June 2024 → June 2026

From Executive Operator to AI-Native Strategist

Two years deep inside AI tools, automation, local models, and AI-assisted development. Not from the sidelines, by building, breaking, shipping, and comparing what works in real workflows.

2026 Velocity · Claude Code

650K+
Lines of Code
net, H1 2026 · my authorship
26B+
Tokens · June '26
~1.1B/day · on pace for 34B
~$32K
Mo. Compute Value
at API list prices, not spend

Since June 2024

10,000+
Hands-on AI Hours
7,260
Commits
across 31 repos
1.12M+
Lines Shipped
production code, lifetime
70+
Tools Tested
5
Live Products
30K+ monthly users
Local LLMs
+ OpenClaw
in active use
  • 31 repos, 7,260 commits across two years.
  • ~80% of 2026's code shipped from six products, led by MeritPlaybook at 323K lines.
  • Heritage Whisper alone: 2,785 commits and counting.

Why this matters

Most AI advice comes from people who've either never operated a business or never built anything real with the tools.

I wanted both. I already knew sales, operations, enablement, incentives, customer experience, and executive strategy. The missing piece was understanding what AI could actually do when you push it hard.

This page is the receipts on that learning curve.

What I'm working on this month

  • Building Heritage Whisper toward production launch.
  • Running a local LLM stack for private workflow experiments.
  • Testing OpenClaw for autonomous browsing tasks.
  • Building a real-time conversational prompt refinement tool that role-plays, scores, and refines prompts before they ship.

Updated monthly.

How It Happened

Five phases. Two years. A lot of late nights.

01
Phase 01

The Curious Beginner

Understanding feasibility and breaking the "magic" barrier.

Commit Velocity ~5/week
First Questions
Asked Claude "Can you remember across chats?"
Started on a Windows machine with zero coding background. Struggled to differentiate between the Python shell and Command Prompt. Asked, "Can websites be coded in Python?"
First Automation
Birthday card pipeline with Make.com + GPT API
Built a birthday card automation pipeline using Handwrytten, Make.com, and GPT API. Early awareness of the tool ecosystem: "Could I use make.com for step 3?"
The Pivot to Code
No-code tools were limiting. Installed Python.
Realized no-code tools were limiting custom vision. Installed Wing Pro IDE and Python. First build: "Help me code a weather app with APIs." Learned that APIs are not magic, but structured contracts.
Python Make.com GPT API
June – August 2024
02
Phase 02

The "Big Bang"

Rapid experimentation. Learning by shipping. Breaking things.

Commit Velocity ~25/day (Oct)
6 Projects in One Month
todo-app, ZeroEffortAI, TheFinalsLoadout...
October 2024 explosion: todo-app (React practice), ZeroEffortAI (prompt library), TheFinalsLoadout (Vanilla JS), premium-response-generator, and Heritage Whisper V2 initialization.
763 Commits in October
~25 commits per day average
Heritage Whisper V2 initialization (Sept 30) followed by explosive development. State management bugs cascade lesson learned: "We went backwards. Now it's spinning the wrong direction."
The "Great Error Log"
18 errors in one terminal session
May 8, 2025: A massive multi-agent coding session (Claude + GPT pipeline) generated 18 recorded errors. Hit hard limits on context_length_exceeded and max_tokens. Learned to chunk tasks and "fail loud".
Next.js 14 React Supabase TypeScript Zustand
September 2024 – May 2025
03
Phase 03

Architecting Heritage Whisper

Production-grade architecture, security, and strategic constraint.

Commit Velocity ~10/day
Voice-First Strategy
"Seniors just talk, no typing"
Strategic pivot for elderly users. Prompt engineering breakthrough: "The long prompt is way worse than the short prompt." Zero-shot voice prompts outperformed elaborate instructions.
Security Architecture
Supabase Auth + RLS + WebAuthn
Implemented Row Level Security (RLS) to enforce tenant isolation. Added WebAuthn passkeys (TouchID/FaceID for seniors). Ethics decision: Rewrote public copy to avoid overpromising on compliance.
Stripe Integration
No "heroic wiring." Native integration.
Rejected manual API wiring in favor of Vercel's native Stripe integration. Forced webhooks to Node runtime (not Edge) because raw body verification failed in Edge environments.
Next.js 15 OpenAI Realtime WebAuthn Stripe Drizzle ORM
August – October 2025
04
Phase 04

The AI-Native Operator

Scale, SEO, advanced automation, and "AI Leader" operations.

Commit Velocity ~11/day
MCP Integration
Supabase + GitHub + Vercel MCPs
Connected Model Context Protocol servers for Supabase, GitHub, and Vercel. Discovered Claude Code's UI didn't support specific OAuth flows, requiring CLI configuration.
SEO Strategy Launch
4 content clusters defined
December 2025: Defined four SEO clusters (Memory Book, Legacy Book, Family Memory, Competitor Comparisons) to target high-intent traffic. Multi-tier personalization system designed.
PR Strategy & Soft Launch
Japanese American publications outreach
February 2026: Executed soft-launch messaging targeting Japanese American publications, leveraging the 442nd RCT family story. "Treat automation like production infrastructure."
Claude Code MCP Remotion React Native PDFShift
November 2025 – February 2026
05
Phase 05

The Multiplied Operator

One operator, output like a team: multi-agent orchestration, a production AI cluster, and teaching what actually works.

Token Velocity · Claude Code ~1.1B/day
Local AI Cluster, Production-Grade
Four nodes, always on
Spring 2026: Wired a Mac Mini orchestrator to a Mac Studio M3 Ultra (512GB, running Qwen), an M4 Max (OpenClaw), and a 4090 PC (Stable Diffusion). Private inference, hardened to run like production infrastructure.
Multi-Agent Pipeline
Claude codes, Codex cross-checks
H1 2026: ~1.1B tokens a day through Claude Code, 26B+ in June alone, with a second model adversarially reviewing every risky diff. 650K+ lines of code shipped this year, net, my own authorship.
Teaching GEO in Public
5-part course + a 10K-citation study
June 2026: Shipped a free five-part GEO course and a first-party study of 10,000+ AI citations answering what engines actually cite, alongside an operator field-notes series. "Demand decides what gets cited."
Claude Code Codex Local LLMs OpenClaw IndexNow
March – June 2026

Built With

Claude
Anthropic
Cursor
Vercel
Supabase
Next.js
React
TypeScript
Tailwind CSS
Stripe
GitHub
Framer
Google Gemini
ElevenLabs
Drizzle
Cloudflare
Redis
PostgreSQL
Docker
Radix UI
shadcn/ui
Zod
Perplexity
Ollama
v0
Replit
GitHub Copilot
Cline
Suno
Notion
Firebase
Wix
Brave
Resend
Make.com
Excalidraw
NotebookLM
OpenAI
VS Code
Midjourney
Aider
Remotion
Runway
AssemblyAI
Grok
Qwen
DeepSeek
Open WebUI
Factory AI
Lovable
Exa
Stability AI
Voicemod
Adobe
Canva
DocuSign
SendGrid
AnythingLLM
SearXNG
OpenClaw

The Tuition

What it actually took to become AI-native.

No bootcamp, no CS degree, no playbook. 17 months of testing every frontier tool, burning through tokens, and building a local cluster to run it all.

10,000+
Hours
Hands-on AI work
70+
Tools Tested
17 months
$20K+
Tuition
Frontier subs & APIs
30B+
Tokens Burned
≈ 39M pages

Top Spend · 17 Months

Five tools over $2,000 each.

Where 17 months of frontier testing actually went.

ChatGPTOpenAI · Pro + API
$2,935
KlingAIVideo gen · Jan–May 2025
$2,689
GrokxAI · Active
$2,410
ClaudeAnthropic · Daily driver
$2,000+
CursorAI editor · Active
$2,000+

Breadth

Ten categories. One operator.

Most "AI users" live in one app. Real fluency means knowing when to reach for which tool, and which to leave behind.

Frontier LLMs

5

Claude · ChatGPT · Gemini · Grok · Perplexity

Local LLM

6

Qwen 3.5 397B · Qwen 3.6 35B · DeepSeek R1 · LM Studio · Ollama · Qwen 2.5

AI Coding

9

Claude Code · Cursor · Replit Agent · v0 · Factory AI · Copilot · Codex · Aider · Cline

Image Gen

7

Midjourney · DALL-E · Gemini · Flux · Stable Diffusion · Nano Banana · Promptchan

Video Gen

6

KlingAI · Runway · Sora · HeyGen · Beautiful.ai · Gamma

Voice / Audio

8

Wispr Flow · ElevenLabs · Auphonic · Suno · Whisper · AssemblyAI · Otter · Voicemod

Agent / Orchestration

4

OpenClaw · LiteLLM · Open WebUI · AnythingLLM

MCP / Search

6

Ref.Tools · Exa · SearXNG · Serper · Context7 · Brave

Build / Design

9

Google AI Studio · NotebookLM · Whimsical · Excalidraw+ · Mobbin · Canva · Aura · Lovable · Pompeli

Infrastructure

12

Vercel · Supabase · GitHub · Cloudflare · SendGrid · Adobe · Make.com · Resend · Notion · DocuSign · Wix · Firebase

Discernment

I keep what works. I drop what doesn't.

A 50/50 split between active and dropped tools isn't churn. It's how you learn what actually earns its place in a workflow.

Kept · Active Daily

35
  • ClaudePrimary thinking partner
  • ChatGPTPro + API
  • GeminiWorkspace + mobile
  • OpenClawCluster orchestrator
  • ElevenLabsVoice synthesis

Tried · Dropped

35
  • Otter.ai→ Whisper API
  • GammaQuick trial, didn't stick
  • HeyGenAvatar quality
  • Replit AgentHeavy use, then graduated
  • DeepSeek R1 671BBenchmarked, moved on

Volume isn't taste. The dropped column is where the taste lives.

The AI Rig

Four machines. One cluster.

$15K+ in hardware. Networked, orchestrated, running 24/7. Most AI users rent. This stack runs at home.

Online

Orchestrator

Mac Mini

Apple Silicon · Tailscale

Runs OpenClaw. Dispatches jobs across the cluster.

Online

Frontier LLM

Mac Studio

M3 Ultra · 512GB RAM

Runs Qwen 3.5 397B locally via LM Studio.

Online

Worker Node

Mac Studio

M4 Max · 64GB RAM

Second OpenClaw worker. Parallel agent execution.

Online

Image / Video

4090 PC

RTX 4090 · 24GB VRAM

Runs Stable Diffusion + a second local LLM.

Evolution of Thinking

The long prompt is way worse than the short prompt.

After testing Pearl AI. Complexity was hurting performance.

Pearl can not have any opinions. Is that clear in here?

Defining the AI as a "witness," not a therapist.

Let's stop the incremental patches. The system is fundamentally broken.

Recognizing architectural debt. The shift from "Help me make X" to systematic engineering.

If it runs 24/7, treat it like production infrastructure.

Feb 2026, while hardening OpenClaw automation.

The Journey Visualized

What I'm Building Now

Three operator-built products in active development.

Heritage Whisper

Voice-first platform helping families preserve stories across generations.

MeritPlaybook

Personalized, research-backed scholarship strategy plans and recommendations.

CredibilityOS

AI visibility platform helping mortgage loan officers see and fix what weakens trust.

If you have questions about this journey, want to share your own experience, or just want to connect, I'd love to hear from you.

Connect on LinkedIn

The thinking behind the build:

Two Questions The Wrong Plane