For most of my career, "new tool" meant a new place to draw rectangles. Claude Fable 5 is the first tool I've adopted that argues back. After a month of daily pairing — design crits, component code, accessibility passes, naming debates — I have notes. They're optimistic, but they're earned, so let me show the work.
Where it's genuinely brilliant
The blank page is gone. Every project has a cold-start phase: the empty Figma frame, the unscaffolded repo, the "what do we even call this feature" meeting. Fable 5 collapses that phase from days to minutes. I describe the product, the audience and the constraints, and I get back five framings I can disagree with. Disagreeing with something concrete is ten times faster than agreeing with nothing.
It's the best accessibility reviewer I've worked with. I've spent years doing WCAG work, and I still miss things — a focus trap here, an aria-label that lies there. The model doesn't get bored on the fortieth component, and boredom is where most accessibility bugs are born. I paste a component, ask "what breaks for a keyboard user?", and it finds the thing I would have found in QA, three weeks later, with a stakeholder watching.
It refactors without ego. Handing over a five-year-old SCSS file and asking for modern, token-driven CSS used to be a week of careful archaeology. Now it's an afternoon of review. The model doesn't defend the old code, because it didn't write it. Honestly, neither did I — past me did, and past me had questionable opinions about specificity.
Where taste still decides
Here's the part the doom headlines skip: the better the model gets, the more valuable judgment becomes, because judgment is now the bottleneck.
Fable 5 will happily generate four plausible directions for a dashboard. All four will be competent. None of them knows that your users are field engineers wearing gloves, that the CFO judges the product by one specific chart, or that the brand died a little the last time someone used a gradient. Plausible is the model's ceiling and the designer's floor.
The model produces options. The designer produces the opinion. Products are made of opinions.
Restraint is the other human monopoly. The model, asked for a hero section, will give you a hero section — it will not say "this page shouldn't exist, merge it with settings." Knowing what not to build remains the most senior skill in the room, and a month of pairing has only sharpened my conviction that it doesn't transfer.
What changed in my process
- I write briefs again. To get good output you must articulate constraints precisely. This discipline leaked back into my human collaborations — my tickets got better because my prompts did.
- Crit comes earlier. I used to polish before showing anyone. Now the first crit happens at minute five, against the model, before ego attaches to the work.
- I ship the unglamorous stuff. Empty states, error messages, loading skeletons — the corners that always got "TODO" now get done, because the marginal cost dropped to almost nothing.
The honest scorecard
Out of roughly sixty design tasks this month, I'd say the model fully handled about a third, accelerated another half, and actively misled me on a handful — always confidently, which is exactly why review isn't optional. That ratio would have sounded like science fiction to 2017 me, hand-coding AngularJS directives at midnight.
The fear was that tools like this make designers interchangeable. My month says the opposite: it makes the average output interchangeable, which makes the differentiated output — the opinionated, context-soaked, restrained work — stand out more than ever. The floor rose. The ceiling is still ours.
I'm keeping the rule. Day thirty-one starts with a conversation too.