By Hassan Ezz, Engineering Lead 8 min read
For a while the conversation about generative tools was stuck between two bad answers: that they would replace us, or that they were a toy beneath us. Neither held up in practice. What actually happened in our studio was quieter and more interesting — the work did not disappear, it moved. The hours we used to spend making marks, we now spend making decisions.
The work moved, it did not vanish
Producing a first option used to be expensive: hours of drawing, rendering, comping. Now it is nearly free, and we can have forty options before lunch. But forty options is not progress — it is a new problem. The scarce resource is no longer the ability to produce; it is the judgment to choose, edit, and reject. The bottleneck moved upstream, toward taste.
This is not the first time. Photography did not end painting; it freed it from the job of recording and pushed it toward interpretation. Synthesizers did not end musicianship; they relocated it. Every tool that automates production raises the premium on the decisions production used to hide.
What the machine is good at
Used honestly, these tools are extraordinary at the early, divergent, low-stakes part of the work: exploring a space quickly, escaping a blank page, generating raw material to react against. They are a sparring partner, not an oracle. Their output is a question — "is it anything like this?" — and the answer is almost always "no, but now I know what I actually want."
Where we still draw the line
- Final craft — the last ten percent is human; the machine cannot sweat the details it cannot perceive.
- Meaning — a brand has to mean something to specific people; the model averages, and meaning is not an average.
- Accountability — someone has to be able to say why, and a prompt is not a reason.
- Attribution and consent — we do not pretend the provenance question is solved.
Taste is the bottleneck
The uncomfortable truth for our field is that generative tools reward the people who already had taste and punish the people who were coasting on production skill alone. If your value was "I can make the thing," that value is being commoditized. If your value was "I know which thing is worth making, and why," it just became more scarce, and more valuable.
So we train for judgment, not just output. We argue about why one option is better, not merely which looks nicer. We treat the machine as leverage on a point of view we are responsible for — never as a substitute for having one.
The machine can generate a thousand options in a second. Choosing the right one, and knowing why it is right, is still the entire job. That part is not going anywhere.
