Fable Is Back. Here's What I'm Doing With It Before I Build Anything.

Back in May I wrote about a brief window with Fable — a long-horizon model that held a sprawling attribution problem together when my daily driver couldn't. Then it shipped for real on June 9, disappeared on June 12 while Anthropic sorted out export controls, and came back yesterday.

I'm not going to re-review it. The May post still stands: same shape, same lesson. What I want to write about is what I'm pointing it at, because it's the least glamorous work I do and it's exactly where a model like this earns its keep.

The quiet project

For most of this year I've been helping friends put AI to work in their businesses on nights and weekends — a VA claims practice, a behavioral services firm, a CPA. It's been informal. At some point informal stops being honest: people want an invoice, a contract, an entity on the other side of the handshake.

So I'm going to formalize it. An LLC, a name, a bank account, the stack. But I've watched enough founders (and been one of the people cleaning up after them) to know the expensive mistakes happen before you file anything. Wrong state. A name that collides with a trademark. A bank you outgrow in six months. Books you have to rebuild in April.

This summer is the research phase. Nothing gets filed until I understand the whole path.

Why this is a long-horizon problem

Forming a company sounds like a checklist. It's actually a chain — and the links depend on each other.

• Where you form determines your publication requirement, your registered agent, your annual fees, and your tax filings.
• What you name it depends on a state entity search, a federal trademark search, what domains exist, and what already owns the Google results.
• Where you bank depends on the entity type, what your bookkeeping looks like, and whether you want an API your tools can read.
• How you keep books depends on how you'll be taxed, which depends on the entity, which depends on where you form.

Every one of those is a medium-hard question. Ask them one at a time and you get clean answers that quietly contradict each other. That's the failure mode I described in May — not too hard, too long.

The research phase of starting a business is one connected chain, not a list. Answers you lock in early have to still be true forty questions later.

What I'm actually running

I've got a running workspace with Fable where the whole chain lives. A few of the threads:

• New York vs. Delaware for a one-person consultancy that lives in Manhattan. (Spoiler: the internet's default advice is aimed at venture-backed startups, not me.)
• New York's publication requirement — the six-weeks-in-two-newspapers rule that can cost four figures in Manhattan and a few hundred dollars a county or two away.
• Name clearance across the NY Department of State, the USPTO, and the domain registrars — all at once, so I stop falling in love with names that are already taken somewhere.
• The tool stack: domain, email, banking, CRM, bookkeeping, and how each one talks to Claude.

The thing I notice, versus doing this with a general model: I can come back three days later, ask "given everything we've settled, what does the publication requirement look like now?" and the answer reflects everything we settled. It doesn't re-derive the state decision. It doesn't forget I ruled a name out.

The honest caveat

I'm still on Opus for most of my day. Drafting, debugging, decisions — short, self-contained work — doesn't need endurance. This is a specific project with a specific shape, and I'm using the tool that fits the shape. If you take one thing from this post, take that.

What's next

I'll write the whole thing up as I go — formation, the bank, the books, the CRM, the brand — as a short series, so the next person doing this in New York has a real map instead of a listicle.

If you've formed a company recently and hit a surprise I should know about, tell me in the comments. And if you're deciding whether to formalize your own side work, ask — I'd rather answer your questions than guess at them.