What Happens to Craft When Everyone Has the Same Model

There's a predictable pattern to how new capability moves through an industry: exclusive at first, valuable because it's rare, then table stakes the moment it becomes available to everyone. Generative tools have moved through that arc faster than almost anything else in design tooling's history. Eighteen months ago, having a team fluent in these tools was a genuine edge - it meant faster exploration, more variations, cheaper first drafts than teams still doing everything by hand. That edge has mostly closed. The tools are the same tools, available to a competitor for the same subscription price, and access alone no longer separates anyone from anyone else.

This is worth sitting with because it's a bigger shift than 'the tools got cheaper.' When a capability is scarce, the strategic question is who has it. When it's universal, the strategic question changes entirely to who uses it well, and those are different games with different winners. A team that spent the last year building internal fluency with a generative tool now shares that fluency, roughly, with every competitor who bothered to try the same tool. The thing that took effort to acquire stopped being the thing that creates advantage almost as soon as everyone acquired it.

What's left, once raw access stops differentiating, is taste - and it's worth being precise about what that word is doing here, because it can sound like a vague compliment rather than a real skill. Taste is knowing which of thirty generated variations actually solves the brief and which twenty-nine are merely plausible. It's knowing when to combine two mediocre outputs into something better than either. It's knowing, sometimes, that none of the machine's suggestions are right and the correct move is to throw the batch out and start from a constraint the model was never given.

Picture two product teams given the identical prompt and the identical model to generate onboarding flow concepts. Both get back a dozen structurally similar options within minutes - that part is now commoditized and roughly equal across both teams. What happens next is where they diverge completely. One team ships whichever option looks cleanest in isolation. The other recognizes that all twelve share a flaw invisible to the model - they assume a first-time user already understands the product's core metaphor - and uses that recognition to brief a thirteenth version by hand. The tool was identical. The outcome wasn't.

This should be genuinely good news for designers who spent years building judgment rather than raw output speed, and it's worth saying plainly because the discourse around generative tools has mostly framed this moment as a threat to designers rather than a validation of a particular kind of designer. The people who are hardest to replace right now were never the fastest producers of pixels. They were the ones who could look at ten options and say precisely why nine of them were wrong - a skill that doesn't get commoditized just because the options got cheaper to generate.

The honest counterargument is that taste itself might eventually get automated too - that a model trained on enough critique, enough editorial decisions, enough examples of 'this one, not that one, because,' could start approximating judgment rather than just output. That's a real possibility worth taking seriously rather than dismissing. But it's a meaningfully harder problem than generating variations, because judgment is contextual in a way raw generation isn't - it depends on a specific brand's history, a specific audience's expectations, decisions nobody wrote down because they lived in a person's accumulated experience. That gap won't close as fast as the access gap did, and betting otherwise is premature.

There's also a version of this shift that looks like decline if you're only measuring the wrong thing. A junior designer whose main value used to be turning around fast, competent first drafts is watching that specific value proposition shrink in real time, and pretending otherwise doesn't help anyone plan their career. But the response to that isn't panic, it's redirection - toward developing the critique muscle earlier than that career path used to require, toward being the person in the room who can articulate why an option is wrong, not just produce more options faster than the person next to them.

In practice, this shows up as a change in how design leads spend review time. Teams that have absorbed this lesson are running longer, more deliberate critique sessions on generated work than they ever ran on hand-made work, precisely because volume went up and the bottleneck moved to selection. A design lead who used to spend fifteen minutes reviewing three hand-drawn concepts now spends forty minutes working through twenty generated ones - not because the review got sloppier, but because the sorting problem got bigger and now requires the kind of sustained judgment that used to be optional.

It's tempting to read 'invest more in critique' as a platitude, so it's worth being specific about what that investment actually looks like on a real team. It looks like protecting time for structured review instead of letting generation speed eat the schedule that critique used to occupy. It looks like senior designers narrating their reasoning out loud during review - not just picking the winner, but naming what made it the winner - so that judgment becomes something the team can learn rather than something locked in one person's head.

The teams that will end up ahead in a year aren't the ones with the best model access, because that variable is about to stop existing as a variable at all. They're the ones who treated this moment as a reason to get more rigorous about editorial judgment, not less - who used the extra volume generative tools produce as more raw material for taste to operate on, rather than as a replacement for having taste in the first place.

None of this means the tools don't matter. They compress time in ways that are genuinely valuable, and a team with no access at all is at a real disadvantage against one that has it. But 'has access' stopped being a sentence that finishes the thought. The sentence that matters now is longer, and harder to shortcut: has access, and knows what to do with what comes back. That second half was always the scarce half. It just used to be easier to hide behind the first.

So the honest way to close this out is with a reframe rather than a warning. Craft didn't disappear when everyone got the same model - it moved to a place that was always more defensible than tool access ever was, and it's now doing its work in plain view rather than getting credited to whichever team had the fanciest software. The differentiator was never going to be the tool. It was always going to be the person deciding what the tool was for.