Anir SurenClaude Fable 5 is the most powerful model you can use right now, and the window to use it as part of your normal paid plan is closing. After that it moves to usage credits, which means every heavy job costs you extra. So the smart play this week is simple: give Fable the five jobs that actually benefit from a frontier model, the ones with a lot of context, a lot of judgment, and a real payoff. These are the five I'd run first, with the exact prompts ready to paste.
Quick setup note before you start: prompts 2, 3, and 4 work by letting Claude see your real data. Connect your calendar and email once under Customize → Connectors in the Claude app (Google Calendar and Gmail), and everything below just works. Prompt 1 runs in Claude Code, the rest in a normal chat.
1. The codebase audit prompt
This is the single best use of a frontier model if you build anything. Open Claude Code in your project and paste this. The "do not fix anything" line matters: you want the full damage report first, not a hundred surprise edits.
Read this whole codebase and find real bugs, broken edge cases, and anything that falls over in front of a user. List everything by severity, and do not fix anything until I say go.
Fable hands you a ranked bug report of things you didn't know were broken: race conditions, unhandled inputs, flows that die on mobile. Then you go down the list and say "go" on the ones worth fixing, one at a time, so you stay in control of every change.
2. The meeting prep prompt
This one turns your calendar into a briefing doc. With the calendar connector on, open a chat and paste:
Look at my calendar for this week, pull up everything you can find on each person I'm meeting with, and give me one page per meeting about who they are, what we've talked about, and the one thing I should push on.
You get one page per meeting: who the person is, the history you have with them, and the single thing to push on. It's the prep work a chief of staff would do, done before you finish your coffee.
3. The spending audit prompt
The uncomfortable one. Point Claude at your inbox (or paste in your statement) and let it show you where the money actually goes:
Break down my spending from the last 30 days into a table and a pie chart, and flag every subscription I barely use so we can cancel them one by one.
You get the table, the chart, and the list of zombie subscriptions you forgot existed, teed up to cancel one by one. If you want to go deeper on this whole workflow, I broke it down fully in the spending expose.
4. The inbox prompt
With the Gmail connector on, this clears a week of email in one shot, and nothing sends without your sign-off:
Go through my inbox from the last 7 days and sort everything into needs a reply, needs action, and can be ignored, and then draft replies for the first group but don't send anything until I approve.
Claude triages everything into three piles and drafts the replies that matter in your voice. You just read, tweak, approve. If you like this one, the full Gmail assistant setup makes it run every morning automatically.
5. The content repurposing prompt
For anyone who posts content. Upload any video (a talking head, a podcast clip, a tutorial) and paste:
Take this video and turn it into a week of content across Instagram, Twitter, and any other platform, and match my voice from the last 10 posts so I can pick what goes out.
One video becomes a week of platform-ready posts, captions, hooks, and threads, written to sound like you because it studies your last ten posts first. You pick what goes out, and your content calendar fills itself.
The honest part
Two things worth knowing. First, all five prompts work on any Claude model, so nothing here breaks when the Fable window closes. Fable just does them noticeably better, especially the codebase audit and anything with a lot of context, which is exactly why you run these now while it's included in your plan. Second, the connector prompts only see what you give them access to, and the inbox and codebase prompts are built to draft and report, not act, until you approve. Keep it that way. The approval step is the whole reason you can trust the output.
The best way to feel what a frontier model actually is isn't a benchmark chart. It's handing it a real week of your life, your code, your calendar, your inbox, and watching it come back with the work done.
Run the codebase audit first if you build, and the meeting prep one if you don't. Either way you'll get more out of Fable this week than most people got out of it all month.
– Anir
If this is the kind of AI thinking you want working on your own business, apply to work with me. And to catch the next breakdown, join the newsletter.