Fable 5 Plus Ultracode: How to Actually Use Claude's Strongest Combo

In plain terms: The company behind the Claude AI just released its smartest version yet, plus a setting that lets it split big jobs across a whole team of AI helpers working at once. This article explains when that combo is worth using and gives you the exact instructions to copy.

Two Anthropic releases landed within two weeks of each other, and together they change what one person can delegate to AI in an afternoon.

The first is Claude Fable 5, released June 9, which Anthropic calls its most capable generally available model (announcement). The second is quieter: a Claude Code setting called ultracode, introduced in late May (Anthropic's post). Most coverage treated them separately. The interesting part is what they do together.

A quick honesty note before the tips: this combo lives in Claude Code, which runs in a terminal. You do not need to be a programmer, but you do need to be comfortable installing a tool and typing commands. If that is not you yet, the concepts below still tell you what is now possible, and it is worth knowing.

What ultracode actually is

Plain English: normally, an AI assistant works on your task by itself, one step at a time. With ultracode on, Claude is allowed to act like a project manager instead. For a big task, it writes a plan, spins up a team of smaller AI workers, runs them in parallel, checks their work, and hands you the combined result. Anthropic calls these runs dynamic workflows, and they can coordinate up to 16 workers at a time, up to a thousand across one big job (docs).

Two ways to use it:

• Type /effort ultracode in Claude Code, and the whole session works this way: maximum reasoning, and Claude decides on its own when a task deserves a team.
• Or just include the word ultracode in a single request, and that one task gets the full treatment.

Worth being precise about: ultracode is a Claude Code feature, not a separate model and not an API setting. And it is not exclusive to Fable 5. It works on Claude's recent top models too. The reason the combo matters is simple: the workers are only as good as the model running them, and Fable 5 is the strongest model Anthropic has shipped (effort docs).

When it is worth it, and when it is overkill

Anthropic itself is direct about the tradeoff: workflow runs use a lot more compute, take longer, and are the wrong tool for routine work (docs). Our rule from daily use:

Use the combo when the task is wide. Research that should check many sources. A review across many files or documents. Anything where you would naturally say "look at all of these and tell me what matters."

Skip it when the task is deep but narrow. Drafting one email, fixing one document, answering one question. The regular model does that faster and cheaper.

A useful test: if you would have split the work across several assistants, it is a workflow task. If you would have done it in one sitting, it is not.

Three prompts that earn the cost

These are patterns we actually run, trimmed to their working core. Adapt the nouns to your business.

1. The market sweep. "ultracode: Research how businesses like mine are using AI for [your function]. Check vendor sites, news, and community discussions separately, cross-check claims against each other, and give me a short report with sources and an honest section on what is hype."

2. The document audit. "ultracode: Here is a folder of [contracts, listings, reports]. Review each one against this checklist, flag anything that fails, and summarize the patterns across all of them, worst problems first."

3. The competitor map. "ultracode: Look at these five competitor websites. For each, capture offer, pricing, and positioning. Then compare them against ours and tell me where we are clearly different and where we are invisible."

The shared trick: each prompt names the fan-out (many sources, many documents, many sites), asks for cross-checking, and demands sources. That is the shape that makes a team beat a single assistant.

The part we tell clients

This power is also the reason for care. A workflow that can touch many files or take many actions needs the same guardrails we wrote about with AI agents and trading: give it a sandbox, start it on copies rather than originals, and review before anything irreversible happens. Capability has outrun habits. The operators who win with this are the ones who build the habits early.

We run these workflows daily in our own practice, including to produce the research behind this article. If you want help figuring out whether your work has the wide tasks this combo was built for, that is exactly the kind of mapping we do with operators.