From design taste to professional judgment.
Eight chapters. Six domains. One universal pattern.
What if the patterns that teach AI your design taste could teach it your legal judgment, clinical instinct, or sales voice?
Chapters
An Interactive Framework
We built three systems that teach AI professional taste through collaboration. Now we ask: what if these patterns work everywhere?
The Thesis
Runtime Learning taught us that AI can learn one person's aesthetic through building together. Brand Studio proved that iterative artifact generation with feedback loops produces better creative work. Design Sprint showed that parallel multi-agent exploration surfaces ideas no single perspective would find.
The shared meta-pattern: Human specifies intent, AI explores, Human decides, AI refines, Portable artifacts exported.
This story maps that pattern to six professional domains where the same dynamics apply — where expertise is subjective, context-dependent, and impossible to capture in a static prompt.
What You'll Explore
Chapter 1
Three eras of AI alignment, the taste gap, and why parametric encoding changes everything.
Evolution
How we teach AI what humans want has gone through three distinct phases, each dramatically narrowing the gap between generic capability and personal relevance.
Learn everything from the internet. Broad capability, zero personalization. The model knows what the average of all humans wants.
Humans rate outputs. The model learns general preferences — helpful, harmless, honest. Better alignment, but still generic. One size fits all.
AI learns from ONE user through building together. Your feedback shapes a parametric taste profile that encodes your specific judgment, preferences, and style.
The Problem
Every professional has a concept of "thorough." But "thorough" means wildly different things in different domains. A single word carries an entire worldview of professional judgment — and no prompt can capture it.
The Solution
Instead of trying to describe your taste in words, encode it as numbers. A slider value between 0 and 1 captures more nuance than a paragraph of instructions — and it's machine-readable, composable, and portable.
"I want contracts that are thorough but not overly cautious. Flag major risks but don't bury the client in edge cases. Use plain language where possible."
The parametric profile isn't just more precise — it's learnable. Each time you approve, tweak, or reject an artifact, the numbers update. Over time, the profile converges on your actual taste, not your description of it.
Chapter 2
Two systems that proved the concept: Brand Studio for iterative refinement and Design Sprint for parallel exploration.
System 1
An iterative artifact generation engine with feedback loops. You write a brief describing your brand vision. The system generates seven distinct artifacts — logo, colors, typography, imagery, voice, layout, and motion. You approve, tweak, or regenerate each one. Your feedback trains the system on your taste.
System 2
A parallel multi-agent exploration engine. Four expert perspectives — Strategist, Designer, Technologist, User Advocate — each independently explore the problem space. Their concepts are evaluated against a rubric, scored, and the winning approach is built into a polished prototype.
Business model, market positioning, competitive analysis
Visual language, user experience, interaction patterns
Technical feasibility, architecture, performance constraints
User needs, accessibility, real-world usage scenarios
The Shared Pattern
Both systems share the same core loop — and this loop is what generalizes across every professional domain.
Chapter 3
The universal architecture: Brief, Artifacts, Feedback, Profile, Export — five components that adapt to any professional domain.
Component 1
Every domain starts with intent. The brief captures what you want, who it's for, what constraints matter, and what tone to strike. The fields change per domain, but the structure is universal.
Component 2
In Brand Studio, artifacts are logos, color palettes, and typography. The same pattern produces different artifacts in each domain — contracts and risk analyses in legal, treatment plans and evidence summaries in medical.
Component 3
Generation is only the beginning. The real power comes from the cycle: generate artifacts, evaluate them, update the taste profile, generate better artifacts. Each iteration makes the system more attuned to your specific judgment.
Component 4
Four expert perspectives per domain, each bringing a different lens. The roles change names but the archetypes are universal: the analyst, the creative, the pragmatist, and the advocate.
Deep domain knowledge. Evaluates correctness, compliance, and technical rigor. The one who knows the rules.
Divergent thinking. Explores unconventional approaches, finds novel framings. The one who breaks patterns.
Implementation focus. Ensures feasibility, timeline, resources. The one who ships.
Human impact focus. Represents the end user, patient, client, or student. The one who asks "who does this serve?"
Component 5
Every session produces exportable artifacts. Three formats serve different needs: machine-readable JSON for system integration, human-readable markdown for sharing, and domain-native formats for direct use in professional tools.
Machine-readable parametric values. Plugs directly into future sessions, team configs, or CLAUDE.md files.
Human-readable narrative of preferences, decisions, and rationale. Shareable with colleagues and stakeholders.
Contract templates, treatment protocols, sales playbooks, lesson plans — artifacts in the format professionals actually use.
Chapter 4
Where the stakes are highest: legal judgment and clinical instinct.
Every lawyer has a sense for risk. Some are aggressive, pushing boundaries for commercial advantage. Others are conservative, building fortresses of protective language. The system learns where you fall — and every clause it drafts reflects that judgment.
Identifies exposure points, rates severity, suggests mitigations calibrated to your risk appetite.
Marks problematic clauses, suggests alternative language matching your firm's preferred positions.
Prioritized list of negotiation points with fallback positions, ranked by your historical preferences.
Plain-language explanation of key terms and risks, calibrated to the client's sophistication level.
Drag the slider to see how contract language changes at different risk tolerance levels.
Medicine is probability. How much evidence is "enough" before recommending a treatment? How aggressively to treat? When to watch and wait? Every clinician calibrates differently — and the system learns your clinical instinct through each case.
Ranked list of possible conditions with probability estimates calibrated to your diagnostic style.
Relevant research filtered by recency and quality, depth matched to your evidence threshold preferences.
Stepwise recommendations with alternatives, calibrated to your balance of aggressive vs. conservative care.
Explanation of diagnosis and treatment in language matched to the patient's health literacy level.
Adjust to see how treatment recommendations shift with different evidence thresholds.
Chapter 5
Where persuasion meets authenticity: sales voice and brand tone.
Every salesperson has a voice. Some are consultative, building trust through deep understanding. Others are direct, driving urgency and action. The system learns your voice and generates outreach, proposals, and follow-ups that sound like you — not like a template.
Adjust the sliders to calibrate your sales voice. Watch the email preview update in real time.
Brand tone is the subtle art of saying the right thing the right way. Playful or authoritative? Aspirational or grounded? Technical or accessible? The system learns your brand's personality across multiple dimensions and generates consistent copy that never sounds off-brand.
Calibrate your brand's voice across multiple dimensions.
Chapter 6
Where human connection matters most: customer service and education.
The best support agents balance empathy with efficiency. They know when to slow down and listen, and when to cut to the solution. The system learns your team's voice and escalation instincts, generating responses that feel human while resolving issues fast.
Drag to see how support responses shift between empathetic acknowledgment and efficient resolution.
Teaching is an art of calibration. How much to scaffold versus challenge? When to explain versus let students discover? Whether to use examples, metaphors, or formal definitions. The system learns each educator's pedagogical instinct and generates materials that match their teaching philosophy.
Calibrate your teaching style to see how lesson materials adapt.
Chapter 7
Taste profiles become portable, shareable, and self-improving.
Per Domain
Each domain produces a taste profile that encodes professional judgment as parametric values. These profiles are the bridge between one session and the next — they carry your expertise forward.
| Domain | Profile Encodes | Example Parameters |
|---|---|---|
| Legal | Risk appetite, language preferences, clause priorities | risk_tolerance: 0.65 |
| Medical | Evidence thresholds, treatment aggressiveness, communication style | evidence_threshold: 0.80 |
| Sales | Voice formality, urgency calibration, technical depth | formality: 0.40 |
| Marketing | Brand tone, audience targeting, creative boundaries | playfulness: 0.70 |
| Service | Empathy-efficiency balance, escalation thresholds | empathy_weight: 0.60 |
| Education | Scaffolding level, abstraction preference, assessment style | scaffolding: 0.45 |
Interactive
Select a domain to see its taste profile structure with sample data. This is what gets exported, shared with teammates, and embedded in configuration files.
The Bigger Picture
Taste profiles don't just serve the individual. They feed a flywheel that makes every future session better: profiles improve with use, get shared across teams, embed in project configurations, and eventually create organizational knowledge that persists beyond any single person.
What started as teaching AI one designer's taste becomes a universal framework for encoding professional judgment. Six domains, one pattern, infinite applications. The future of AI isn't just capable — it's personally calibrated.