Viking Design Framework
Good intent-first UX starts with thinking, not screens.
Viking is a design methodology for AI-native products. It gives your team a clear sequence from value assessment to engineering handoff, so you build the right thing before you build the thing.
The problem
Most design frameworks were built before AI existed.
They assume your product does what you tell it to do. AI products don't work that way. They fail in unexpected ways, require human judgment at odd moments, and need a different kind of design thinking to get right.
Teams jump straight to screens before asking whether AI is even the right tool for the problem.
Designers work on surfaces without mapping the logic underneath. The AI behavior surprises everyone.
Engineers get wireframes instead of contracts. What the AI should do, and when, gets lost in translation.
The methodology
Five steps. In order. Every time.
Viking is a sequence, not a toolkit. Each step builds on the last. You don't skip to screens until the thinking is done.
Before anything else, assess whether AI is the right tool. Five questions covering Value, Precision, Logic, Risk, and Context that save you from building the wrong thing entirely.
Do you actually need AI? → Fit assessmentMap your product's object model before touching the canvas. What are the things? What states can they be in? What rules govern them?
Logic mappingA seven-role skill chain that walks you from strategy to engineering handoff. Each role has a clear mandate and hands off to the next.
Design processThink procedural generation in video games. You define the rules and the system assembles the right UI for the moment. No fixed screens. Just surfaces that fit the context they appear in.
Interface designA quality audit to catch generic, AI-generated aesthetics before they ship. Clean typography and real design judgment. Not pattern soup.
Quality auditPrinciples
Three rules under every step.
When AI does work that matters, it shows its homework. The reasoning and the way to override it get designed alongside the action.
AI content always sits between two layers of fixed copy: a line that sets trust on top, a way to check or act on the bottom.
Fix the instance. Reflect the pattern. When AI fills a gap a person should have filled, the record gets fixed but the habit doesn't. Show the pattern to the person who owns it.
Why predictive UX needs a reflective mirror →The High Council
Seven roles. One sequence.
Each council member owns a step in the process. Work flows in one direction and nothing ships until every role has signed off.
Sets the creative mandate and business goals. Leads the 5 Gates review. The strategic lead and aesthetic guardian.
Runs the fit assessment. Audits mental models, user vocabulary, and whether the problem is worth solving.
Translates business goals into user opportunities. Defines the user hats and frames the design problem clearly.
Maps Things, States, and Rules. Defines where AI lives in the product and where humans stay in the loop.
Designs the interaction recipes. Builds flows that account for edge cases, error states, and AI surprises.
Writes the UX copy. Warm, direct, and clear. Uses the Sandwich Pattern to frame AI actions in human terms.
Delivers component contracts and design tokens to engineering. The work doesn't leave design until Thor says it's ready.
Worked example
One ask. Four jobs. Two of them aren't AI.
The ask: "Add AI to our expense tool so people can snap a receipt and be done." The 5 Gates split that into four jobs, and each one landed somewhere different.
Fuzzy, low risk, and easy to fix. Passes every gate. If the model isn't sure, the field stays empty instead of guessing.
AIA good AI job, except the total has to be exact. The model fills it in; the person confirms it, and it's matched against the card charge.
AI with a human check"Over $75 needs an itemized receipt" is a checklist. Fails the Logic gate. A rules engine does it every time, for free.
Rules, not AIThe data lives in each manager's spreadsheet. Fails the Context gate. It stays manual until the data moves.
Not yetEvery deliverable from all seven roles, from the brief to the QA checklist: read the full run on GitHub →
Intellectual foundation
Built on the right shoulders.
Every agent in the High Council draws from a shared canon of UX, design, and product thinking. Not as decoration, but as working constraints that shape what gets built and why.
Affordances, signifiers, feedback loops, and the seven stages of action. The vocabulary for how people understand and interact with systems.
The five-plane model (Strategy, Scope, Structure, Skeleton, Surface) that governs the council's work sequence. No surface before structure.
Self-evident design, scan-pattern thinking, and the ruthless omission of words. If it needs an explanation, reconsider the design.
Fitts, Hick, Miller, Peak-End, Zeigarnik, Doherty. Psychological laws applied as hard design constraints, not loose guidelines.
Hypothesis-driven design and lean UX research. Define success before you design. Segment ruthlessly. Ship the smallest thing that generates real signal.
Every design decision needs a rationale. What you made, why you made it, how it makes things better. "Looks clean" doesn't survive a stakeholder review.
The baseline quality review at every milestone. Visibility of status and user control are blockers on AI features. No exceptions.
POUR principles baked into every deliverable. Contrast, keyboard access, touch targets, and screen reader parity are specced from the start, not audited after ship.
Full reading list: kartooner.com/recommended-books →
Usage
Nine commands. Start anywhere.
Once installed, type these as slash commands at the AI prompt in Claude Code CLI or Gemini CLI. Viking automatically detects tools in your environment (Figma MCP, Pencil, and others) and incorporates them into the workflow.
/viking-design
Open the Discovery Map. Start here if you are new to the framework.
e.g. "I'm starting a new AI feature for our support tool. Where do I begin?"
/viking-design gate
Run the 5 Gates assessment. Determines whether AI is the right tool before you build anything.
e.g. "Should we use AI to auto-categorize incoming support tickets?"
/viking-design frame
Problem framing. Translate business direction into a design brief, define user hats, and size the opportunity.
e.g. "We want to reduce support ticket volume by 30%. Frame the design problem."
/viking-design map
Logic mapping. Walk through Things, States, and Rules for your product.
e.g. "Map the object model for a ticket that can be open, pending, escalated, or resolved."
Uses Miro MCP if available.
/viking-design shape
UI recipes. Build task flows and surfaces from the system map, with the unhappy paths designed first.
e.g. "Design the flow for when the AI isn't sure which team a ticket belongs to."
Uses Pencil or Figma MCP if available.
/viking-design content
UX copy. Write warm, direct interface copy that makes AI behavior clear and legible to real people.
e.g. "Write the copy for our AI triage confirmation screen."
/viking-design craft
Handoff specs. Component contracts, acceptance criteria, and QA checklists engineering can build from.
e.g. "Write the handoff spec for the triage review card."
Uses Figma MCP if available.
/viking-design chain
Activate the full Council chain. Runs all seven roles in sequence from strategy to handoff.
e.g. "Run the full design process for our AI inbox triage feature."
/viking-design audit
Run the AI Slop Test. Quality audit for aesthetics and design judgment before you ship.
e.g. "Review these screens before we hand off to engineering."
Uses Figma MCP, Pencil, or Playwright if available.
No CLI? Browse the agents and skills on GitHub and paste any prompt directly into Claude, Copilot, or ChatGPT.
Start with thinking. Ship with confidence.
Add Viking to your Claude Code or Gemini CLI environment. Works with any AI product team, any stack.
Built on Impeccable and Design with Intent. Original contributions by Erik Sagen. MIT licensed. Draws from Norman, Garrett, Krug, Yablonski, Klein, Greever & Nielsen. Full reading list →