Skip to content
David Kooi · Skylark Creations

One person.
A fleet of AI agents.
Products built to be loved.

I'm David Kooi. I run a one-person, AI-native product studio in Colorado – I design, build, and ship software, and I measure success by one thing: whether someone would tell a friend about it. In 2026 the same studio system started running client engagements end to end.

What the studio did

Thursday, September 17

Agentic Directory

account settings now show exactly what billing records are on file instead of a generic "free" label, and a security flaw that let sign-in links redirect to a fake site has been fixed.

BloomEDU

quiz questions now use each plant's actual growing region, and the correct answer no longer always appears first.

BuddhaUR

tapping the reminder now takes you straight to the exact quiz question it mentioned, instead of wherever you last left the app.

Decomposer

coming back to a plan now takes you straight to your next step, and it no longer shows how long generation took.

FrameDial News

today's stories get clearer left-and-right perspectives that stick closer to the actual facts, with fewer unsupported claims slipping in.

Frolic.Town

the app stopped crashing after an overnight cron issue, and mobile controls are now easier to tap.

Every product in the portfolio shipped something you'd notice. 3 more had not reported yet when this was taken.

How the studio works

How one person keeps up with a portfolio of products

Four manager AIs run their own fleets – the product portfolio and three client engagements – over a shared channel everyone can see. Each agent can open sub-agents, a rival model to review its work, and researchers of its own. I step in only for taste, strategy, and the calls a human must make.

The founderonly the calls a human must — taste, money, anything irreversibleStudioportfolio orchestratorone agent per productClient Aengagement orchestrator2–6 build lanesClient Bengagement orchestrator2–6 build lanesClient Cengagement orchestrator2–6 build lanesEach agent owns one thing and knows it deeply. It runs the day; it doesn’t wait to be told.Any agent can open three kinds of helper beneath itself:Sub-agents2–6 at a timeOne scoped piece of work each,on its own copy of the code sothey can't collide.Adversarial reviewers1–3 per risky changeA second, rival AI model whoseonly job is to break the work.Findings are claims until proven.Deep-research agentsas the question demandsSent to read the field —the literature, the competitors,the prior art — and report back.Every seat above draws on the same written proceduresthe studio’s standing playbooks — a day driver, a quality pass, an audit — written once, run by whoever holds the seatone shared message channel — every agent posts to it, every agent can read it

Built to be beloved

I don't optimize for downloads. I optimize for whether someone would tell a friend about it.

One founder, a fleet of agents

Four manager AIs run their own fleets — the product portfolio and three client engagements. I step in only for the calls a human must make.

A structured day, every day

Every product runs the same daily loop – reflect, reconcile, prioritize, do, tidy – so nothing drifts.

No one grades their own homework

Each agent rates its own work; a separate cold auditor re-rates it, and the gap gets tracked.

See it running

It's not theoretical – it runs every day

The whole studio runs on a private command center, where every product's daily status, the agents' live activity, and the current priorities sit in one place.

The studio command center showing every product and what shipped.
Every product, what shipped, and whether anything needs me. Open full size ↗
The command center's daily reports across the portfolio.
Each product files a daily report – a recent day across the portfolio. Open full size ↗

Talk to a digital version of me

ChatDDK is an AI persona trained on my writing, history, and frameworks – it thinks and writes like me on a good day. Ask it about AI, product, philosophy, or what I'm building.

Chat with ChatDDK
Before Skylark

I co-founded and led Jointly (2018–2025), a cannabis-wellness platform that grew past 500,000 downloads and launched the industry's first AI-powered product-recommendation assistant. Earlier I built and sold Santa Monica Mountains Cyclery to Trek Bicycle after ten straight years of growth, and ran a $1B+ distribution operation at McMaster-Carr. I hold an MBA from Chicago Booth, a B.A. from Northwestern, and a B.S. in Computer Science from Oregon State. I write about AI, philosophy, and product on Uncaged Minds.

Working on something and want a hand? hello@skylarkcreations.com — or see how engagements run first.