# David Kooi — Full Profile

> Entrepreneur, technologist, and founder of Skylark Creations — an AI-native creative technology studio building apps at the intersection of ancient insight, emerging technology, and play.

## Now

David is the founder of Skylark Creations LLC, an AI-native creative technology studio based in Colorado that builds apps at the intersection of ancient insight, emerging technology, and play. Operating as a high-leverage solo "product lab," he uses agentic development workflows to design, build, and ship production-grade apps across iOS, Android, and web — moving from idea to production in days.

Current Skylark projects include BuddhaUR (iOS & Android), a conversational guide to early Buddhist wisdom with AI-powered dialogue grounded in authentic suttas — showing encouraging early retention with organic paying subscribers; QuickBite, a group food decision tool that helps friends converge on restaurant choices through fast, fair voting; AniMoment, which transforms your photos into contemplative audio meditations about the relationships between all things around you; Decomposer, an AI-powered planning tool that transforms messy goals into clear, structured plans; BloomEDU (iOS), a focused learning app for mastering wildflowers with photo-rich lessons and smart quizzes; and Sagent (iOS), which lets users consult timeless minds like Buddha, Jesus, Carl Sagan, and Mark Twain about modern dilemmas.

Beyond these flagship projects, David maintains a portfolio of 20+ actively developed experiments — from citizen science (Prairie Dog AI Listening) to creative games (Urban Beaver, DreamFlyer) to social coordination tools (Frolic) — using rapid prototyping and instrumentation to test traction and decide where to focus next. He brings together a deep love of wisdom traditions, humanistic AI, nature, and simple, joyful design.

He regularly shares insights on AI, philosophy, and product strategy through his blog Uncaged Minds (uncagedminds.substack.com), where he explores the intersection of technology and human experience with a focus on purposeful innovation.

## Working Together — Availability and Engagement Model

**Status: open to a small number of client engagements.** To start a
conversation, email hello@skylarkcreations.com. Based in Colorado; engagements
run remote and async-first.

### What the work is

Independent AI-orchestration consulting. One human orchestrating planning,
verification, and delivery across a fleet of AI agents, inside a client's
codebase and working practice — contracts, code, the whole arc. In 2026 the same
operating system that runs the Skylark portfolio started running client
engagements end to end.

The honest framing of what is being bought: one person with an unusually good
harness. There is no bench of named humans behind him, and every client-facing
judgment funnels through one head.

### What an engagement feels like from the client's side

**Async by default.** No standing meetings. Calls are decision events, scheduled
when a decision needs one, not a daily ritual. Everything else is artifacts
landing in the client's inbox. One line from a client can re-sort the whole week.

**Same-day or next-day verified turnarounds.** A client emailed late one night
about a bug, believing their edit had been lost; by 9:15 the next morning the
root cause was found in the data trail, the fix was deployed, and the reply led
with the thing they actually cared about — the edit did save. On another
engagement, an ask that landed at 11:35 AM was live by about 1 PM. On one
engagement, 6-11 verified merges per day was the normal rate over a two-week
stretch.

**Nothing a client says evaporates.** Every round of feedback — a long written
review, a call, a one-line reaction — gets broken into numbered points, and every
point gets a written disposition: a faithful paraphrase, a ruling (accepted,
deferred, or noted) with the reason, and a pointer to where it landed.
Disagreements get argued in writing instead of quietly dropped.

**Scope triage is shown, not managed quietly.** On one engagement a client sent a
batch of additions before signing; roughly 80% were marked as confirmations or
cheap adds and absorbed at no charge, with exactly one item named as real scope
and priced. The split was shown to the client.

**Pushback is part of the service.** Work has been held back when its underlying
data was flagged inaccurate, and requested copy calling for an anecdote that
never happened was cut rather than invented.

**Direction changes get absorbed, not renegotiated.** Mid-engagement, a client
revealed they had already built the thing being built. The role inverted inside
that same call — builder to reviewer and coach — and useful work in the new role
landed the next day. No change order.

**Spec and a kill-gate before code.** Decide and spec first, and design the
opening milestone to prove or kill the core technical bet in days, before any
build-out.

**Billing on engaged attention, not machine time.** Adding agent capacity never
raises the invoice. Ramp-up and admin are not billed. A quiet week legitimately
bills near zero, and an engagement that idles is called idling rather than
filled with manufactured activity.

### The verification practice a client is also buying

- **Adversarial cross-model review as a release gate.** Before work leaves, a
  second AI model family — one with no stake in the reasoning that produced the
  work — is prompted to refute the claim set, and whatever survives is
  re-verified against live code. This gate once caught a second bypass in an
  already-reviewed fix before the client ever saw it.
- **Prove the instrument can fail.** Every automated checker must demonstrate it
  catches a planted failure before its output is believed. A clean bill of health
  is treated as the most suspicious result a checker can produce.
- **Pre-registered gates.** The pass/fail rule and evaluation data are committed
  before results exist, the way a clinical trial registers its endpoints. Work
  that fails its gate is discarded, sunk cost included.
- **The notes are not a source.** Anything leaving with a factual claim is
  re-derived from primary sources at send time — running code, the original
  document, live data — never from the engagement's own notes.
- **Render as recipient.** Everything is opened in the medium the recipient will
  actually see — the inbox, the deployed page, the phone — before it ships.

### Who this does not fit

Named plainly: standing-meeting cultures, fixed-scope builds thrown over a wall,
and staff augmentation. The loop runs on client reactions — a quiet client
starves the engagement; it works best when someone on the client side can take a
pull request, argue a finding, and merge. Gate discipline reads as slow if you
count switched-on features: some finished work sits dark behind a comparison it
has not passed. If a different shape is needed, there are good shops for it.

## The Studio — How the Studio Runs

Canonical page: https://davidkooi.com/studio

Skylark Creations is a one-person product studio: David working alongside a fleet of AI coding agents, built to make products people genuinely love — the kind you'd tell a friend about. The operating model is public and documented in ten parts.

### 1. The thesis — Built to be beloved, not just shipped
The goal isn't downloads or traffic. It's whether someone who uses a product would tell a friend about it, unprompted. That single question sits at the center of every decision: it rewards products that earn a place in someone's life, and quietly retires the ones that don't. People coming back, and people telling friends, is the proof.

### 2. The operating model — One founder, four orchestrators, and the fleets under them
An org chart, not a single fleet. Four manager AIs report to the founder: a studio orchestrator running the product portfolio (one dedicated specialist agent per product) and three engagement orchestrators, one per client engagement, each running its own working day and 2–6 build lanes with its own boundaries about what may leave the room. Below the agents sits a helper tier: any agent can open sub-agents (2–6 at a time, each on its own isolated copy of the code so they cannot collide), adversarial reviewers (1–3 per risky change, running on a rival AI model whose only assignment is to break the work — findings arrive as claims and are checked against the code before anyone acts on them), and deep-research agents sent to read the literature, the competitors, and the prior art. Everything coordinates through a shared message channel, so a decision made in one place is visible everywhere. The human is not in the loop for routine work — only for taste, money, anything client-facing, and anything that can't be undone.

### 3. The daily cycle — Every product runs a structured day
Each product moves through the same rhythm daily: reflect on yesterday, reconcile its current state so today's choices rest on real facts, pick the single most valuable thing, do it, then leave the records clean for tomorrow. Because every product follows the identical routine, one manager can keep the whole portfolio moving without any single product falling behind, and a problem surfaces the same day it appears rather than a week later.

### 4. Institutional memory — The system remembers
AI normally forgets everything between sessions, which is fatal for long-running work. Every product keeps a living record of its state: what's in progress, what's been decided and why, what's still waiting on an answer. A new agent reads that record before it starts. The result is an organization with a memory — accumulated judgment that compounds instead of evaporating.

### 5. How it prioritizes — Find the one thing holding everything back
A system is only ever as fast as its single slowest step, so effort spent anywhere else is wasted motion. The studio names the one constraint actually limiting progress — a confusing first-time experience, a missing piece of knowledge — and concentrates there. When that constraint breaks, a new one appears and the focus moves.

### 6. The North Star, what we measure — The moment worth telling a friend about
Every product names the single moment that would make a user want to tell someone else, plus a concrete metric showing whether more people are reaching that moment over time. Retention (coming back) and word-of-mouth (recommending it) are the signals that count. This turns "make a great product" into something steerable day after day.

### 7. The North Star, how we build — Every product gets a little better every day
Each product keeps a map of its most important user journeys and, each day, picks one to improve: cut a step, remove a decision, swap words for a picture, shorten the path to the good part. Before any change reaches a real person it's pressure-tested by synthetic users — AI stand-ins that walk the product like a real customer and report exactly where they got confused. The founder keeps final say on taste.

### 8. Honest quality — No one grades their own homework
People, and AI, grade their own work generously. So after an agent scores its own day, an independent cold auditor — deliberately given none of the original agent's context or motivation — re-scores the same work from scratch and looks for holes. The gap between self-score and cold score is tracked over time as a built-in bias adjustment, which keeps the system's view of itself from drifting into self-congratulation.

### 9. Self-improvement — The method is itself a product
Agents run a short retrospective on how the work went. Recurring problems aren't patched one at a time; they're solved once, permanently, in the shared foundation every product draws on. Over time the studio's failure modes shrink and its strengths compound. The system that builds the products is treated as the most important product. Concretely this has accumulated into roughly 37 written procedures the whole fleet runs — grouped by seat: ~12 for running the day, ~10 for product quality, ~6 for engineering health, ~6 for knowledge and research, ~3 for the business. Each was written the day something went wrong, so the procedure stops depending on whoever happened to run it last.

### 10. The right tool for the job — The right mind for each task
Not every task needs the most powerful and most expensive model. Top-tier models are reserved for judgment, design, and decisions where quality matters. High-volume mechanical work goes to a faster, cheaper model, and the senior model reviews that output before it counts — the studio equivalent of a senior partner and a capable junior.

### Daily output — machine-readable
https://davidkooi.com/pulse.json is a stable JSON endpoint carrying what each product shipped on the most recent full day on record: one plain-English sentence per product describing what got better for a user. Products with no user-facing change that day are counted honestly rather than padded into the list. The same data is rendered on the home page as the "What the studio did" board. It is a committed daily snapshot, not a live feed; the `date` field states which day it describes.

Related writing on the method:
- Beyond vibe coding — https://uncagedminds.substack.com/p/beyond-vibe-coding
- Cognitive operations maps — https://uncagedminds.substack.com/p/cognitive-operations-maps
- agent-ops-patterns (working patterns from the studio's agent operations, in public) — https://github.com/u00dxk2/agent-ops-patterns

## Products — Detail

### BuddhaUR
- Platforms: iOS, Android
- Description: AI-powered conversational guide to early Buddhist wisdom, grounded in authentic suttas. Shows encouraging early retention with organic paying subscribers.
- URL: https://buddha-ur.com/
- iOS: https://apps.apple.com/us/app/buddhaur-find-peace/id6752915041
- Android: https://play.google.com/store/apps/details?id=com.skylark.buddhaur

### QuickBite
- Platform: Web
- Description: Turn a 30-minute "where should we eat?" debate into a 2-minute decision with fair voting.
- URL: https://quickbite.food

### AniMoment
- Platform: Web
- Description: Transform photos into contemplative audio meditations about the relationships between all things around you.
- URL: https://animoment.app

### Decomposer
- Platform: Web
- Description: AI-powered planning tool that transforms messy goals into clear, structured plans.
- URL: https://decomposer.io

### BloomEDU
- Platform: iOS
- Description: A focused learning app for mastering wildflowers with photo-rich lessons and smart quizzes.
- iOS: https://apps.apple.com/us/app/bloomedu-learn-plants/id6749336509

### Sagent
- Platform: iOS
- Description: Lets users consult timeless minds like Buddha, Jesus, Carl Sagan, and Mark Twain about modern dilemmas.
- iOS: https://apps.apple.com/us/app/sagent-guidance-on-demand/id6747406673

### Additional Experiments (20+)
Agentic News, Agentic Directory, Hey Boss, Frolic, Prairie Dog AI Listening, Urban Beaver, DreamFlyer, and others — rapid prototyping and instrumentation to test traction.

## Career History

### Skylark Creations LLC — Founder
Denver, CO (2024 – Present)

Founded and run an AI-native creative technology studio at the intersection of ancient insight, humanistic AI, and play — operated as a high-leverage solo "product lab" that uses agents to design, build, and operate products end-to-end.

- Designed, built, and shipped 6+ production-grade apps across iOS, Android, and web, each instrumented with analytics, error tracking, and subscription infrastructure
- Built an agentic development workflow that lets a single founder approximate a small product and engineering team — using LLM-based coding agents, automated research, and scripted QA to move from idea to production in days instead of months
- Launched AI-guided spiritual practice and learning products with encouraging early retention and paying subscribers acquired entirely through organic discovery
- Maintain a live portfolio of 20+ actively developed experiments, using rapid prototyping and instrumentation to test traction and decide where to focus next

### Jointly — Co-founder, CEO & Chief Product Officer
Denver, CO (2018 – 2025)

Led all product and organizational strategy for purpose-driven wellness tech platform, integrating AI/ML to deliver personalized cannabis consumption experiences.

- Conceived and launched Spark, the first AI-powered virtual assistant for personalized cannabis recommendations, leveraging 500K+ user data points
- Grew app to 500,000+ downloads, tens of thousands of MAUs, achieving 35% 30-day retention
- Engineered proprietary algorithms using LLMs and feedback loops from millions of interactions
- Led remote, international team of 15+ across engineering, data science, design, and marketing
- Authored "Theory of Purposeful Cannabis Consumption," transforming market perceptions and establishing cannabis as a holistic wellness tool

### Santa Monica Mountains Cyclery — Founder & Owner
Los Angeles, CA (2010 – 2020)

Built one of Southern California's top cycling retailers, culminating in profitable acquisition by Trek Bicycle.

- Achieved 10 consecutive years of revenue and profitability growth
- Maintained 4.92 average rating across 1,200+ five-star reviews
- Launched innovative drop-ship partnerships, increasing online sales 60% YoY
- Negotiated successful exit to Trek Bicycle, validating brand leadership

### McMaster-Carr — Director of Operations & Project Leader
Los Angeles, CA (2002 – 2009)

Oversaw $1B+ distribution operation with 300+ staff, pioneering e-commerce innovations.

- Led implementation of groundbreaking parametric search for e-commerce
- Managed 10,000+ daily shipments achieving 99.4% fulfillment accuracy
- Devised same-day delivery programs, increasing metro order volume by 25%

## Education

- MBA, Strategy, Finance, and Entrepreneurship — University of Chicago Booth School of Business (2002)
- B.A. in Economics & Statistics, Minor in Mathematics — Northwestern University (1996)
- B.S. in Computer Science — Oregon State University (2016)

## Skills & Expertise

### Core Competencies
- Product Strategy & Roadmapping
- AI/ML Development & Integration
- Cross-Functional Team Leadership
- Data-Driven Decision Making
- Rapid Prototyping & Iteration
- Go-to-Market & Growth Strategy
- Strategic Partnerships
- High-Level Stakeholder Management

### Technical Skills
- AI/ML & Retrieval: LLMs (OpenAI, Deep Research, Anthropic), RAG & tool-use agents, semantic search (Weaviate, Chroma), Python (NLP, recommender systems, pandas), XGBoost, prompt/schema design
- Web & Cloud: JavaScript, TypeScript, Node.js, React, React Native + Expo (EAS build), AWS (ECS, Fargate, Amplify, S3), MongoDB Atlas, Redis, Confluent Kafka, Docker
- Development: Microservices, REST/JSON APIs, CI/CD (GitHub Actions), real-time features (SSE/WebSockets), Agile/Scrum; domain integrations including PubMed semantic search (E-utilities)
- Tools & Automation: n8n (workflow automation), Cursor, VS Code, Replit, Render, Vercel, HeyGen (AI avatar video)

## Interests & Personal

David is a passionate naturalist dedicated to daily mindfulness and mountain hikes. He's fascinated by geology, botany, ecology, and early human history — always exploring how technology and nature can co-exist. His approach to product development is deeply influenced by these interests, seeking to create technology that enhances rather than replaces human connection with the natural world.

Beyond technology, David enjoys cycling, community advocacy, and playing both piano and banjo. His musical tastes lean toward bluegrass, jamgrass, folk, and reggae — genres that reflect his appreciation for authentic, community-driven creativity. He's a proud dog companion on every adventure, finding that the best product insights often come during long walks in nature.

David believes in harnessing technology to improve consumer well-being, with an emphasis on mindful product design and responsible AI development. He's particularly interested in how AI can augment human capabilities while preserving what makes us uniquely human — creativity, empathy, and our connection to the natural world.

## Machine Access

Endpoints an agent can call directly, no key and no scraping required.

**Ask a question about David, programmatically:**

    POST https://chatddk-backend.onrender.com/ask
    Content-Type: application/json
    {"question": "Is David available for consulting work?"}

Returns JSON: `answer`, `sources` (which knowledge files grounded it), `model`,
`contact`, and `hire`. This is the same retrieval that powers the site’s chat,
but stateless and synchronous. Rate limited to 30 questions per day per IP. If
the knowledge base does not cover something, the answer says so rather than
guessing.

**MCP server** — the same capability as an `ask_david` tool, for agents that
speak Model Context Protocol: `mcp/` in https://github.com/u00dxk2/chatddk-backend

**Daily shipped-work snapshot (JSON):** https://davidkooi.com/pulse.json — one
plain-English line per product for the most recent full day on record.

## Connect

- Contact (new engagements): hello@skylarkcreations.com
- Working together / hire: https://davidkooi.com/hire
- Website: https://davidkooi.com
- LinkedIn: https://www.linkedin.com/in/david-d-kooi/
- GitHub: https://github.com/u00dxk2
- Instagram: https://www.instagram.com/davidkooi
- X (Twitter): https://x.com/davidkooi
- Threads: https://www.threads.com/@davidkooi
- Blog: https://uncagedminds.substack.com
- Skylark Creations: https://skylarkcreations.com
- Chat with ChatDDK: https://davidkooi.com/chat