Source: raw/Claude_Fable_5.2_-_How_I_d_Make_Money_With_It.md

Creator: Greg Isenberg, The Startup Ideas Podcast (solo episode) · URL: https://www.youtube.com/watch?v=G3i2k6DdI0o · Duration: not recorded · Platform: YouTube (auto-captions, fetched 2026-11-11). No sponsor segment. He says the full prompt list goes in the description, with bonus prompts in a pinned comment if the video reaches 5,000 likes.

Recorded on Fable 5.2’s release day, this is a prompt list rather than a review: twelve prompts in three groups (making money, being productive, video and content) plus a hardware-design demo, several run live in the Claude app. The common thread is selling finished outcomes at a fixed price instead of hours. Isenberg’s claims about the model are his own impressions or his relay of Anthropic’s launch post; the model facts belong in the Fable 5.2 model article. See Claude Fable 5.2 + Mythos 5.2 for the model facts.

Key Takeaways

  • His thesis: sell the job, not the hour. He repeats it across the freelance, legacy-software and content prompts: work people paid 100 an hour for can be sold “per output” as a productized service.
  • Two prompts were run live with visible results.
    • The business audit on his design agency (latecheckout.agency) returned a ranked list he calls “to the point”, adding that it has been on every company he incubates.
    • The freelance prompt, told to find its own job, picked an Upwork posting (an equity analyst report, $450 fixed price, intermediate level, posted two days earlier) and started doing the work. He did not show the finished report.
  • The 100-niche subagent prompt was started but not shown finishing. He says he did not know whether it would finish during recording.
  • Red-teaming his plan beat a human advisor, in his view. He gave it his agency’s vision deck and goals; “in some ways the 5.2 feedback was even better” than feedback from “a really well-known sort of advisor”.
  • The hardware demo produced a full device document from one prompt. Asked for a minimalist phone, it proposed “Plain One”: a pocket 4G handset limited to calls, texts, contacts, notes, clock and alarm, with a 45-key thumb keyboard, a 2.7-inch monochrome sharp memory LCD, and no browser, feed or app store. The document included a parts list, circuit design (power rails, pin map, modem connections, display connectors, keyboard matrix), firmware and a printable enclosure. He calls this “vibe manufacturing” and still says not to trust it 100%.
  • Content economics he states: faceless explainer channels where a video costs “2, 4, $5”. He compares channels to low-capital real estate: they earn AdSense and followers “you can later sell”. These are his figures, not measured.
  • Two friends’ videos are cited as evidence (a documentary-style 30-40-second video from “Riley” and an explainer from “Hunter”, each from one prompt). Neither is shown or linked in the transcript.

The prompts (as read on air)

Making money

  1. Business audit. “Audit my entire business, the website, pricing, onboarding, emails and top five competitors, work until it’s actually finished, keep a running log, give me a ranked list of the 10 changes most likely to make money with the evidence for each.”
  2. Freelance job end to end. “Do the whole job as if you were a freelancer, then tell me which parts you’re confident in, which parts a human should check, and what you charge for it as a fixed price service.” (Paste a job posting, or tell it to find one.)
  3. 100 niches at once. “Spin up a subagent for each one of these 100 niches. Each one finds the top five complaints people have, what they pay for today, and how hard it is to reach. Give me one ranked table and the five I should build for.”
  4. Rebuild the ancient software. “Here’s the spreadsheet my client has run their business on for 12 years. Figure out what it actually does, rebuild it as a simple web app, move the data over, and write a one page guide for the staff.”
  5. Deal checker for buying a boring business. “Here’s the seller’s P&L on [the] listing. Find everything that doesn’t add up. Mark each number as verified, unverified, or suspicious, and write the 15 questions I should ask before making an offer.”

Being productive

  1. Grill me. “Interview me with 20 questions about how I make decisions in my business one at a time, then turn my answers into a skill file you’ll follow from now on.” He calls this the one prompt “literally everyone” should use.
  2. Run the apps that never had an API. “Open our old invoicing app, export last month’s invoices, match them against the bank statement and list anything unpaid, and then ask me before sending anything.”
  3. Red team my plan. “Here’s my plan for the next quarter, argue against it like an investor who wants to pass, then tell me the one thing that would change your mind.”

Video and content

  1. Short-form editor. “Here’s a raw four minute talking head video. Cut the silences and retakes, find the strongest 45 seconds, add captions and simple motion graphics and export it to vertical video, and then show me three different hooks.”
  2. Niche explainer videos. “Make a 60 second animated explainer on this topic for this niche. Write the script, build the animation, encode and render it, then make five variations with different first lines.”
  3. One episode, every format. “Here’s the transcript, make a PDF playbook, five tweets in my voice, a newsletter, a free tool someone could use in two minutes, and then flag any claim you couldn’t verify.”

Hardware

  1. Single-purpose device. “Design a small device that does one job for one niche. Give me the parts list with the prices, the circuit design, the firmware, and an enclosure I can print.” Follow-up: “Tell me what it costs to make 100.”

Try It

  1. Run prompt 1 on your own business first. It is the one he shows working and recommends for anything you are building. Keep the “running log” and “evidence for each” clauses; they are what make the output checkable.
  2. Price one productized service with prompt 2 on a real posting in your field, and read the “which parts a human should check” section before quoting anyone.
  3. Use prompt 6 once and keep the skill file. Re-run it when your decision rules change.
  4. Keep the human gates he wrote in: prompt 7 ends with “ask me before sending anything”, and prompt 11 ends with “flag any claim you couldn’t verify”. Copy those endings into your own prompts.
  5. For prompt 12, treat the output as a draft for an engineer. He says not to trust it 100%; a parts list and circuit need checking before you order a batch of 100.

Open Questions

  • Model capability claims in the episode (working 10-12 hours unattended, the freelance-benchmark figure he quotes from Anthropic’s blog, better subagent handling, saying “I don’t know”, computer use, Anthropic’s “three working devices from a single prompt”) are relayed by the host and were not checked here; they are sent to the Fable 5.2 model article for verification.
  • The finished Upwork report, the 100-niche table and the friends’ videos were not shown.
  • He says “Fable 5.5” once while introducing the productivity prompts; from context he means 5.2.
  • “Astra”, which he compares the hardware result to, is read here as OpenAI’s GPT-6 Astra; he does not say which product he means.