24
Named capabilities
AI capability map
I thought I was prompting. I was building an AI-native way of working.
Across Green Filter, HAAM, Raw Drafts, Proxy Society, WCD 2026, AI Grant Finder, and personal data projects, the visible prompt was only one small part of the system.
6
Connected systems
1
Research-to-product loop
Prompting is the visible layer
The deeper work happens before, around, and after generation.
Before
Choose the problem, outcome, audience, evidence, risks, and constraints.
Around
Supply context, tools, sources, memory, permissions, and human approval.
After
Inspect quality, compare alternatives, test failure modes, revise, and ship.
Across time
Preserve what was learned so the next project starts with more intelligence.
System 01
Direction
Deciding what AI is for, what it should protect, and where it should stop.
01.1
AI problem framing and product judgment
01.1
AI problem framing and product judgment
What it means
Finding where AI could change an entire activity, then choosing whether it should act as a companion, researcher, matching engine, interface, agent, or automation layer.
Evidence in the work
Green Filter moved from sustainable shopping to financial wellbeing and later agentic commerce. HAAM ideas repeatedly become testable products instead of remaining feature lists.
01.2
Human-centered AI design
01.2
Human-centered AI design
What it means
Designing conversational behavior, trust, explainability, human control, handoffs, authority, disclosure, and cross-device interaction around the model.
Evidence in the work
Green Filter explored how a financial AI companion should speak, explain trade-offs, reduce anxiety, and move between web, Apple Watch, and browser-extension contexts.
01.3
Responsible AI and governance
01.3
Responsible AI and governance
What it means
Making permissions, consent, representation, uncertainty, escalation, accountability, and cultural differences visible in the experience.
Evidence in the work
Proxy Society treats agent etiquette and delegated authority as social design problems, not merely technical settings.
01.4
Decision intelligence
01.4
Decision intelligence
What it means
Using AI to expand the option space, expose assumptions, compare trade-offs, and build criteria without pretending the model should make the final call.
Evidence in the work
This pattern appears across career choices, product priorities, travel, education, collaborations, personal finance, and technology adoption.
System 02
Orchestration
Turning a conversation into a supervised system of models, tools, context, and review loops.
02.1
Prompt engineering
02.1
Prompt engineering
What it means
Refining instructions through constraints, examples, roles, sequencing, format rules, and repeated evaluation rather than searching for one magical prompt.
Evidence in the work
Directions such as “go deeper,” “too generic,” “not specific enough,” and “use my voice” function as compact evaluation signals that progressively shape the result.
02.2
Context engineering
02.2
Context engineering
What it means
Designing everything the AI needs around the immediate prompt: memory, source files, messages, project history, personal preferences, retrieval, and structured data.
Evidence in the work
The emerging personal context architecture connects Gmail, Drive, GitHub, data exports, screenshots, writing samples, project histories, health information, and travel plans.
02.3
Agent orchestration and delegation
02.3
Agent orchestration and delegation
What it means
Breaking outcomes into research, design, implementation, review, deployment, and follow-up while routing work across different tools and retaining approval gates.
Evidence in the work
HAAM work regularly coordinates ChatGPT, coding agents, GitHub, Vercel, email, Drive, image tools, web research, and domain-specific systems.
02.4
AI evaluation and quality control
02.4
AI evaluation and quality control
What it means
Checking coverage, evidence, taste, regressions, edge cases, failure modes, and whether a technically valid output is actually worth shipping.
Evidence in the work
The recurring loop is generate, inspect, reject, specify, regenerate, test, and publish. The idea that every coding agent needs a regression agent makes this explicit.
System 03
Research and data
Using AI to investigate, structure, compare, and turn messy evidence into something navigable.
03.1
AI-assisted research
03.1
AI-assisted research
What it means
Forming questions, refining queries, mapping landscapes, conducting company diligence, synthesizing sources, and turning discovery into action.
Evidence in the work
Research spans AI labs, Chinese apps, universities, technologies, grants, events, companies, collaborators, and professional opportunities.
03.2
Mixed-methods and data analysis
03.2
Mixed-methods and data analysis
What it means
Combining surveys, interviews, self-testing, prototype evaluation, segmentation, thematic synthesis, quantitative comparison, and iteration.
Evidence in the work
Green Filter included more than 900 survey responses, 675 valid responses, 32 interviews, 32 prototype tests across seven schools, and more than 100 self-tests.
03.3
Data curation and knowledge architecture
03.3
Data curation and knowledge architecture
What it means
Creating taxonomies, records, filters, relationships, metadata, directories, and lightweight knowledge graphs from fragmented information.
Evidence in the work
Examples include the 244-entry WCD 2026 directory, the AI-labs database, grant matching, app atlases, contact maps, and connected project archives.
03.4
Privacy, data rights, and personal AI infrastructure
03.4
Privacy, data rights, and personal AI infrastructure
What it means
Treating personal data exports, continuous access, ownership, provenance, and model permissions as part of the AI system rather than administrative leftovers.
Evidence in the work
The GDPR work is evolving toward a personally controlled data plane that could continuously supply context to a personal model.
System 04
Building and publishing
Converting research and decisions into deployed interfaces, tools, systems, and public evidence.
04.1
AI-assisted software development
04.1
AI-assisted software development
What it means
Directing implementation through behavioral, structural, interaction, and visual specifications, then debugging, reviewing, versioning, and deploying the result.
Evidence in the work
The portfolio includes Next.js sites, Three.js environments, Chrome extensions, TestFlight apps, interactive directories, GitHub workflows, and Vercel deployments.
04.2
Knowledge-product publishing
04.2
Knowledge-product publishing
What it means
Turning conversations and research into durable public artifacts that other people can navigate, use, critique, cite, or build upon.
Evidence in the work
Outputs become websites, directories, maps, essays, databases, proposals, case studies, tools, design systems, and searchable archives.
04.3
AI portfolio and ecosystem architecture
04.3
AI portfolio and ecosystem architecture
What it means
Treating small apps as experiments inside a larger product family with shared infrastructure, visual language, interaction patterns, and strategic meaning.
Evidence in the work
The haam-* portfolio, shared design systems, multiple domains, and hundred-app ambition turn individual prototypes into a coherent ecosystem.
04.4
AI-assisted communication and business development
04.4
AI-assisted communication and business development
What it means
Researching people and organizations, finding the real collaboration angle, and adapting outreach to the recipient, relationship, evidence, and desired next step.
Evidence in the work
AI supports journalist pitches, job messages, partnership proposals, follow-ups, contact databases, and research-backed outreach waves.
System 05
Creative direction
Using generated material to discover preferences, direct style, preserve voice, and build multimodal worlds.
05.1
Negative preference as a design method
05.1
Negative preference as a design method
What it means
Generating an imperfect possibility, noticing what feels wrong, converting that reaction into a constraint, and iterating until the preference becomes visible.
Evidence in the work
AI output often functions as a thinking surface. The ability to reject precisely becomes a method for eliciting taste that was difficult to describe from nothing.
05.2
Multimodal AI
05.2
Multimodal AI
What it means
Working across text, screenshots, photos, diagrams, maps, SVG, code, audio, video, 3D environments, medical-device readings, and spatial interaction.
Evidence in the work
Projects include a 3D Venice memory space, sailing simulations, animated photo histories, generated illustrations, filmstrip tools, and visual story analysis.
05.3
Generative creative direction
05.3
Generative creative direction
What it means
Defining mood, texture, density, typography, symbolism, motion, narrative framing, and the rules that keep generated assets inside one coherent world.
Evidence in the work
Raw Drafts, the HAAM seasonal design system, the cyberpunk octopus world, and Proxy Society all use AI as material under human art direction.
05.4
AI-assisted writing and editorial direction
05.4
AI-assisted writing and editorial direction
What it means
Finding the angle, building narrative tension, retrieving forgotten connections, preserving personality, and editing generic model output into publishable work.
Evidence in the work
Raw Drafts uses AI for research archaeology, structure, revision, titles, visual concepts, and computational autobiography, while final editorial judgment remains human.
System 06
Personal and cultural systems
Teaching AI how one person, language, culture, body, and learning process differs from the average user.
06.1
Voice and personalization design
06.1
Voice and personalization design
What it means
Separating factual memory from stylistic imitation, collecting representative examples, modeling recurring patterns, and correcting outputs that sound generically competent but personally wrong.
Evidence in the work
The writing corpus, voice guide, personal history, and repeated editorial corrections are becoming a retrieval system for representing Kris more faithfully.
06.2
Localization and cross-cultural AI
06.2
Localization and cross-cultural AI
What it means
Adapting language, politeness, authority, trust, finance, sustainability, and agent behavior to cultural context rather than performing literal translation.
Evidence in the work
Work across Estonia and Taiwan includes Traditional Chinese localization, multilingual products, and research into how acceptable agent behavior changes between societies.
06.3
Quantified-self and personal-data synthesis
06.3
Quantified-self and personal-data synthesis
What it means
Combining incompatible health sources into timelines, separating raw signals from interpretation, preserving provenance, and forming better questions for professionals.
Evidence in the work
ECG devices, Wellue, Qaly, Welltory, blood pressure, genetics, exercise, symptoms, and medical records are being treated as one longitudinal information problem.
06.4
AI-assisted learning and curriculum design
06.4
AI-assisted learning and curriculum design
What it means
Turning curiosity into sequenced lessons, examples, exercises, simulations, and projects, then using active work as the curriculum itself.
Evidence in the work
AI has become an adaptive teacher across design, languages, dance, sailing, architecture, technology, health literacy, and software development.
Skills hiding under the skills
The most valuable abilities are easy to mistake for personality.
They look informal because they are exercised through judgment, rejection, curiosity, and repeated making. Together they are the operating layer that keeps the tools useful.
AI taste
Recognizing when an output is technically acceptable but conceptually dead, socially wrong, visually generic, or not worth publishing.
Synthetic team management
Assigning outcomes, giving context, reviewing work, correcting direction, choosing tools, and deciding what is ready to ship.
Externalized cognition
Making an incomplete thought visible so it can be inspected, challenged, connected, and developed into a stronger idea.
Epistemic calibration
Sensing when an answer is shallow, unsupported, suspiciously confident, incomplete, or in need of another source and another pass.
Artifact-first thinking
Turning an idea into a page, prototype, map, database, or interaction because judgment improves once something concrete exists.
Boundary design
Defining where AI should stop, which actions need approval, who remains accountable, and what happens when the system is uncertain.
Continuous context building
Creating a personal knowledge environment where new work can draw on years of projects, messages, research, preferences, and evidence.
Research productization
Converting open-ended exploration into something other people can navigate, reuse, critique, collaborate on, or act upon.
Proof trail
The capability map is visible in shipped work.
These are not isolated experiments. Each project exercises several parts of the same research-to-product system.
Research and AI experience design
Green Filter
Six years of problem reframing, mixed-methods research, segmentation, conversational UX, trust design, cross-device prototyping, and evidence-led iteration.
- Human-centered AI
- Mixed methods
- Trust
- Product judgment
AI-native product operation
HAAM
A growing product and service system built through specifications, coding agents, design review, GitHub, Vercel, shared infrastructure, and human approval.
- Orchestration
- Software development
- Evaluation
- Ecosystem architecture
Computational autobiography
Raw Drafts
A publishing environment where AI helps recover connections, test angles, direct multimodal concepts, and turn unfinished thinking into durable public work.
- Editorial direction
- Voice
- Creative direction
- Knowledge publishing
Agent governance and social design
Proxy Society
A research and design frame for disclosure, delegated authority, etiquette, cultural plurality, accountability, and the social architecture around AI agents.
- Governance
- Boundary design
- Cross-cultural AI
- Systems thinking
Structured opportunity systems
WCD 2026 and AI Grant Finder
Large, messy landscapes transformed into directories, matching logic, comparable records, contact paths, and interfaces for finding the next useful action.
- Data curation
- Research
- Taxonomy
- Productization
Longitudinal personal intelligence
Personal data plane
An emerging system connecting GDPR exports, writing, finance, health, communication, projects, and life history under personal control.
- Context engineering
- Privacy
- Personalization
- Data sovereignty
Strongest current profile
AI-native product researcher and designer who turns ambiguous human problems into agent workflows, multimodal prototypes, evaluation loops, and shipped knowledge products.
Human-centered AIContext engineeringAgent orchestrationResearch synthesisAI evaluationRapid prototypingResponsible AIKnowledge architecture
Separate specializations
This is serious AI work without pretending every AI discipline is the same.
The current evidence is strongest in product, research, interaction, orchestration, evaluation, governance, and publishing. It does not yet establish the following specialist claims:
- Training or fine-tuning foundation models on a properly constructed dataset
- Production-scale MLOps, inference optimization, and GPU infrastructure
- Formal benchmark and evaluation-suite engineering at large scale
- AI security engineering and adversarial model testing
The through-line
I do not just use AI. I design the conditions in which AI can become useful work.
The value comes from combining human context, research, product judgment, creative direction, technical execution, evaluation, and responsibility into one operating loop.