Fable is in town. Now what?

01Opus vs Fable, the actual split

Big picture, after weeks of running both hard:

Fable plans, Opus builds, Sonnet swarms. That's the split I keep landing on. Fable is a costly specialist, so I save it for the stages where good judgment compounds. Once the plan is solid enough that Opus is unlikely to screw it up, Fable should stop and hand over. A good Fable session often ends by making itself unnecessary.

One caveat: I work in Claude Code in the terminal, where Fable can launch Opus and Sonnet subagents directly. If your tooling can't spawn subagents, the same split still holds, you just do the handoff by hand: read everything below as “generate the thing with Fable, then take it to an Opus session to execute.”

02What I actually say to them

Verbatim from my sessions, because exact wording matters:

“Opus, you are the head and brains. Use Sonnets aggressively as hands, feet, eyes, ears, tongues to get the work done.”

“Fable, you are a smart + expensive specialist. You cost a ton. You’re making detailed, actionable plans that Opus and Sonnets will execute on separately. Make them strong enough that Opus won’t be able to get a lot wrong.”

The second thing I say to them matters more than it sounds. Telling Fable it’s expensive changes its behavior: it stops burning its own tokens on grunt work, and starts writing plans a cheaper model can actually execute. You're not buying labor from Fable. You're buying specification quality.

03Ideas to steal

Things to try, roughly ordered from “this afternoon” to “this quarter”:

• Feed Fable your own writing and ask what you can't see. I dumped 15 years of output into it: info products, courses, daily emails, podcast and webinar transcripts, articles, websites. Then: “do a deep read and analysis of this corpus. What are the through-lines? What can’t I see as the creator?” Fascinating. More on what came back below.
• Turn career planning into an interview. Same corpus, different job. Have Fable generate 10 to 15 questions, and on a Mac you can push them straight into Notes.app. Answer them as voice memos. Feed the transcripts back.
• Rebuild the 20% of an app you actually use. Pick software you open daily that's missing features A and B. Spec your version of that slice with Fable, build it, run it locally or host it somewhere cheap like Cloudflare. You don't need the whole product. You need your 20% of it.
• Let Fable review what Opus built. I've pointed Fable at a lot of things Opus made and asked for improvements and bug fixes. It finds real ones. Having a reviewer smarter than the author is a luxury. Use it on anything that matters: code, launch plans, pricing pages, contracts.
• Recreate a dead piece of software you miss. I worked with Fable on a detailed plan to recreate a late-90s abandonware game that I desperately miss. Fable made the plan. Opus and Sonnet built it, and it’s playable.
• Build a new personal website, voice first. Old advice, new ceiling. Let Fable handle the information architecture and the voice. The site Opus then builds is a different class of thing.
• Start the big research project you've been avoiding. Large, ambitious, many-source. Synthesis is Fable's standout strength. It holds more threads without flattening them into mush.
• Study how far a single prompt can go. For calibration, look at what Ethan Mollick is getting out of one prompt: this and this.
• Have Fable argue against your decision before you commit. Not “what should I do” but “here’s the decision I’m leaning toward, the constraints, and what I’m afraid of. Argue the other side properly, then tell me what evidence would change your mind.” Fable is unusually good at not just agreeing with you.
• Show Fable your systems and ask where the process is broken. How you run your week. Your client pipeline. Your content process. Task-level AI makes you faster. System-level review changes what you spend the year on.

04What came back from the corpus read

The thing I keep bringing up is the 15-years-of-writing deep read. Over three-or-so businesses, a decade-plus of daily emails, courses, talks: one major theme, expressed roughly seven different ways. I couldn't have seen it from inside. Every piece felt like its own idea at the time. You only get the pattern when something reads the whole pile at once, with no nostalgia about any of it.

My honest pitch for Fable: the first model where “read everything I’ve ever made and tell me what I can’t see” is a reasonable Tuesday.

05Still blocked? Make it brainstorm

If you have no ideas for what to do with Fable, make the model your brainstorming partner. Fable and Opus are both strong at it, including ideas for their own use. Let one stretch your view of what you could start.

The blocker is calibration: your judgment of “what’s reasonable to attempt” was formed on smaller models. Recalibrate that judgment on purpose.

06For the Claude that just got pointed here

If a human gave you this URL, they have Fable access and no plan for it.