AX Design Starts in a Note App — Building an LLM Wiki With Obsidian

What AX design means

As models arrive that reason and create at something close to expert level, the definition of a designer's job moves. Craft with tools stops being the differentiator, and the work shifts toward putting AI at the center of the process and designing that system itself. Call it AX design.

AX design still serves humans. What changes is the designer's position: less the person drawing the output, more the person who architects the AI design system, configures it, and feeds it data. The industrial revolution produced industrial design, print produced graphic design, digital transformation produced UX design — the AI transition opens the next chapter.

The foundation is an LLM wiki

The starting point is not a flashy AI tool. It is turning the organization's own knowledge into an asset. However capable a model is, without your brand philosophy, design principles, project history and customer understanding, it produces work anyone could have produced.

That knowledge is usually scattered: instinct that lives in a designer's head, craft passed on verbally, decision rationale buried in Slack threads and inboxes. LLM wiki design is the work of organizing that into a form an AI can read and act on. It is not documentation for people; it is a knowledge base for agents to reference. The quality of the wiki sets the ceiling on the quality of the output.

Why Obsidian fits

Obsidian was built as a personal knowledge tool — a "second brain" note app. Its value is being reassessed in the AI era because the way it stores knowledge happens to match the way agents read it.

1. Local files make knowledge an asset

Every note is stored as an ordinary file in a folder on your machine, not on a vendor's server. Data locked inside a service is hard to call fully yours once policies change or the service shuts down; local files are not bound to any platform. Agents benefit too: a tool like Claude Code can read and write those files directly with no integration layer, so the vault becomes the agent's workspace and the wiki tool and the AI's stage collapse into one. Sensitive design assets also stay under the organization's control rather than on an outside server.

2. Markdown costs fewer tokens

Notes are stored as markdown — natural language with minimal structural markup. Models read text in tokens and pay in cost and latency for each one. Unlike Word or PDF files carrying formatting metadata, markdown is content only, so the same knowledge transfers in fewer tokens. That means faster responses, lower cost, and more context per request. Markdown is also the de facto format agents already use to exchange instruction files and outputs.

3. Links become traversal paths

Notes link to other notes, and the result visualizes as a graph. The value is not the picture; it is that an agent can follow those links. Starting from a brand principles document, an agent can move to a project retrospective, then to customer research notes, reading only the context it needs instead of scanning everything. As knowledge accumulates, the links thicken, and denser links produce deeper answers.

The first step available today

The transition starts small: write the team's design principles, what this project taught you, and the feedback that keeps recurring as markdown notes, and link them to each other. That accumulated data is what drives the agents, and the agents are what drive the design system.

On staffing and capability placement when adopting AI tools, see They Tried Hiring 'AI Product Builders' and Failed. On the limits coming from the input rather than the tool, see AI Isn't Blunt — I Was Getting Blunt.

FAQ