What happens to craft when nobody says it isn't good yet.

Unearned yeses, human and machine.

On January 21 Jaden Smith unveiled his collection as men’s creative director of Christian Louboutin. He had painted his face and torso the house red for the campaign. The collection built around the working man, an homage, he said, to stonemasons, scribes and doctors, to people who build something through discipline and effort over time.

The role had not existed four months earlier.In the 15 years the men's line has existed, no one but Christian Louboutin himself had ever directed it. The house invented the title in September, announced Smith, and by Paris Fashion Week he was presenting waterproof tactical loafers and silver-capped cowboy boots. The reviews and feedback on internet were loudly critical, although the reviews by press were warmer. When his second collection showed in June, the reaction ran the same script, louder.

And the script is always the same. Did he earn it. Would he be here without the name. Is the talent real or is it all access. This is the fairness question and it is the question every nepo baby argument has been stuck on since the term was made popular in 2022. Fairness is a fine question but it has no available answer. You cannot know for fact how Jaden’s career would evolve had he grew up in Ohio with different parents, so every verdict about what he deserved ends in speculation, which is why the argument never ends.

But fairness has a second flaw, and this one matters more. Fairness only measures access. It asks how someone got in and usually that is where questions end. Both sides of the debate agree the nepo baby won something and fight about whether the win was deserved. Ask a different question and the whole discourse inverts: what did they lose on the way in?

The answer comes from the science of getting good at things. Anders Ericsson, the psychologist whose research on expertise underpins most of what we know about skill acquisition, spent decades studying how people become excellent at hard things. His model has three requirements: a well-defined task, the opportunity to repeat it, and informative feedback you can use to correct the error. Remove the third and the other two stop working. The mistake has to land as a mistake, has to arrive as information, or the loop never closes and the skill never forms. Feedback in this sense is not encouragement. It is the signal that tells you where you actually are instead of where you think you are, and it costs the giver something to deliver and the receiver something to hear.

Now watch what nepotism does to that loop. Everyone reads inherited access as a head start, and it is. But it is also a severing mechanism for a person wanting a career in art, entertainment, creative industries in general. The collaborators are too well paid, or too dazzled to say the work is thin. The project ships regardless, because the deal was signed on the name, not the output. The press shows up because coverage was guaranteed before the work existed. And the family money can absorb the cost of every miss, so the miss never reaches its maker as consequence. The nepo baby ends up the best-resourced and worst-informed person in the room. Nobody in their orbit has a reason to tell them the truth to their face, and expertise cannot form in the absence of truth.

This is the mechanism the fairness debate keeps missing, and it explains the cases the fairness debate cannot. Think about talented people who happen to have famous parents and without a doubt that gave them a head start. But I highly doubt their talent now could be associated to closeness to the privilege only: Jamie Lee Curtis, Zoë Kravitz, Maya Rudolph and more. The name may have opened the door but the practice walked through it. I would call it the feedback loop in action.

Which brings us to the uncomfortable part, because the severed loop is no longer an inheritance. It is an AI product. Here goes the disclaimer that many things could be done with AI when it comes to automation, data triaging, research and so on, but my focus in this piece is on genuinely original creative work, new unique business creation, distinctive world-building, real imagination development. In other words, my focus is on creating work that will take more time and effort than a few prompts and a workflow creation.

For most of history, the only people who got to skip honest feedback were born past it. Now you are told you can create potentially anything from apps to art. The LLMs that a growing share of creative work passes through are documented flatterers. A Stanford and Carnegie Mellon study published in Science this spring tested eleven leading language models and found they affirmed users roughly 49% more often than humans do, endorsing the user’s position even in scenarios involving clearly harmful behavior. Anthropic researchers identified the root cause back in 2023: models trained on human preference ratings learn that people reward agreement, so agreement is what they produce. The most telling finding in the Stanford work is not about the machines at all. Participants rated the sycophantic models (AI systems that prioritize flattery and validation over truth and objectivity) as more trustworthy and said they would return to them, even as those models made them more convinced of their own rightness and less willing to repair conflicts. When OpenAI shipped a GPT-4o update in 2025 so aggressively flattering it had to be rolled back within days, users had, by the company’s own accounting, been rating the flattery up.

The flattery is only one layer. Recommendation feeds surface the audience most likely to approve. The negative signal, the one Ericsson’s entire model depends on, gets filtered upstream at every stage, so it never arrives as information back to the person who actually needs feedback.

You can now produce a song, a film, a brand, or a book without ever sitting in a room where someone tells you it isn’t good yet. The obvious workaround is to ask the model for brutal honesty, and the models will oblige, for a while. But criticism you commission is criticism you control, and the research shows models progressively cave when users push back. Feedback that softens the moment you argue with it is not feedback. It is a simulation of the loop with the consequence removed, adjustable by the same person whose judgment it was supposed to correct.

I also wonder what it takes to create objectively good work. One component is apprenticeship. We are loosing the idea of doing the same thing over and over again to build the muscle and taste of what good looks like. Yes, you can ship anything within a few sentences you type into a LLM but should you? What value would it have for you or for your customers? Would it have any value a year from now?

Scarcity did the same work by force. When a career offers three real shots, every rejection has to carry maximum information, so you read each one like scripture. The person with unlimited shots learns nothing from any single one. Abundance breaks the loop that scarcity closes.

If Ericsson is right that the painful loop is the only thing that ever made anyone good, then a culture that engineers the loop out of every surface it touches should expect a very specific output: enormous volume, high confidence, and craft that stops compounding. Talent is still there but a mistake you never feel is a mistake you never fix, and we are getting extraordinarily efficient at not feeling them.

The strange result is that unfiltered truth is becoming something you have to be either lucky or disciplined to encounter at all.

Always have people around who could tell you without any hesitation that the work is not great yet.