Greg Isenberg's solo episode is a tour of 12 ways to put Claude Fable 5.2 to work, from a business audit and freelance research to video editing and a minimalist phone design. The most useful part is not a promise of automatic income. It is a set of concrete job briefs. We transcribed the 12 prompts from the screenshots supplied with this article and grouped them below so you can copy one, provide the missing inputs and review the result.
Start with the prompt that produces a result you can inspect: a ranked business audit, a red-team memo or a draft edit from footage you own. Give Claude access only to the necessary files and tools. Verify the numbers, facts, rights and final actions yourself. Greg's Upwork example is a draft in progress; his phone is a design concept. Neither is shown as a completed sale or manufactured product.
Watch Greg's 12-Prompt Walkthrough
Prompt provenance: Wording follows the six screenshots provided for this article, with punctuation and line breaks normalized. Greg's live results segment shows the business audit and an Upwork report draft still running. The use notes under each prompt are our editorial checks, not additions to Greg's prompts.
What Fable 5.2 Can and Cannot Prove Here
Anthropic's launch post reports stronger performance on its Remote Labor Index: 41.3% for Fable 5.2 at high effort versus 21.3% for Opus 5.5 on that evaluation. This is a benchmark score, not a 41% chance of winning a freelance contract. The launch recommends starting with Opus 5.5 for most everyday coding and knowledge tasks, reserving Fable for more ambitious, open-ended work. At the time of this article, Anthropic lists API pricing at $8 per million input tokens and $40 per million output tokens, with a higher-priced fast mode. Long runs and many subagents can therefore cost real money even before human review.
Greg describes 10- to 12-hour autonomous jobs and better expressions of uncertainty. The video demonstrates promising work, but it does not establish a guaranteed run length or eliminate hallucinations. In finance, legal, client deliverables and hardware, "I don't know" is welcome only when backed by source checks and a qualified reviewer.
Prompts 1-5: Making Money and Finding Work
These five prompts can reveal opportunities, scope services and prepare drafts. They do not validate demand, authorize access to client data, or replace a professional sign-off.
1. Audit your entire business · 01:00
Audit your entire business
What to check: Greg runs this against his design agency. Treat the ranking as hypotheses: confirm pricing, competitor comparisons and source dates, then measure one change at a time. Watch this chapter.
2. Do a freelance job end to end · 02:05
Do a freelance job end to end
What to check: In the episode, Claude finds a $450 Upwork equity-report listing and starts a draft. No client acceptance, payment or completed professional review is shown. Never submit financial analysis without checking filings, calculations and applicable platform rules. Watch this chapter.
3. Research 100 niches · 03:46
Research 100 niches
What to check: Supply the actual niche list and source constraints. One hundred parallel searches can be expensive and shallow; validate the final five with interviews and real willingness-to-pay evidence. Watch this chapter.
4. Rebuild old business software · 04:33
Rebuild old business software
What to check: Begin with a copy and map formulas, permissions, exceptions and hidden manual steps. Test totals and exports against the original before any migration; retain rollback. Watch this chapter.
5. Check a business deal · 05:25
Check a business deal
What to check: Use the labels as a request for evidence, not as a completed audit. A model cannot verify a seller-provided number without independent records. An accountant or deal adviser must review consequential findings. Watch this chapter.
Prompts 6-8: Decisions and Legacy Apps
The productivity prompts are strongest when the agent learns your rules but does not silently acquire authority. Keep the generated skill reviewable and any money movement or outgoing message behind approval.
6. Grill me on decisions · 08:53
Grill me on decisions
What to check: Review the generated skill before installing it. Remove private customer details and write the exceptions and approval boundaries explicitly. Watch this chapter.
7. Operate a legacy app · 09:33
Operate a legacy app
What to check: Use a test account or read-only export first. Bank matches need date, amount, currency and partial-payment rules; the send step remains human-approved. Watch this chapter.
8. Red-team my plan · 10:07
Red-team my plan
What to check: Ask for the strongest counterargument and the evidence that would settle it. Do not mistake a persuasive critique for an actual investor decision. Watch this chapter.
Prompts 9-11: Video, Explainers and Repurposing
Here the bottleneck moves from generation to editorial selection. Greg suggests very low per-video costs, but the episode does not provide full production receipts; pricing depends on tools, retries, assets and review time. A publishable video also needs legitimate footage and music rights.
9. Edit a short-form video · 12:39
Edit a short-form video
What to check: Give the agent the original footage and edit rights. Review cuts for meaning, captions for names and facts, and music/graphics for licensing before publishing. Watch this chapter.
10. Make a niche explainer · 13:05
Make a niche explainer
What to check: Replace the placeholders with a specific audience and a source brief. Check claims, readability on mobile and whether the animation actually teaches the point. Watch this chapter.
11. Repurpose one episode · 13:24
Repurpose one episode
What to check: This is a production brief, not a permission to publish. Review attribution, voice, factual claims and the free tool before distributing any format. Watch this chapter.
Prompt 12: A One-Person Hardware Company?
12. Design a one-person hardware product · 13:56
Design a one-person hardware product
What to check: The phone Greg sketches is a concept and mockup, not a built or certified handset. Recheck live component availability, electrical safety, manufacturing yield, assembly labor, testing and certification with engineers. Watch this chapter.
Greg's minimalist phone brief produces a specification and mockups in the episode. That is a useful design starting point, not evidence that the phone works. Anthropic's separate research with Diode Computers did produce three working electronics prototypes from short prompts: a speakerphone, voice-assistant hub and virtual pet. Engineers still assembled and tested them; Anthropic says they have not undergone electrical validation, antenna tuning, thermal testing or certification, and were not designed for mass production. Anthropic also discloses a commercial relationship with Diode.
Watch Anthropic's voice-hub prototype video. This is a different project from Greg's minimalist phone concept.
A Responsible First Pilot
- Pick one result and its owner. Choose a prompt whose output you can judge this week, such as a ten-item audit or a 45-second edit.
- Prepare the inputs. Supply actual files, a list of niches or a real job brief; redact private client and financial data unless access is authorized.
- Set boundaries. Define sources, budget, time limit, permitted tools, deliverable format and actions that need approval.
- Review against a baseline. Compare the output with the existing process on accuracy, useful time saved and cost per accepted result.
- Sell only what you can stand behind. Quote a fixed price when scope and acceptance criteria are clear; include your own QA, revisions and support.
That is the durable business opportunity in Greg's list: a service the buyer can trust, not just an impressive run of Claude.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | Why Greg picked these prompts |
| 01:00 | Business audit |
| 02:05 | Freelance job |
| 03:46 | 100 niches with subagents |
| 04:33 | Rebuild old software |
| 05:25 | Business deal checker |
| 06:54 | Results of the live tests |
| 08:53 | Decision interview |
| 09:33 | Operate legacy apps |
| 10:07 | Red-team a plan |
| 12:39 | Short-form video editor |
| 13:05 | Niche explainer |
| 13:24 | One episode, every format |
| 13:56 | One-person hardware company |
| 16:42 | Minimalist phone concept |
| 18:31 | Closing thoughts |
Sources and Scope
Checked 11 November 2026. The episode and supplied transcript establish what Greg said and demonstrated. The 12 prompt texts are transcribed from the six screenshots supplied by the reader, not reverse-engineered from a generated output. Anthropic's launch supports model availability, pricing, benchmark context and its Opus-first recommendation. Anthropic and Diode's electronics write-up supports the three working prototypes and states their engineering limits. We did not rerun Greg's prompts, inspect his Upwork deliverable or independently validate production cost and earnings claims.