I have spent the last year watching people argue about whether AI is coming for creative work. From where I sit — booth, studio, edit bay — that argument is already over. It arrived. The interesting question is what it actually took, and what it turned out it could not touch.

The numbers nobody argues with

Generative AI in the creative industries is on track to move from roughly $4.06 billion in 2025 to about $5.38 billion in 2026 — a 32.3% growth rate. That is not a trend. That is a re-pricing of an entire category of labour. And text-to-image still leads the pack, which tells you something about where the money went first: the parts of the job that were always the most repeatable. I run MLO Technologies, so I do not get to be romantic about this. I have to look at what a tool costs, what it saves, and whether the output survives contact with a client.

What it actually took from me

Not the voice. Not the idea. It took the tedium. Cleaning room tone. Generating forty layout variations so I can reject thirty-nine. Drafting the client email I have written four hundred times. Roughing out a storyboard so the director can tell me it is wrong. Across 60+ film credits and years in commercial design — including work for Pepsi — I can tell you that the majority of a creative career is not creating. It is preparing, revising, exporting, and re-exporting. AI ate that layer, and I am not mourning it.

What it did not take

The industry consensus for 2026 is shifting toward specialised, domain-specific systems over giant general-purpose models — and that shift exposes the limit. A model is a compression of what already exists. It is very good at the average of everything that has been done. It is structurally incapable of the thing you do when you are in the booth at 2am and you decide to read a line wrong on purpose because wrong is truer. When I voiced for Takki on Netflix, the direction that mattered was never technical. It was somebody saying: less. Again. Now mean it. No system predicts that, because that decision is not in the training data. It comes from a room, a risk, and a person willing to be embarrassed.

How I am actually using it

Simple rule at Raw Studios: AI handles volume, humans handle judgement. It drafts, we decide. It generates the ninety, we choose the one. The moment it starts deciding, the work goes flat — polished, competent, and completely forgettable. The creatives getting wiped out in 2026 are not the ones who refused to use AI. They are the ones whose entire value was volume and polish. If that was your moat, it is gone. If your moat was taste, judgement, and the nerve to make a call nobody asked for — you just got handed leverage.

The machine can make a thousand versions. It still cannot decide which one is worth your name.