I Taught My AI to Run 9 Projects

A storytelling playbook: narrative spine, identity framing, content plan — researched across 8 master builders and the entire AI content landscape
6 March 2026
Projects
9 active
2 paying, 3 pilots
Output
53 reports
5 weeks
Dunbar
71/150
tracked with context
Risk
No narrative
invisible output

I. The Problem

Eric runs 9 active projects, 2 paying clients, 3 live AI pilots, and a research engine that’s published 53 reports in 5 weeks. Nobody knows. The world sees “AI consultant.” The reality is a persistent AI operating system running an entire practice at 5–10x typical output. The story hasn’t been told.

This playbook synthesizes global storytelling research — 8 master builders, anti-pattern analysis, HK market scan, identity framing analysis, format benchmarking — into one narrative spine, a meeting toolkit, and a 7-day content plan.

II. What Actually Exists

Hard Output (5 weeks: Feb 1 – Mar 6)

CategoryCount
Research reports (deployed, live)53
CLI commands24
Skills (research/automation)21
Paying clients2 (HK$40K/mo)
Live WhatsApp pilots3
Production PRs to client codebase2 (same evening)
Eval frameworks2
Math questions transcribed + verified225
Contacts in CRM (Dunbar-tracked)71/150
Meetings transcribed + analyzed30+
Field-tested prototype1
Fine-tune (identity reprogramming)1 (10.5 seconds)

Invisible Infrastructure

1. PCRM — Persistent context across 71 contacts. The AI remembers what Penny said 3 weeks ago when writing tonight’s briefing.

2. Diff engine — WhatsApp, calendar, Cursor sessions, action items diffed against last run. Nothing falls through cracks.

3. Checkout system — Every hour attributed to a project with cost, to the dollar.

4. Research pipeline — Question → 50-source HTML report → deployed in 2–4 hours.

5. Meeting intelligence — Transcribe → segment → hallucination detection → structured insights → state updates.

Compound Capabilities
Built for AReused in B
Sourcy eval frameworkTalent Coop math eval
WhatsApp diff engineDonna pilot infrastructure
Research pipeline (53 reports)Client deliverables
OpenClaw deploymentDonna WA server for B2B
PCRM contact trackingConsulting at scale
The Honest Tensions
  • • Revenue: HK$40K/mo consulting, $0 product
  • • 9 projects = context-switching tax
  • • 19-day content drought
  • • Closing is slow (weeks of relationship building)
  • • Token cost: $473/week on AI compute

III. Global Storytelling Research

The Masters

BuilderSpineMechanism
Elon Musk“Accelerate sustainable abundance”Master Plan docs — every company = a phase
Greg Isenberg“Community-first businesses, $0 capital, free playbook”Podcast clips → Reels, give away the recipe
Pieter Levels“One person, one laptop, multi-million dollar portfolio”Revenue screenshots, radical transparency
Guillermo Rauch“I build the tools that ship the web — live, in front of you”Dogfooding as content
Sahil Lavingia“I failed at a billion-dollar company and found something better”One iconic failure essay
Jack Butcher“How is value created, transmitted, and verified?”Visual identity IS the connective tissue
Dan Koe“HUMAN 3.0 — a map to the top 1%”Framework as chapters
Matt & Ari“Building a 9-figure startup from our living room”Cinematic build-in-public documentary

Identity Framing

Nobody 10 steps ahead calls themselves “consultant.” The universal pattern: Founder + portfolio as menu.

PersonExact Language
Matt Gray“Founder & CEO, Founder OS”
Steven Bartlett“Founder: Steven.com. Our Studio: X. Our Fund: Y.”
Hormozi“Founder Acquisition.com” — verbs: “buying and building”
Gary Vee“I day trade attention and build businesses”
Codie Sanchez“I own a portfolio of Main St companies”
Shaan Puri“founder, investor, & creator”

Language IN: Founder, Builder, Operator, Portfolio, HoldCo, “I own,” Systems.
Language OUT: Consultant, Entrepreneur, Agency owner, Serial entrepreneur, CEO (alone).

What Cuts Through vs What Fails

What Cuts Through

  • Revenue receipts with real numbers
  • Named artifacts people can visit
  • Agent-as-protagonist storytelling
  • Show the failure rate honestly
  • Hardware + cost specificity ($473/wk)
  • Process, not announcement
  • Compound narrative: “A → B”

What Fails

  • “I replaced my team with AI” without specifics
  • Buzzword soup (“agentic workflows”)
  • $97 prompt courses
  • “Delighted to announce” posts
  • Stock-photo founder + generic bio
  • 52% reduce engagement on suspected AI content3
  • 82% negative sentiment on “slop”2
The HK White Space

Nobody in Hong Kong is doing builder-transparency content. The scene: corporate professors, startup pitches, community organizers. Zero people showing “here’s my stack, here’s what the AI does, here’s the daily workflow.” The slot is empty.

Video Reference: Matt & Ari

Matt Ao + Ari Bapna, co-founders of Eden (eden.so). CEO = Dan Koe. ~17K subs. Cinematic startup vlog — documenting building a real product from their living room.

The format: 10–15 min episodes. Dedicated videographer (Jason). Real work on camera: YC demos, late-night bug fixes, rebuilding after 80K users. One playlist: “Building in Public.”

Why it resonates: The content IS the work. Not content about content. Not advice — documentary. High production quality but genuine. Recurring location anchor (“our living room”). Narrative arc across episodes (launch, aftermath, pivot, rebuild).

IV. The Narrative Spine

Selected Frame: “I Taught My AI”

Spine: I taught my AI to run 9 projects. Here’s what happened.

Identity: Founder. Builder. NOT consultant. The verb is BUILD, not advise.

Bio format: “Founder. I build AI operating systems — mine first, then clients.”

Content series: “I taught my AI to…” — infinite, serializable, each post standalone AND compounding.

Why This Wins

  1. True — every word backed by artifacts
  2. Active — “I taught” implies mastery, not just usage
  3. Serializable — “I taught my AI to…” is infinite content
  4. Differentiated — nobody frames it as teaching; everyone frames it as using
  5. Cinematic — the image of someone training an AI that runs 9 things
  6. Defeats scatter objection — teaching implies deliberate curriculum, not ADHD

Aesthetic North Star

Matt & Ari (@matt-ari): warm minimalism, cinematic but real, calm authority. Not hustle. Not loud. Clean, tasteful, quiet confidence. The building IS the content.

V. Story Architecture

Origin

After co-founding GenieFriends and building a 9,000-member community, I watched my startup close down while AI changed everything around me. I had 68 real relationships, zero product revenue, and one question: what if I taught an AI to remember everything I do?

Five weeks later, it writes me a daily briefing at 8am. It tracks 71 relationships. It produces 50-source research reports in hours. It runs eval frameworks that prove which chatbot actually works. And it does this across 9 active projects — 2 paying clients, 3 live pilots — without dropping a single follow-up.

I didn’t hire a team. I taught an AI.

Five Proof Points

#“I taught my AI to…”The Number
1…write me a daily briefing from 300 WhatsApp messages, my calendar, and all work sessions. Every morning at 8am.71 relationships, zero missed follow-ups, 5 weeks
2…turn one question into a 50-source research report, stress-tested with failures and counter-examples. Deployed live.53 reports in 5 weeks10
3…evaluate 3 competing chatbots against 14 real customer personas. Not opinions — data.67% → 100% pass rate across 5 iterations
4…ship production code the same evening a school VP said “we want to trial.” Backend + frontend, deployed by morning.685 lines, same evening
5…remember every conversation, meeting, and action item across 9 projects. It knows what we discussed 3 weeks ago.24 commands, 21 skills, 71 contacts

The Tension

I run 9 projects and people think I’m scattered. The truth is every project teaches the AI something new. The eval framework I built for a sourcing company became the methodology for an education company. The WhatsApp integration for my CRM became the pilot infrastructure for 3 B2B clients.

The portfolio doesn’t scatter — it compounds. But nobody can see that from the outside. That’s the story I haven’t been telling.

The Stakes

Right now in Hong Kong, every startup is pivoting to “AI agents.” Every SaaS company is bolting on a chatbot. And the people I meet — founders, CTOs, business owners — correctly recognize that 90% of it is noise. They’re looking for someone who can show them what’s real. Not a demo. Not a pitch deck. The actual daily operating system of a person who runs on AI.

Key data: Gartner projects 40% of AI agent projects will be cancelled by 2027.1 Human-generated content receives 5.44x more traffic than AI slop.4 The bar for credibility is higher than ever — and rising.

VI. The Pitches

60-Second (Meeting Version)

I build AI operating systems. Right now I’m running 9 active projects — two paying clients, three live pilots, and a research engine that’s published 53 reports in 5 weeks.

The way it works: I taught an AI to track all my conversations, meetings, and projects with persistent memory. Every morning at 8am, it writes me a briefing — what changed overnight, who I need to reply to, which projects need attention. It’s like having a senior partner who never forgets.

When a client asks me to evaluate their AI chatbot, I build an eval framework that tests it against real customer conversations. When a school says they want to trial, I ship production code the same evening.

Everyone talks about AI agents. I just show people what mine does every day.

15-Second Hooks

Content hook: i taught my AI to write me a daily briefing from 300 whatsapp messages, track 71 relationships, and run 9 projects. my team costs $0/month.

Meeting hook: I taught my AI to run my entire practice. Let me show you what it wrote me this morning.

Meeting Toolkit

“Show me” moment: Pull up research.ericsan.io on phone. Then pull up today’s daily briefing (redacted).

Follow-up: Send LinkedIn carousel + research.ericsan.io link.

Objection Handling

They SayYou Say
“Everyone does agents”“40% get cancelled.1 Mine runs in production daily. Let me show you this morning’s briefing.”
“How is this not a demo?”“Pick any of these 53 reports.10 20–50 sourced references each. That’s output, not demo.”
“What’s different?”“Persistent memory. Most AI forgets between sessions. Mine remembers every conversation across 9 projects.”
“Nine projects = scattered”“They compound. The eval from a sourcing company became the methodology for an education company.”

VII. Content Plan

Flagship: LinkedIn Carousel

10 slides, 1080×1350. Matt & Ari aesthetic: warm minimalism, all-lowercase, generous whitespace.

SlideContent
1i taught my AI to run 9 projects. here’s what happened.
2everyone talks about AI agents. 40% of projects get cancelled (Gartner).1 most demos are vaporware.
35 weeks ago i asked: what if i taught an AI to remember everything i do?
4i taught it to brief me. screenshot of redacted morning briefing.
5i taught it to research. screenshot of research.ericsan.io. 53 reports, 20–50 sourced references each.
6i taught it to evaluate. 3 chatbots, 14 real customer conversations, data not opinions.
7then everything compounded. eval for client A → methodology for client B. WhatsApp engine → pilot infra for 3 B2B clients.
85 weeks. 2 paying clients. 3 live pilots. 53 reports. 24 commands. 21 skills. 71 contacts tracked.
9AI doesn’t replace you. it gives you persistent memory. that’s the actual superpower.
10want to see the system? link in comments.

7-Day Sprint

DayPlatformHook
1 (Sat)LinkedInCarousel: “i taught my AI to run 9 projects.”
2 (Sun)Threads“i taught my AI to write me a briefing every morning at 8am” + screenshot
3 (Mon)LinkedIn“a school VP said ‘we want to trial.’ i shipped code the same evening.”
4 (Tue)Threads“everyone asks how i handle 9 projects. the answer isn’t discipline. it’s memory.”
5 (Wed)LinkedIn“i taught my AI to evaluate 3 chatbots with 14 real conversations.” + screenshot
6 (Thu)Threads“53 research reports in 5 weeks. 20–50 sourced references each. no templates.”
7 (Fri)LinkedIn“what i learned building an AI operating system for 5 weeks”

Content Flywheel

Daily briefing → “what my AI caught that I missed.” New reports → “i taught my AI to research X.” Client milestones → “i taught my AI to solve X.” New capabilities → “i taught my AI to…”

Minimum cadence: 2 posts/week (1 LinkedIn, 1 Threads). 30 min/week using existing material.

VIII. Video Path

The Matt & Ari Model

The highest-impact format long-term is cinematic build-in-public documentary — filming the actual system in action.

Format: “A Week Inside My AI Practice” — Cursor IDE, daily briefing rendering, research pipeline running, WhatsApp diffs appearing, eval results. 10–15 min episodes. Narrative arc across episodes (like Matt & Ari: “building, launching, pivoting, rebuilding”).

What to show: Screen recordings of the AI working in real-time. The briefing appearing. The eval running. The report deploying. Narrated with the honest tension — what works, what breaks, what compounds.

Why: This is what nobody in HK is doing. The content IS the work. Not content about content. Not advice. Documentary proof that one person can run 9 projects with an AI that remembers everything.

Prerequisite

Camera setup or high-quality screen capture workflow. This is a Q2 investment, not this weekend’s move. LinkedIn carousel ships first.

IX. Critical Assessment

Authenticity

Every claim backed by existing artifacts. Tensions are honest. Nothing stretched — 53 reports sounds unbelievable but all are live at research.ericsan.io.10

Differentiation

Nobody in HK. <5 globally at this transparency level. “Persistent memory” is genuinely different from “I use AI to code faster.” The compound narrative (A → B) defeats the scatter objection.

Comparable solo operators: Aaron Sneed runs 15 custom GPTs from his couch7; Rushiraj Jadeja runs 5 AI agents on a Mac Mini M28; Felix Craft’s OpenClaw agent earned $14,718 in 2.5 weeks.9 Eric’s differentiation: persistent memory across 9 projects, not isolated agents.

Sustainability

Every week produces new material. The spine grows stronger as more projects ship. Compatible with convergence: “I explored 9, converged on 3, and the AI made both phases possible.”

Verdict

Spine: I taught my AI to run 9 projects. Here’s what happened.

Identity: Founder. Builder. I build AI operating systems.

Hook (15s): i taught my AI to write me a daily briefing from 300 whatsapp messages, track 71 relationships, and run 9 projects. my team costs $0/month.

Ship first: LinkedIn carousel (this weekend). Doubles as content + pre-meeting material for Hampton Thu.

#1 risk: “Nine projects = scattered.” Always pair with compound story.

What makes it undeniable: A client acquired through the content itself — the AI that runs the practice also generates the leads.

Sources & References

[1] Gartner (2025). “Predicts 2025: AI Agents Transform Enterprise Operations.” 40% of AI agent projects will be shelved or cancelled by 2027 due to escalating costs and unclear ROI.
[2] Brandwatch (2025). “AI Slop Sentiment Analysis.” “Slop” mentions grew 200%+ in 2025; 82% carry negative sentiment.
[3] MainTouch Consumer Research (2025). 52% of consumers reduce engagement when they suspect content is AI-generated.
[4] NP Digital / SEO research (2025). Human-generated content receives 5.44x more organic traffic than AI-generated equivalents.
[5] LinkedIn carousel engagement benchmarks (various, 2024–2025). Carousels generate 3.7x text-post engagement; 596% higher reach than text-only posts.
[6] Mark Firth, LinkedIn. Viral post roasting “AI guru” archetype — the exact anti-pattern this playbook avoids.
[7] Aaron Sneed, Business Insider (2025). Solo founder running 15 custom GPTs across his business operations from his couch.
[8] Rushiraj Jadeja (2025). Runs 5 AI agents on a Mac Mini M2, documenting the full stack publicly.
[9] Felix Craft / OpenClaw (2025). Autonomous agent earned $14,718 in 2.5 weeks — verifiable on-chain revenue.
[10] research.ericsan.io. 53 live research reports, deployed Feb 1 – Mar 6, 2026. Each with 20–50 sourced references.