Having a new model is one thing. Using it well is another. Anthropic just published a guide on working with Fable 5 in Claude Cowork — and the most interesting part isn’t technical. It’s a question of attitude.
Spell out less
The central advice: stop breaking tasks into little chunks and prompting Claude for each step. Fable 5 sustains long, multi-step work — even across days. Earlier models lost the thread over long stretches and needed reminding. Fable 5 plans the workflow up front and checks its results against that plan as it runs.
Anthropic compares it to working with a capable colleague: you explain the situation, you agree on what a strong final result looks like — then you let them work. Less of your time goes to checking each step, more of it to deciding what the work should be in the first place.
When to use Fable 5, and when not
Important: Fable 5 is not the default in Cowork — you have to select it. The default is Claude Sonnet 5, and for everyday tasks that’s the right call. Opus is a solid choice for deep work with a clear shape. You reserve Fable 5 for the complex, ambiguous projects that were out of reach before. It thinks longer and spends more of your limit — but that can be worth it on work that’s costly to get wrong.
Then there’s the effort setting: at higher effort, Fable 5 plans more before starting and checks in more along the way. In Anthropic’s testing, Fable 5 at low effort often matched or beat earlier models at their highest effort.
Context over a list of rules
The nicest idea in the post: constraints only tell Claude what not to do. Context tells it what the work is for — so it can make the right call when a situation comes up that your rules didn’t anticipate.
That’s why Anthropic recommends three ways to delegate: the approach (hand over the material, describe the outcome), the procedure (a skill teaches Claude your team’s process), and the timing (describe the outcome, and Claude turns it into a recurring scheduled task).
And one practical tip I’m taking with me: old skills and memory entries written for earlier models often carry corrections that model needed. With a new model, that baggage can hold it back. Anthropic’s suggestion: let Fable 5 audit itself — “Go through my skills and memory. Which still fit, and which were written for an older model?”
My take
For clauding.de this is home turf — this blog runs on a Cowork schedule itself. And the core matches my experience: the less I micromanage and the clearer I state the goal, the better the result. The hard part isn’t technical. It’s letting go.
The post, by the way, was written by Josefina Albert on Anthropic’s Education team — not marketing, an actual how-to. Worth reading if you use Cowork seriously.
Sources: