Claude Opus 5 ('Fable') Is Back: Five High-Leverage Use Cases Before Your Tokens Run Out
The podcast centers on a hands-on demo of what the host calls 'Fable' — apparently Anthropic's top-tier Claude model — which has returned to subscription access after an 18-day absence, with availability capped at July 7th and further constrained by weekly usage limits. The host frames the session around five use cases: discovering high-value tasks, getting business/life advice, shipping code, planning new features, and refactoring large codebases.
The most emphasized use case is using the model as a strategic business advisor. The host recommends feeding it a structured plan document covering yearly goals, decision principles, business positioning, energy levels, and financial context — then connecting live data sources like bank accounts, newsletter stats, YouTube analytics, and website traffic to generate a grounded three-month focus plan. He describes this as the highest-leverage thing you can do with the model, since it shapes months of work.
On the technical side, the host ran a bug-finding exercise on a fitness app and the model spun up five parallel agents, passed all unit tests, but still surfaced over 12 major bugs — including a session data leak that could expose one user's data to another. The host claims this significantly outperformed GPT and Opus on the same prompt. A separate feature-planning demo produced a detailed, phase-by-phase nutrition tracking spec, including a recommended free USDA food data API.
Several practical concerns were raised throughout: - Ultra-high effort mode burns token limits extremely fast - The model can loop and waste tokens if not monitored - Large-scale tasks like mining a content archive may consume an entire weekly limit in one session - A Stripe anecdote (migrating 50M lines of Ruby in a day) was cited as a benchmark for codebase refactoring potential, though this claim was unverified
The bottom line the host lands on: use cheaper models (Opus, GPT) for setup, API connections, and execution; reserve Fable strictly for tasks requiring deep reasoning — strategic planning, bug audits, architecture decisions, and large refactors where the quality delta justifies the token cost.
The host demos five practical use cases for Claude's most powerful model tier while it's temporarily available on subscription plans. The core message: save 'Fable' for deep thinking tasks, plan with it, and execute with cheaper models. Token limits are tight, so prioritize ruthlessly.
Using Fable to get life and business advice is the highest leverage activity you can do with Fable because it can impact what you do over the next three months, six months or even a year.
Use Opus or GPT to prep the right plan document and hook up all the APIs and MCPs to get things ready for Fable. Don't use Fable for boring work like hooking up an MCP or API.
Plan with Fable and execute with another model. Fable is really great at making detailed plans; once you have a detailed plan, pass it to a cheaper model to do the actual execution.
Claude Fable 5, the world's best AI model, is finally back after being banned by the US government for 18 days.
Fable is only available through July 7th on your claw subscription and you can't use it anymore after you hit 50% of your weekly usage limits.
After hitting 50% of weekly usage limits, Fable will only be available through pay as you go API credits.
Fable found in memory that the host recently left his job to work on his business full-time and also found a bunch of skills built to get advice on the business.
LLMs are very good at analyzing large corpus of data and Fable can take this to the next level.
The advisor skill escalates to a council skill that has three different personas: the actual customer being built for, a skeptic about the plans, and an execution-minded operator.
Fable spun up five agents to find bugs, ran all unit tests and everything passed, but then found over 12 major bugs and a bunch of minor bugs in the fitness app.
If you sign out involuntarily by shutting off the app, you could potentially leak one user's data into the next user's account.
When running the same bug-finding prompt on GPT 5.5 or Opus, it does find a few bugs but nowhere near the amount that Fable found.
Fable found an API to pull food and nutrition data from the USDA from the government and recommended using USDA FDC only as the food data source.
Stripe used Fable to refactor 50 million lines of Ruby code and migrate it in a day where it could have taken a whole team over two months to do.
Fable spun five different agents to audit the personal OS skill system and found 13 things to improve.
Fable is back only until July 7th on your cloud subscription.
Mining the archive into a content road map is probably going to burn my limit in one shot.
If you run Fable with ultra high effort all the time, your limit is going to run out almost immediately.
Fable is really smart, but sometimes it can get into a rat hole and just keep looping and burning your tokens, and you don't want that to happen since the tokens are so limited on your subscription.
Five use cases for Fable: using Fable to find Fable worthy work, getting life and business advice, making your project ship ready, planning for the next big thing, and refactoring a large codebase.
Before asking Fable for life and business advice, give it a plan document with sections for: overall goal for the year, decision-making principles, business positioning, energy assessment, and additional life and financial context.
You are Fable 5, the most capable AI model available. Look through my projects and memory. List the top five tasks that require deep thinking that you think are generally worth running by you. Please do not share any confidential information because I'm making this video and giving a demo.
Connect Fable to Mercury for bank information, Substack for newsletter stats, YouTube for YouTube information, Vercel for website analytics, and Google Workspace for plans and documents.
I asked Fable: I'm about to ship this fitness app that lets users track their workouts. I want you to find everything wrong with it. Read the whole codebase. Look for real bugs, broken edge cases, anything that fall over in front of a user and list all the things that you think need to be fixed and maintain a high quality bar.
I asked Fable: I want you to draft a detailed plan for a new nutrition tracking tab in this app. Please use searches online to find relevant resources and then lay out the full plan, the key phases, the decisions we need to make, the risks and the open questions that I need to answer. Flag anything that could sink the project if I get it wrong. Give me a plan clear enough that a cheaper or simpler model could execute step by step.
“Fable Sladvisor take a detailed look at my plan document and my content schedule. Feel free to pull other information as well. Then I want you to write a detailed one pager assessing my business and what I should focus on for the next three months.”
Shared with Earmark · earmark-ai.com