Build Your Brand System in an Afternoon, No Agency Required

TLDR: A Substack writer at AI Adopters Club built a structured AI skill that produces a full brand strategy artifact — discovery interviews, token architecture, live HTML proof, and designer briefs — in one session. What agencies charge $25,000 and three months to produce, you can now run yourself on a Sunday afternoon. The trick is directing AI to become expert on brand strategy methodology so you can do the orchestration and judgment on top.

Summary:

Let me paint a picture most founders know too well. You have three browser tabs open: your personal site (dark, editorial, built by a designer in 2022), the product you launched last year (bright, cheerful, completely different palette), and your newsletter running on Substack defaults you never actually chose. Three properties, all yours, none of them looking like they belong to the same person. You've been quoted $75,000 by a full agency. The boutique wants $12,000. Neither fits the calendar. So nothing happens, and the brand fragments a little more each week.

The author of this piece, writing from AI Adopters Club, decided to build a real fix. Not a Pinterest mood board, not a PDF with some vague direction words, but a proper brand strategy artifact — the kind Pentagram or Wolff Olins produces for major clients. The key insight is interesting and worth sitting with: you don't have to personally master every adjacent discipline to produce work in it. Instead, you direct AI to become expert on the thing. In this case, that meant having the AI study Wally Olins on brand architecture, David Aaker's Brand Relationship Spectrum, the Pentagram MIT Media Lab precedent, and Wolff Olins case studies on Tate and Uber. The human does the orchestration and judgment on top of what the AI surfaces.

The resulting skill runs in Claude Code or a browser-based Claude environment, and it does serious work. It starts with up to 41 structured discovery questions across six rounds — and it refuses to synthesize anything until you've answered at least ten honestly. That's the right call. Generic AI brand output almost always comes from skipping the interview phase, so making it non-optional is a genuine design decision. From there, it visually audits your live properties by screenshotting them at various breakpoints and extracting computed color, font, and typography tokens via JavaScript. A separate subagent downloads your canonical logos, fonts, and favicons, and flags which properties are missing proper Open Graph images.

The core output is a three-tier token system: Tier 1 covers family DNA (typography, spacing, grid, radius, motion, the inviolable stuff shared across all your entities), Tier 2 maps those primitives to per-entity semantic roles, and Tier 3 handles component-level tokens derived from Tier 2. It also produces an architecture decision — monolithic, endorsed, or house of brands — argued with specific rationale rather than just listing your options and leaving you to guess. Then comes a live HTML showcase where you click between your entities and watch the same components re-skin themselves via token swap. That demo alone is worth something: it's the proof that convinces your team the system is real, and it's what you hand to a contractor so they start real work on day one instead of spending their first two weeks extracting from you what you already know.

The author frames four concrete payoffs. Inconsistency between properties creates a trust tax on every first impression — a visitor who clicks from your newsletter to your product site and lands somewhere that feels disconnected pays that toll subconsciously, and it compounds. Brand coherence across properties makes marketing spend compound rather than scatter. The structured discovery artifact eliminates the expensive early weeks of a design engagement, the phase agencies bill at full rate while they figure out who you are. And the open questions file the skill produces at the end names every decision it couldn't make for you, giving you a clear decision surface instead of vague "we'll figure that out later."

Key takeaways:

AI brand work produces generic output when you skip discovery; this skill enforces the interview before any synthesis
The three-tier token system (DNA / per-entity expression / component tokens) is the same structure a senior design system architect would spend a month building manually
Directing AI to master a discipline you don't know yourself, then doing the orchestration on top, is a repeatable pattern for other domains beyond branding
The live HTML showcase serves as proof-of-concept and a contractor handoff artifact in one
A proper brand architecture decision (monolithic vs. endorsed vs. house of brands) needs argued rationale, not just a list of options
The open-questions.md output is itself valuable — it names exactly what's unresolved so you can decide explicitly

Why do I care:

As someone who has watched design systems get built (and torn apart) in large frontend codebases, the token architecture described here is legitimately correct. The separation of primitive tokens, semantic tokens, and component tokens is exactly how mature design system teams structure things — and it's genuinely hard to get right without experience. The interesting thing about this skill is not the AI part, it's that it encodes institutional knowledge about brand strategy methodology into a repeatable workflow. Most developers and architects working on their own projects have fragmented personal brands for the same reason they have fragmented side projects: no time and no clear starting point. A structured artifact that forces you to answer the hard questions first, then derives the technical outputs from those answers, is how you break that paralysis. The designer handoff brief alone would save real money on any engagement.

How to Build Your Brand System in 45 Minutes, Without the Agency