms and video sites have their own rules on AI-made work and disclosure. Read them first.
- Confidential data. Client spreadsheets, invoices and a seller’s accounts are sensitive. Check your agreements and your AI plan’s data settings before uploading them.
- Computer use. Letting an AI operate apps is powerful. Keep a “check with me before sending or paying anything” rule, as Greg’s prompt does.
- No guaranteed income. These are ideas from one creator’s experiments. Whether any of them earns money depends on your market, your pricing and your quality checks.
How to test one Fable 5.2 business idea this week
- Choose one prompt from the table, ideally the business audit or the grill-me prompt.
- Decide your model. Use Fable 5.2 only if the job is long and hard. Otherwise start with a cheaper model.
- Run it on your own business and save the output.
- Check every claim. Mark what was right, wrong or vague.
- Write down the time and cost. Would someone pay a fixed price for this, after your checks?
- Repeat once more before you offer it to anyone else.
Frequently asked questions
Can you make money with Claude Fable 5.2?
Possibly, by selling finished work such as audits, research or rebuilt tools at a fixed price. Greg Isenberg shares ideas and early demos, not proven results, so treat it as a test rather than a promise.
Do I need Fable 5.2 for these prompts?
No. Long, complex jobs benefit most from Fable 5.2. Prompts like “grill me”, red-teaming a plan or repurposing a transcript work well on cheaper Claude models.
Is it allowed to use AI for freelance jobs?
It depends on the platform and the client. Check the marketplace’s rules and the job brief, and be open about how you work.
Can I trust Fable 5.2 with financial figures?
Use it as a first pass that highlights what to question. It can still be wrong, so have an accountant check anything you rely on.
Where are Greg’s exact prompts?
He shares them in the video description on his YouTube channel. Our table above summarises each one in our own words.
The bottom line
Claude Fable 5.2 makes some genuinely new business ideas possible, mainly long, well-defined jobs you can package and sell as a finished output. Greg Isenberg’s 12 prompts are a useful menu. The sensible path is to try one on your own business, learn how much checking it needs and only then decide whether it is worth selling.
Your next step: run the business audit prompt on your own website and offers this week, then read our Fable 5.2 review to decide whether the bigger jobs justify the price.
Related posts
- Claude Fable 5.2 Review: Is It Really the World’s Best AI Model?
- Which Claude Model Should You Use? Haiku, Sonnet, Opus or Fable
- How to Use AI to Plan a Simple Electronics Gadget: Parts List, Code and 3D-Print Case

