Medium: Testing Claude Fable 5 on real UX/UI design projects
Michal Malewicz, designer and founder of Hype4Academy, published this article in June 2026 after running Claude Fable 5 through three actual client projects — two focused on pure UI work and one involving a more complex, multi-step UX platform. The piece is a direct challenge to the way AI design tools are typically evaluated: optimized prompts, hand-picked showcases, and one-shot generative landing pages that look strong on first glance.
What the test involved
Malewicz chose to work with projects over two years old specifically to account for AI’s knowledge lag. He notes that the models average internet data from roughly one to two years prior, which means using current projects would create an unfair advantage for human designers who have access to recent reference material. Working with older briefs removes that variable and gives a cleaner read on what the tool can actually do in a professional context.
The three projects were used without prompt optimization. Each was treated as a real commission: given a brief, expected to produce output that would hold up to client review, not just screenshot well.
What the results showed
The main finding is that AI has a legitimate role in professional design workflows, but that role is probably not what most demonstrations suggest. Generative landing pages — the showpiece of most AI design tool videos — often look convincing in isolation and fall apart under closer inspection. The tasks where AI contributed more reliably were specific and contained, not open-ended generation from scratch.
Malewicz stops short of listing every finding in granular detail, but the pattern across the two UI projects and the UX platform was consistent: AI is more useful as a contributor to specific workflow stages than as a replacement for the judgment that stitches those stages together.
Why this framing matters
Most AI design tool content is produced either by the tool’s makers or by creators with an incentive to demonstrate impressive results. Testing on real projects with real constraints — including outdated briefs, messy requirements, and the kind of ambiguity that professional design always involves — produces different conclusions. The article is valuable precisely because it refuses to flatten that complexity into a before-and-after comparison.
Who this is useful for
Designers who have tried AI tools in controlled settings and found the results underwhelming in actual production, as well as those deciding whether to incorporate AI into client work rather than internal prototyping. It also provides a realistic check for teams discussing what AI can replace versus what it can assist, particularly in studios where the pressure to appear up to date with AI tools is running ahead of the evidence for where those tools genuinely help.