Habr AI shows how to build a startup brand platform in 8 hours without a designer

Habr AI has published a detailed breakdown of why a brand platform can be useful for a startup even before a full MVP. The author shows how to build, in eight hours, a system of text constraints from which the landing page, app screens and ad creatives are then derived, without constantly returning to a vague "make it look nice."

Why brand comes before the MVP

The article's main thesis is that, at an early stage, a brand is not a logo or a decorative shell. It works as a system of constraints: it sets the product's voice, the type of visual decisions, the promises it can make and the way it talks to its audience. Without this system, every new artifact, from a banner to a social media card, has to be generated from scratch. As a result, the model keeps landing on the same averaged, "expensive-looking" design that looks convincing but hits a specific user poorly.

The article demonstrates this with a test case: a PWA for overlanders traveling off-road in Russia, Mongolia and China. A quick landing page built in five minutes came out with a dark theme, invented metrics, two CTAs and fake social proof. The version made through the pipeline in eight hours was calmer but more honest: one CTA, real usage scenarios, and a visual focus on the terrain and the language of the audience. For the author, what matters is not an instant wow effect but the ability to keep applying the same approach to new screens and creatives.

"20 artifacts at 5 minutes each is 20 lotteries. 8 hours once is a system."

What the pipeline consists of

The author suggests starting not with Figma or with generating a "pretty landing page," but with a product ontology: what exactly the product changes in a person's relationship with their environment, what its archetype is, and where the line falls between usefulness and false promises. From this level the brand platform is assembled, then the rules of visual semantics, the identity system, tokens and only then the specific artifacts. In such a chain, every decision can be explained in words and checked against the overall logic.

• Product ontology and JTBD
• Brand platform with invariants
• Principles of visual semantics
• Visual identity, tokens and components
• Screens, landing page and creative specification

The key idea is that prompts stop being a set of taste-based wishes. Instead of the word "pretty," they contain concrete constraints: why a dark theme is not needed, why there is one CTA, how to show how old the data is, and when people in the frame are acceptable and when they are not. This approach is closer not to free-form generation but to constraint-driven production: a specification is set first, and interfaces, illustrations and marketing materials are then derived from it step by step. The more artifacts a product needs, the more worthwhile this preliminary layer becomes.

Where AI goes wrong

One of the most telling moments in the breakdown is an error by the model itself inside a carefully built pipeline. At one stage, the AI decided that visuals should contain no faces: the terrain supposedly had to remain the main subject of the frame, and the interface had to speak in the first person. On paper the explanation looked logical, but it contradicted what is known about the audience. For overlanders, trust in data is tied to its source, which means people in the frame are sometimes essential: the person who shows the road, discusses a river ford next to the vehicle or shares an observation is part of the product.

The author uses this episode not to criticize models but to draw a harsher conclusion: prompt engineering works as a multiplier, but it does not create domain knowledge out of thin air. The pipeline helps record decisions, check them for consistency and see where the AI went convincingly wrong. But the final check is still done by a human who understands the market, the audience and the context of use. That is why the article describes the brand platform not as a replacement for a designer or product manager, but as a way to put professional intuition into text and make it reproducible.

What this means

For solo founders and small teams, the takeaway is simple: before the landing page and banners, spend a few hours on a one-page constraint file with principles, anti-examples and the brand's voice. It will not replace testing hypotheses on the market, but it will cut the chaos in generations and help you assemble a coherent system faster instead of a set of random images.