PRIORITY MAP
Workstreams · automation · signals · knowledge
Q1 2026
COA World Systems
Network
Insights
Signals
Knowledge Engine
Cards
The Real Move: From Operator to System
Building the machine that runs COA without you in every loop
▼
Every payment, SOW renewal, and check-in lives in your head. That's a scaling ceiling. Once email is in Google, everything becomes programmable.
1
Email → Google Workspace
Do Now
▼
2 hours of DNS work unlocks months of automation.
↗ Opens: Gmail API, Apps Script, everything below
2
Build the Label System
Do Now
▼
Labels = database tables for email. Every email becomes structured data.
★ Two parallel tracks
3A
Money Becomes Visible
Next
▼
Apps Script scans "Invoices" daily → Sheets ledger. Claude Skill for smart categorization.
↗ Feeds: Burn rate, runway
3B
Time Becomes Managed
Next
▼
You respond to prompts instead of remembering dates.
4
See Everything: The Dashboard
Next
▼
Every strategic question gets a number behind it.
5
You Work ON the Business
Goal
▼
Attention shifts to products, deals, and creative work.
Two Products Waiting on One Fix
Supabase gates Portal, Simple, AND the Knowledge Engine
▼
Fix once, ship three things. Cleanup sprint + enable pgvector = Portal deploys, Simple gets its API layer, Knowledge Engine gets its data home.
1
Supabase Cleanup Sprint
Do Now
▼
2-3 day sprint. Fix RLS, consolidate, enable pgvector.
★ Three paths
A
Portal → Retainer Leverage
Ship
▼
Once Nike logs in daily, that's switching cost.
B
Simple → Sales Weapon
Ship
▼
Generate a loyalty deck live in the DTC meeting.
C
Knowledge Engine → Second Brain
Build
▼
pgvector + items table = searchable, connected knowledge. See the Knowledge Engine tab for the full deep dive.
Building Your AI Team
Each Skill is a hire that costs nothing
▼
The gap isn't knowing AI exists — it's having specific skills built around your actual workflows.
1
Learn the Format
Study
▼
SKILL.md = YAML + instructions. Use skill-creator to scaffold.
2
Skill #1: Invoice Parser
Build
▼
Replaces 30+ min/week.
3
Skill #2: Status Digest
Build
▼
Clients get updates you never wrote.
4
Skill #3: Shopify Audit
Build
▼
A deliverable that justifies your retainer.
5
Skill #4: Knowledge Ingester
Build
▼
OCRs, summarizes, embeds anything you throw at it. The knowledge engine's intake pipe.
The Bigger Picture
What all of this is actually building toward
▼
You're building an agency that operates like a product company, powered by a knowledge engine that compounds over time.
Portal = retainer defense. Clients depend on it daily. Switching costs baked in.
Simple = sales weapon. Generate decks live. Sell capability, not services.
Knowledge Engine = compounding intelligence. Every book, article, and reference makes you smarter. Six months from now: 10,000+ searchable insights.
Study areas converge. Supabase fixes products AND powers knowledge. Skills automate ops AND become deliverables.
DTC deal proves it all works. Close with Simple, dashboard clarity, automations, knowledge. Business model validated.
What to Study and Why
Ranked by what unblocks the most value
▼
1
Google Workspace APIs
Start Here
▼
Learn by building the payment tracker.
2
Claude Skills
Multiplier
▼
The difference between "if X then Y" and "understand this."
3
Supabase + pgvector
Products + Knowledge
▼
Unblocks three products. 13 tables + vector search waiting.
4
Claude API
Simple + Engine
▼
Powers both Simple's deck gen and Knowledge Engine's processing pipeline.
5
Financial Modeling
Dashboard
▼
Know what to track before building.
6
Shopify + Loyalty
When DTC Moves
▼
Study when the deal gets real.
What You're Seeing
Patterns from your research — what they mean for COA World
Trend
Workflows are the new moat
The gap between "knowing AI exists" and "having AI run your operations" is where all the value lives.
Every custom Skill, every automation, every Supabase workflow = proprietary infrastructure no competitor can copy.
Opportunity
One person + systems > small team
The return on wiring tools together is nonlinear. First automation saves 30 min. The tenth saves a full-time hire.
You're running an agency with major clients AND building products. That's only possible if ops get automated.
Trend
Creative tools are becoming generative
Tools that generate, not just edit. The creators who thrive bring taste to the output.
Simple = your editorial eye + Claude's generation. Hand-crafted quality at infinite scale.
Watch
Model convergence shifts value
Value moves to proprietary data, unique workflows, and speed.
Your Skills, client data, workflow IP, knowledge library = the moat. AI is the engine, not the chassis.
Study Queue
Tactical patterns worth pulling apart
1. Repeatable AI ops playbooks — Document → repeat → automate. Your Claude Skills strategy.
2. AI creative pipelines — Break production into AI steps, keep taste human. The Simple architecture.
3. Data pipelines — Supabase, Gmail, Sheets = all pipelines. Core skill for everything.
4. Productize → validate → scale — Retainers fund products, products create leverage.
2. AI creative pipelines — Break production into AI steps, keep taste human. The Simple architecture.
3. Data pipelines — Supabase, Gmail, Sheets = all pipelines. Core skill for everything.
4. Productize → validate → scale — Retainers fund products, products create leverage.
Deep Dive
KNOWLEDGE ENGINE
A personal library that grows with you — images, books, references, bookmarks — all searchable by concept, not filename. Start from how you'll USE it, then design how you capture.
Part 1 — Start here
How You Actually Use This Thing
Most knowledge tools fail because they're designed around capture. That's a filing cabinet with a search bar. This system is designed around retrieval moments — the specific situations where you need something, fast.
The five retrieval moments:
Moment 1
"What did I read about retention loops?"
You're prepping the DTC pitch and know you bookmarked something three weeks ago. Don't remember where, who, or what site. You just remember the concept. The engine finds it by semantic meaning across all content types.
Type a question
→
Vector search
→
Ranked results
→
Jump to source
Moment 2
"Show me everything about Supabase RLS"
Cleanup sprint — want all reference material in one view. Bookmarks, Stack Overflow screenshot, book page photo, conversation notes. One query, all formats, same result set.
Topic search
→
Mixed results
→
Images + text + links
→
Visual grid
Moment 3
"I just found something — save it NOW"
On your phone, scrolling. Or at a bookstore photographing a page. Capture needs to be under 3 seconds. Share sheet on mobile. Drag-and-drop on desktop. Processing happens in the background.
Share / drop / scan
→
Instant save
→
Background process
→
Searchable in minutes
Moment 4
"What connects to what I'm working on?"
Deep in Simple's API integration and the engine surfaces a bookmark about structured outputs from two months ago. The system noticed the connection and brought it to you.
You're working
→
Context detected
→
Related items surface
→
"You might need this"
Moment 5
"Pull up that design I liked"
Saved a screenshot of a dark UI with editorial typography. Can't remember when or where. Describe it and the system returns images by visual similarity.
Describe it
→
Image embeddings
→
Visual similarity
→
Thumbnail grid
These five moments define the entire product. If retrieval works, capture method almost doesn't matter — because anything you throw at it becomes findable. Smart retrieval, dumb capture.
Part 2 — The right form factor
PWA: Desktop + Mobile, One Codebase
Progressive Web App is the answer. One Next.js codebase, deployed to Vercel (which you already have), installs on every device as a standalone app. No App Store needed.
The killer feature is mobile share sheet. Reading an article in Safari → hit Share → "Save to Library" → done. Under 3 seconds. Camera access for book scanning. Offline browsing via service workers. Start as a web app, graduate to installable PWA with a manifest file and service worker.
Next.js → Vercel → add manifest + service worker → installable everywhere. You can add an "Install App" button in the UI.
Part 3 — How things get in
Capture: Fast, Dumb, Everywhere
Capture requires zero thought. You just put things in. OCR, summarization, embedding, tagging happen after.
Camera / Screenshot
Your #1 intake. Book pages, whiteboards, designs. Claude Vision OCRs text, describes images.
URL / Share Sheet
Fastest on mobile. Share any URL → scrape, summarize, tag, embed. Original URL preserved.
Dropbox Sync Folder
Bulk imports. Drop files in a Dropbox folder → edge function processes automatically via webhook.
Quick Note
Your own ideas. Type a thought — gets embedded alongside everything else. Your thinking becomes searchable.
Drag & Drop (Desktop)
Power user flow. Drag images, PDFs, text files onto the window. Batch processing.
API Endpoint
For later. REST endpoint for Zapier, Make, or scripts. Signal analyzer pipes in directly.
Part 4 — Under the hood
Architecture: The Stack
Everything runs on tools you already know. No new infrastructure.
1
Next.js PWA Frontend
Search bar is the primary UI. Grid view for browsing. Detail view for reading. Camera view for scanning. Drag zone for imports.
Next.js + Vercel + Service Worker
↓
2
Supabase Backend
items (raw content + metadata), insights (extracted nuggets), tags (many-to-many), vector column via pgvector. Auth + file storage included.
Supabase + pgvector + Storage
↓
3
Processing Pipeline
Edge Function triggered on insert. Routes by type: URL → scrape. Image → Claude Vision. PDF → extract + chunk. All paths end the same: embed, tag, store.
Edge Functions + Claude API
↓
4
Search Layer
RPC function: embed query → vector similarity → ranked results across all types. Filing cabinet → research assistant.
pgvector cosine similarity + RPC
↓
5
Connections Engine (v2)
Nightly job. Finds high-similarity matches between new and existing items. Your library starts thinking for you.
Scheduled Edge Function + Claude
Part 5 — The data model
Schema: Simple and Expandable
Three core tables. New content types = zero schema changes. Just a new `type` value.
| Table | Key Columns | Purpose |
|---|---|---|
| items | id, type, title, content, source_url, file_path, summary, embedding, created_at | Every piece of content. One row per item. |
| insights | id, item_id, text, embedding, created_at | Extracted nuggets. A book = 40 insights. |
| tags | id, name + junction table | Flat tags, many-to-many. |
| connections | item_a_id, item_b_id, similarity_score | Auto-discovered relationships. |
Infinitely scalable: adding audio transcripts = zero schema changes. New `type`, new processing branch, everything else works. Vector embeddings don't care what produced them. 100 items or 100,000, same query speed.
Part 6 — Build plan
Four Phases
Get search working first, then add capture methods one at a time. Each phase delivers something usable.
Phase 1 — Foundation (Week 1-2)
Database + Search + URL Paste
Enable pgvector. Create tables. Build search RPC. Minimal UI: search bar + "paste URL." End state: paste a URL, find it by asking a question.
Phase 2 — Visual (Week 3-4)
Image Upload + OCR + Grid
Drag-and-drop images. Claude Vision for OCR + description. Visual grid for browsing. Screenshots and book scans become searchable.
Phase 3 — Mobile (Week 5-6)
PWA + Share Sheet + Camera
Web manifest + service worker. Share target for mobile. Camera capture. Library in your pocket. Capture anything, anywhere, 3 seconds.
Phase 4 — Intelligence (Week 7+)
Connections Engine + Dropbox Sync
Nightly connection discovery. Dropbox folder sync for bulk. Library starts thinking — surfacing connections, reminding you of things you forgot.
The big insight
Why This Gets Better Over Time
Most tools decay — the mess grows faster than the value. This system is the opposite. Every item makes every other item more findable. A bookmark about retention loops becomes more useful when you later add a book chapter about habit formation — the connection engine links them.
And because it's built on Supabase + Next.js, it shares infrastructure with Portal, Simple, and your automation stack. One Supabase instance. One deployment pipeline. One set of skills that compound across everything.
Start with Phase 1. Paste 10 URLs. Search by concept. If that moment feels like magic, keep building. If it doesn't, adjust search before adding more capture methods. Retrieval quality is the only metric that matters early on.
Active
Queued
Studying
Automation & Delegation
Leverage
Email → GWS
Step zero. Makes everything programmable.
Payment Tracker
Gmail → Sheets ledger → burn rate.
Calendar Engine
SOW, payments, cadences → auto GCal.
Claude Skills
Invoice parser, digest, audit, ingester.
Products
Build
Simple
AI deck gen. Sales weapon → SaaS.
COA Portal
Client hub. Retainer defense.
Knowledge Engine
Second brain. Ingest anything, search by concept.
Agency
Revenue
Retainers
Nike, Pepsi, Gatorade.
DTC Outreach
Proof the system works.
Shopify Sites
Performance, conversion.
Finance
Money
Burn Rate Dashboard
See everything. Decide with numbers.
Google Workspace
Automation backbone.
Systems
Backend
Supabase Sync
One sprint. Three products unblocked.
Audio & Dev
Chain complete.
Studying
Growth
GWS APIs
Learn by building the tracker.
Claude Skills
AI team, one skill at a time.
AI / LLM
Powers Simple + Knowledge Engine.
Supabase + pgvector
Products + Knowledge Engine.
Financial Modeling
Know what to track first.
Shopify + Loyalty
When DTC gets real.