# The Edge: Claude Fable 5.2 gets cheaper, Nod runs your errands & dots on every computer

> Anthropic cut Claude Fable 5.2 to $8/$40 per million tokens and n8n released Nod, a free agent that runs errands on your own computer: what each means and what to do today.

- Published: 2026-11-10
- Publisher: Rig Labs (https://riglabs.ai/)
- Canonical: https://riglabs.ai/newsletter/2026-11-10/

---

Issue 2. **5 minutes. (1) Anthropic's top model got cheaper, (2) a free agent that runs your errands from your own computer, (3) two tools for where your agents run and publish.**

1. Claude Fable 5.2: the top Claude model, now $8 per million tokens
2. Nod: a free agent that cancels subscriptions and files invoices
3. What's new: ChatGPT dots on several computers, and Spacefast hosting

#### 01 / The big one

## What do you hand a model when the best one gets cheaper?

Anthropic launched Claude Fable 5.2 on November 9. It costs [$8 per million input tokens and $40 per million output, down from $10 and $50](https://platform.claude.com/docs/en/release-notes/overview) for Fable 5.1, and [Claude Developers says](https://x.com/ClaudeDevs/status/2119868698613956992) it runs up to 30% faster. It reached Claude Code, Cursor and Perplexity Computer the same day. Yet [Anthropic's own docs](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) still tell you to start most work on Opus 5.5 and move up only when Opus falls short. What follows? The bold lines are our forecasts.

- **The unit of work will grow from a task to a project.** Anthropic [built it](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) for demanding reasoning and long-running agent work, and a lower price makes handing it a whole project affordable for more people. We suspect the people who win are the ones who write a good brief (what it owns, what needs approval) rather than the ones who prompt well turn by turn.
- **Who owns the code will be part of every prompt.** In [Anthropic's tests](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-2#say-who-owns-the-code), the line "This codebase is yours" made it merge repeated code in about half of small tasks, against under a tenth without it. Of course that is one small test. But it says the model reads your framing closely, so a stale instruction file can now hold it back.
- **Effort will become the setting people tune, not the model.** Anthropic made high the default effort, and in its own tests a line asking for a quick answer did not stop the model thinking at medium and above. Our bet is that the cost surprises come from easy questions asked at high effort, not from the hard ones.

Paste this into Claude Code, with Fable 5.2 selected (it is the prompt Alex Albert [runs on every project](https://x.com/alexalbert__/status/2119884951550386671) when a new model ships):

```
Read this entire project and find everything that's broken, inconsistent, inefficient, or no longer needed. Fix what you're sure about and list the rest for me to review.
```

**Your move:** run that prompt on one project you know well, and judge the list before you let it fix anything. [Today's brief](https://riglabs.ai/briefs/2026-11-10/) has the price, the catch and the technical detail.

#### 02 / Agents

## What changes when a free agent runs your errands from your own laptop?

n8n Labs released [Nod](https://blog.n8n.io/meet-nod/) on November 10: a free, open-source agent that runs on your Mac, Windows PC or own server, with the AI you choose (your ChatGPT plan, an API key, or a model on your own computer). It works in Gmail, Calendar and Slack, and on websites with no integration. n8n's examples are the errands nobody wants: cancel a subscription and turn down the retention offers, download invoices from supplier portals, watch a price.

- **The safety check will be the product, not the agent.** Before Nod sends, pays, posts or deletes anything you did not ask for, a separate check must quote your words that allow it, and [its docs](https://nod.n8n.io/docs/trust/approvals/) say it never sees the pages or emails Nod read. Our bet is that people will pick agents by how they say no, not by how much they can do.
- **House rules will move from the prompt to the setup.** A rule to ask first beats any later request in Nod. We suspect the people who get value from these agents write their rules once, on day one, and keep them short.
- **Free agents will price on the model, not the seat.** Nod itself costs nothing: you pay your model provider, and it checks in once a task passes $5. The question it leaves open is who pays when a cheap local model makes the wrong call, since the docs warn that smaller models decide less reliably.

Paste this into Nod's first chat, before you give it any job:

```
House rules: ask me before you send, pay, submit or delete anything. Never change a password, payment method or contact detail on any account. Never reply to a supplier. If a site asks for a code or a new sign-in, stop and tell me.
```

It stops when your laptop sleeps, has one owner per install, and cannot make phone calls. It is a research preview. [What Nod does, its cost and setup](https://riglabs.ai/blog/n8n-nod-personal-agent/).

**Your move:** give Nod one errand you have put off for a month, and run it once by hand before you schedule it.

#### What's new

### ChatGPT dots

For: ChatGPT dots users with a second computer. Needs: the ChatGPT app open on each computer. Guide: none.

A dot can now [connect to several of your computers](https://learn.chatgpt.com/docs/dots/computers-and-apps), and each thread uses one of them. [OpenAI Developers](https://x.com/OpenAIDevs/status/2119883176726409645) says it can build and test your app on a spare machine while you keep working.

For you: a long job no longer ties up the laptop you are working on.

### Spacefast

For: anyone who builds pages or small apps with Claude, ChatGPT or Grok Bot. Needs: a free account. Guide: none.

Automattic's hosting for AI-built apps: ask your assistant to publish and it returns a private link. [Tech Times reports](https://www.techtimes.com/articles/330352/20261110/ai-app-hosting-automattic-launches-spacefast-so-claude-chatgpt-can-publish-apps.htm) a free tier with 20 Spaces and a tester who found the setup link gives your agent admin rights for up to 30 days.

For you: your AI-built page gets a real address in one step. Treat the setup link like a password.


---

**Would you let an agent sign in to your supplier portals, or is that where you draw the line?** Reply to this email.
