Volume is not the same as good
Every studio is "AI-native" now. It has become a badge you pin on, not a claim you have to earn. So here is the honest version, the one we actually believe:
AI is the best tool we have ever had for producing volume. It is also blind to whether any of that volume is good. Those two facts are not in tension. They are the whole story. A tool that can generate a hundred logos in a minute still cannot tell you which one is worth using. That part has not been automated. It may never be.
Why AI output drifts to the average
There is a structural reason AI-made work feels same-y, and it is not a bug that a better model fixes. These systems are trained on the enormous middle of everything that already exists, and they are very good at returning the center of that distribution. Ask for a startup logo and you get the average of ten thousand startup logos. Ask for a landing page and you get the shape every landing page already has.
The average is a fine place to start. It is a terrible place to stop. Taste is what pulls work away from the center: the decision to cut the obvious feature, to break the expected grid, to pick the one option in fifty that is sharp instead of safe. That decision is a judgment about what is good, and judgment is exactly what a model trained on the middle cannot supply.
Where we use AI, and where we refuse
We are not precious about tools. We use AI everywhere it moves us faster and nowhere it would flatten the work. In practice that line is clear.
We reach for AI on:
- Plumbing: boilerplate, scaffolding, the wiring nobody should hand-write in 2026.
- Volume on demand: first-pass variations to react to, research to compress, drafts to cut down.
- Speed: getting from blank page to something we can judge, faster.
We keep human hands on:
- The cut: what belongs in the product and what does not.
- The one right answer: the single layout, the single name, the single flow that is correct, chosen from everything the volume produced.
- The feeling: whether the thing has any, and whether it is the right one.
Speed from the machine. Judgment from us. The decisions stay human, because the decisions are the work.
Taste is the whole job
Our name is a small argument for this. Imli is tamarind: tangy, sharp, impossible to ignore. That is the standard for everything we make, and it is not a standard a model reaches for on its own, because it is not the average. It is a choice.
When AI can produce infinite competent work for free, competent stops being worth paying for. The only thing left with value is knowing which competent thing is actually good, and being willing to throw out the other ninety-nine. That is taste, it is a skill you can hire for, and it is the entire job.
That is how we work: brand, product, and AI automation, built with AI for speed and taste from us. If that is the standard you want on your work, book a call.