What's the bravest thing a designer can do right now? Apparently, it's refusing to finish.

4Creative — Channel 4's in-house agency — just unveiled a rebrand built around a logo that's deliberately, permanently unfinished. Raw, unpolished, and openly rejecting convention. And it's not alone. This week also brought Anchor's beautifully restrained butter rebrand, and a very public reminder from Nintendo fans that adding more detail to a beloved character is a fast way to lose what made them special in the first place.

Three stories. One shared lesson. And it's a lesson that AI tools are fundamentally bad at learning.

What happened

4Creative's new identity leans hard into imperfection. The logo isn't meant to arrive at a final state — it's designed to feel like it's always mid-process, always evolving. It's a deliberate aesthetic choice that says "we're not done, and that's the point."

Meanwhile, Anchor took the opposite route to the same destination. Their rebrand is so subtle you might miss it — refined typography, cleaner packaging, gentler colour shifts. Nothing revolutionary. Everything intentional. It's a masterclass in knowing what not to change.

And then there's Star Fox. Nintendo revealed a redesigned character that fans immediately rejected as overdesigned — too much detail, too many flourishes, too far from the simple silhouette people loved. The backlash was swift and loud.

Three very different projects. But they all point to the same truth: the hardest, most valuable design skill right now is restraint.

The bigger picture

Here's the thing — restraint is exactly what AI struggles with most.

If you've spent any time with generative AI tools, you'll have noticed their default mode is more. More detail. More polish. More completeness. Ask an image generator for a logo and you'll get something slick, rendered, and thoroughly finished. Ask an AI writing tool to iterate on your copy and it'll smooth every rough edge until it reads like a press release.

AI doesn't know how to leave things unfinished on purpose. It doesn't understand that a rough edge might be the entire point. It can't look at a beloved character's silhouette and feel, instinctively, that the charm lives in what's missing.

This isn't a small limitation. It's a fundamental one.

So much of what makes design genuinely good — the kind of good that 4Creative is reaching for, that Anchor nailed, that Nintendo accidentally violated — comes from subtraction. From knowing when to stop. From looking at a perfectly rendered output and choosing to strip half of it away.

We talk a lot about AI replacing design tasks. But we don't talk enough about how AI's defaults actively push against the kind of editorial judgement that separates competent design from great design. Every time you accept a polished AI output without questioning whether it should be less polished, you're letting the tool make a creative decision for you.

And it's a decision that almost always lands on "more."

Tool spotlight

This is less about a specific tool and more about a technique I've found genuinely useful.

If you're using AI in your branding or identity work — whether that's Midjourney, Firefly, or any of the current crop — try using it as a subtraction engine rather than a generation engine. Here's what I mean:

Generate ten variations of a concept. Then, instead of picking the most polished one, study them all and ask: what's common across every version? That's probably the core idea. Now strip everything else.

Use the AI output not as a destination but as raw material for editing. The value isn't in what AI adds — it's in what the volume of output reveals about what's essential and what's noise.

It's a small shift in workflow, but it completely changes your relationship with the tool. You stop being a curator of AI's taste and start being an editor of your own.

The takeaway

The best design work landing right now is defined by what's been left out, left rough, or deliberately left alone. AI tools will keep defaulting to polish and completion — that's how they're built.

Your job is to be the one who knows when to stop.

Next time you're working with an AI-generated output that looks clean and finished, ask yourself one question: would this be better if it were less? If the answer is yes — and honestly, it often is — you've just done something no AI can do for you.

That instinct for restraint? That's the craft. Protect it.