The human in the loop - and why writers won't be replaced anytime soon

Ever watched a movie and then think to yourself it was average, then you go onto IMDB or Rotten Tomatoes or social media to find that it got flying reviews and people are raving about it.

Then wonder to yourself if you missed something?

Just because people with average taste rate an average movie highly doesn't make it a good movie.

The same thing goes for content...

Everyone has access to the same models and everyone is running roughly the same prompts. The drafts are landing at roughly the same quality. If you are waiting for AI to be the thing that sets your content apart, you are going to be waiting a long long time, because your competitors are using the same tool with the same inputs and getting the same output.

Wide availability changed what differentiation looks like. What separates genuinely good content from the beige sprawl is the person deciding what good looks like.

Taste and judgement. The things you cannot train from an average of everything ever published on the internet.

What AI is good at

AI is extraordinary at speed. It compresses the grinding, structural work of content production - research, outlining, drafting, repurposing - into minutes instead of hours. That and any team that hasn't figured this out yet is spending time it doesn't have.

Speed gets you to a draft faster. Getting to great is still a human job, and it always will be, because great requires the ability to say "this fucking sucks" or "no - that won't do" or "this sounds like the forty-seventh version of this article I read this week" - and then fix it, or simply throw it in the bin and burn it. That's a judgement call, and judgement is not a feature you can add to a large language model.

LLMs are trained on the global average. They are, by design, a distillation of what already exists. When you ask a model to write something insightful, it produces something that resembles insight - shaped from everything it has read. The output is often technically competent and completely forgettable, because it reflects the middle of the distribution. It gives you the consensus, which is nobody's competitive advantage.

The first mile and the last mile is where it counts

You have seen this if you have spent any time reviewing AI-generated content at scale. The draft comes back. It hits the brief. The keywords are in the right places and the wordcount is expensive, the structure is on point and the tone is fine (if you ran it through a well-set-up knowledge base). And it is completely devoid of the kind of specific, lived, particular observation that makes a reader think "yes, that is exactly it." It can't reference an observation you had in a coffee shop earlier, or land the emotion of frustration with a disco-ball logo that just dropped from Spotify. It just doesn't know...

That specificity comes from the person who has done the work, run the campaigns, lost the pitch, sat in the room, and knows how it really goes. It comes from inside me, and inside you. Experience of having spent a decade in content operations and knowing that the real failure mode in content teams is taste, not output volume.

The last mile is where a good writer looks at the draft and spots the sentence that is technically correct but sounds like nobody ever said it. Where they pull in the specific detail that makes the argument land. Where they cut the paragraph that the model loves but makes the reader vomit and hit unsubscribe. That editorial instinct is the skill, now more than ever, because the part before it has been automated.

Good writers are now expensive in the right way

High-judgment writing is now more valuable than it has ever been.

If you were being paid to produce volume - ten blog posts a month, a steady stream of social copy, newsletter drafts - then yes, that specific function has been compressed. A single person with the right setup can now do what a small team used to do. That is just true - sorry, dude.

But if you are the person who knows when something is good, who can hear a brand voice and reproduce it faithfully, who reads a draft and immediately knows the paragraph that needs to go - that person is now the critical node, because the rest of the pipeline has been automated around them.

The brands producing genuinely good AI-assisted content have at least one person like that in the loop, and their job is to make sure everything that goes out is worth reading.

Taste is not teachable through a prompt

Taste is artistic, and it cannot be taught. You either have it or you don't. It's like good music taste, or a good sense of humor. You can write very detailed instructions. You can build a knowledge base that captures your brand voice, your hard rules, your style preferences. Structured instructions have a ceiling, because taste is not a rule set. Taste is a developed sense built from experience, and structured instructions cannot replicate it - reading widely and writing enough to care deeply about the difference between shit, okay and great.

Taste lives in the person reviewing the output and deciding whether it is ready, and that judgement is either sharp or it is not.

For any team producing content at scale with AI: the person with the editorial eye - the writer who has done enough of this to know what great feels like and cares enough to get it there and burn the rest - is the one role the model cannot absorb.

Which means the smartest thing you can do is make sure that human is in the room, and that they have enough space to actually do the job.