What AI actually replaces in a branding project (and what it doesn't)
Every third discovery call now opens the same way: “Before we start — could we just do this with AI?”
It is a fair question and it deserves a straight answer instead of a defensive one. We use AI daily. It has genuinely removed work from our process. It has also created a category of expensive mistake that did not exist three years ago, and the founders who hit it usually do not find out until a lawyer or an investor asks a question nobody prepared for.
Here is the honest split.
What it genuinely replaces
The first hundred bad options. Exploration used to mean a designer producing twenty directions over three days, of which eighteen were dead ends they had to draw to know were dead ends. Generative tools produce that spread in an afternoon. The dead ends are still dead, but nobody paid for the hours it took to prove it.
Moodboarding and reference hunting. Assembling a visual field for a brand — the textures, the type pairings, the adjacent industries worth stealing from — used to be a day of Pinterest and Behance. It is now a conversation. This is a real saving and we pass it on.
Asset variation. Once a system exists, producing the forty-two sizes, the dark-mode variants, the localized lockups, the social crops is mechanical work. Automating it is uncontroversial and every studio doing it manually in 2026 is charging you for typing.
First-draft copy. Not the positioning line — the connective tissue. Alt text, feature blurbs, the microcopy in empty states. AI writes a serviceable first pass that a human then makes actually good, and that is faster than a human staring at a blank field.
Together this is not nothing. It is a meaningful share of what a brand project used to cost, and any studio claiming otherwise is protecting a margin rather than telling you the truth.
What it does not replace
The decision. A hundred logo directions is not a hundred times more useful than five. It is worse, because choosing is the hard part and volume makes choosing harder. The value in a brand engagement was never generating options — it was someone with judgement saying this one, and here is why, and here is what we are giving up by choosing it. Nothing about that has changed. If anything, the flood makes the judgement more valuable.
The constraint. Good identity work comes from knowing what the brand cannot do. Not “calm, modern, trustworthy” — every SaaS brand claims those. The useful constraints are specific and uncomfortable: this brand never uses stock photography of people, because our buyers have been lied to by that exact image for a decade. Constraints like that come from the discovery interviews, from listening to how a founder talks about their competitors when the recording stops. A model cannot attend that conversation.
Consistency over time. A brand is not a set of files. It is the same decision, made the same way, by different people, for years. That is what a design system and documented usage rules exist to enforce. Generating a beautiful set of screens does not produce it, and neither does generating a beautiful logo.
Being wrong in public. When a rebrand lands badly, someone has to sit with the founder and work out whether to hold the line or move. That accountability is the actual product in a brand engagement. It is not transferable to a tool.
The gap most founders find late
This is the part that rarely comes up until it matters.
In the United States, the Copyright Office has held consistently that material generated purely by AI, without sufficient human authorship, is not eligible for copyright protection. Human authorship is the requirement, and prompting alone has not been accepted as enough. Other jurisdictions have landed in similar or stricter places.
Trademark is a separate regime and works differently: it protects a mark as an identifier of source, not as a piece of authorship, so an AI-assisted logo can generally still be registered if it is distinctive and actually used in commerce. That is the reassuring half.
The uncomfortable half is that trademark only stops someone using a confusingly similar mark in your category. Copyright is what stops someone lifting the artwork itself and putting it on a t-shirt, or on a product that competes with nothing you sell. If the artwork has no copyright because no human meaningfully authored it, that second layer may simply not be there.
For most seed-stage companies this is theoretical. It stops being theoretical during diligence, when someone asks what the company actually owns, and the honest answer is “a PNG”. We have watched that conversation happen. It is not fatal, but it is expensive and badly timed.
Two practical notes, and this is not legal advice — the specifics differ by jurisdiction and are worth twenty minutes with counsel before a raise:
- Keep the record of human authorship. Iteration history, the decisions and the reasoning, the redraws. It is the difference between “generated” and “authored with assistance”.
- Get final marks rebuilt as real vector artwork by a person, with the construction documented. This is also just good practice — generated logos tend to have inconsistent optical spacing and curves that fall apart at favicon size.
What we actually do
We use AI in the exploration phase and say so on the invoice. It shortens discovery-to-first-directions from about two weeks to about one. What we do not do is ship a generated mark. Every final logo gets redrawn — corrected optical spacing, real curves, a documented construction grid — because that is what survives contact with a favicon, an embroidered polo, and a 4-metre conference banner. The five-phase logo process has not changed; two of the phases just got faster.
Our prices are public and they came down when this work got faster. That is the whole argument: if a tool removes work, the saving belongs to the client, not to the studio’s margin.
The test worth running
Before paying anyone — us, another studio, or a subscription — ask for the reasoning behind one specific decision. Why this typeface and not the obvious competitor. Why this colour when the category is saturated with it. What the brand deliberately gives up.
An answer that names a trade-off is the thing you are buying. An answer that describes the aesthetic is a description of the picture, which you can now generate for free.
That question sorts the market faster than any portfolio review, and it does not care whether the answer comes from a person or a prompt.
If you want a second opinion on brand work you have already paid for — generated or not — send it over and we will record a 15-minute Loom: what is working, what will break, what would need to change. Free, no follow-up sequence. If you are starting from nothing, the brand identity engagement is the place to look.