8-week cohort · Session 3

Your AI Operating System

The 5 layers behind a second brain — and how I actually use mine. Live.

Egils — host

Today · ~90 min

What we'll do today

① Check-inwhat you built this week
② Why we do thisradiology · travel · why build your own
③ The systemthe 5 layers — where it lives, how to store it
④ See it live3 demos from my own brain

Goal: leave having built your first skill — we do it live in the last 10 minutes.

Every week · the engine

First — what did you build this week?

Each week we go round and share what we shipped. Not to impress — to keep moving.

💡 Why this part matters most: Accountability turns intention into momentum — you move because the group will see. Sharing hands everyone real data and a live, working example to react to — that's where the best ideas, fixes and intros come from. The talking is the system.
No one gets left behind

We build this together

  • The community effect. Watching each other move is what carries you through the empty days — momentum you can't manufacture alone.
  • Be proactive — use us. Egils & Gianluca for 1:1 chats, calls, brainstorming, real consults. Don't sit stuck.
  • This is rare. Future cohorts won't get this level of direct access to us. Use it now.

Ask. Share. Pull each other forward.

Your turn — 60–90s each

So — what did you build or learn this week?

One win, one stuck point. We drop it into the system, live — and the group jumps in.

Why we do what we do · the "1997 of AI"

Same scan. Every era, less friction — now, better answers.

X-ray film on a lightbox
~1990s–2000s · FILM
You carried printed film doctor to doctor — non-urgent scans driven between hospitals by taxi.
Speed
Access
Cost
Insight
Digital PACS workstation
2010s–2020s · DIGITAL
Filmless — by 2007 every NHS England trust. The scan is in the system in minutes, everywhere at once.
Speed
Access
Cost
Insight
AI-annotated scan heatmap
2020s–2040s · AI
AI reads the scan & flags findings. 76% of FDA medical AI is radiology; a 100k-woman trial: AI caught more cancers (81% vs 74%). And we're just at the start.
Speed
Access
Cost
Insight

We're at the "1997 of AI" — and it's arriving faster than the internet did. Build what you own now.

Sources: filmless by 2007 — English National PACS Programme · film-by-taxi — NIH/NCBI · 76% of FDA medical AI = radiology — JAMA Network Open 2024 · more cancers caught, 81% vs 74% — MASAI, Lancet 2026 · AI faster than the internet — St. Louis Fed 2025. Era-1 photo CC BY-SA 4.0 (Infrogmation/Wikimedia) · Era-3 CC BY 4.0 (Rahman et al., Diagnostics 2024).

Why we do what we do · you've lived this one

Booking a trip. Same seat — every era, less friction.

📞
~2000s · TRAVEL AGENT
You called, waited, paid a markup for the same flight. Agent jobs later fell ~70%.
Speed
Cost
Choice
💻
2010s–2020s · ONLINE
You book it yourself in minutes — same seat, no markup. Most trips are now booked online.
Speed
Cost
Choice
🤖
2020s–2040s · AI AGENT
You say "plan my trip" — AI books it. Analysts (IDC) expect up to 30% of bookings by AI agents by 2030. Just the start.
Speed
Cost
Choice

Same trip, same seat — the people who got online early own travel today. AI is that moment again.

Sources: travel-agent jobs −70% & self-serve online — TravelPerk (US BLS) · AI agents ~30% of bookings by 2030 — CNBC 2025, citing research firm IDC (International Data Corporation) · AI adopted faster than the internet — St. Louis Fed 2025.

So why build YOUR own?

The ones who build don't get left behind

Radiology and travel both got compressed. We don't know exactly what 2040 looks like — but we know who wins: the people who learn the new tool early and make it theirs.

  • AI skills already pay. Workers with AI skills earn a 56% wage premium — up from 25% just a year earlier.
  • Early adopters pull ahead. In real workplaces AI lifted output +14% on average — +34% for the newest workers — and cut writing time 40%.
  • We've seen this movie. When search replaced the Yellow Pages, those who moved online found better service for less — by 2011 most had dropped the phone book. AI is that switch again.

Build your own AI OS → you understand it, shape it, own your future — instead of being used by it.

Sources: 56% AI wage premium — PwC Global AI Jobs Barometer 2025 · +14% / +34% productivity — NBER 2023 (Brynjolfsson, Li & Raymond) · 40% faster writing — MIT / Science 2023 · ~70% dropped print directories by 2011 — Yellow Pages (Wikipedia / industry data).

From last week → the full picture

The same brain you saw — now the full 5

① CONSTITUTION · governs everything, loaded first
📸 🎤 📨 📞 💼 📊 🧾 🏦 💊 😴 🏃 📚 🎧 📺 🔖 📄 🌍 📓 MEMORY SKILLS TOOLS outer ring = more tools & sources, over time … Your Second Brain 📝Notes 🎙️Calls 🧬DNA 💡Ideas ✍️Write 🔎Research 📋Plan 🎯Prioritize 📧Gmail 📅Calendar 📁Drive Garmin
⑤ FOCUS · what you do now — where it all turns into action
Layer 1 · loaded first

Constitution — who your AI is

Read at the start of every session — it's why your AI sounds like you, not generic. Four things go inside:

Identitywho you are · your story · your voice
Values + North Starwhat you're aiming at, and why
Rules + boundariesa few concrete "always / never"
Routingwhere things live · what to load when

Lives in whatever file your tool uses — ⚙️ OpenClaw config · 📄 SYSTEM.md · 🟣 CLAUDE.md. Different tools, same job.

Layer 1 · one file, or many?

How to hold it

🗂️ One fileSimplest — start here. Identity + values + rules in a single file.
🗃️ Many files (modular)Scales: a short root "router" + topic files that load on demand. Anthropic's own advice — keep the always-loaded file lean (<200 lines).
🔑 "Won't a big constitution cost a fortune every message?" No — prompt caching: reload the same context and Claude caches it, so repeat reads cost ~90% less (0.1× the price) and start faster. A rich constitution is cheap to keep loaded.

And Skills load on demand (only a short description until you use one) — so you can have many without bloating context.

Sources: Claude Code memory (lean, modular, <200 lines) · prompt caching (cache reads = 0.1× input) · Agent Skills (progressive disclosure).

Layer 1 · my real setup

Mine = 6 files, each one job

🗺️ CONTEXTThe map — what lives where, which command opens which file.
🧭 NORTH STARVision + this quarter's P1 / P2 / P3 → the piece that sorts everything. Every decision: "does it serve P1?"
🟢 NOWRolling focus, 7 spheres. I review it 2–4×/month — but the AI jots updates itself.
⚙️ PROTOCOLS · 🫀 SOUL · 👤 USERMy skills + tools list · personality · who I am.

All auto-load each session. This "lean router + load-on-demand" shape is exactly what Anthropic + practitioners recommend — one canonical source per topic, no padding.

Layer 2

Memory — what your AI knows

Pure life data, kept in folders. Mine = 7 spheres:

🏃 Body🧠 Mind👨‍👩‍👧 Family🤝 Network💼 Business💰 Money🌿 Environment

7 is a starting point, not a rule — you shape it to your life:

  • Heavy on business? Split it into clients, offers, content.
  • Only about you? Drop business entirely.

In practice: tell your AI "remember I'm training for a marathon" → it writes that into Body → next time you ask about food, it already knows. Add a folder only when it outgrows its parent — each builds their own.

Layer 2 · where it actually lives

Your data — own the files

✅ Source of truth: plain MD files YOU ownOn your own computer, or always-on (Raspberry Pi / VPS / hosting). Portable, vendor-lock-free, still readable in 20 years. The AI reads & writes these directly.
☁️ Backup + heavy files: cloud (Google Drive)A second copy — and the warehouse for big stuff: large PDFs, videos, whole books. Not the source of truth (sync breaks, formats lock you in).
💡 My setup: the books sit on Google Drive; my summaries live as MD in my Memory layer. The AI works the MD — Drive just stores the heavy originals.

Principle: "File over app" — Steph Ango, CEO of Obsidian: if you want digital things that last, keep them as files you control, in formats easy to read.

Layer 2 · folder structure

Steal a structure — then make it yours

🏛️ AIOS · 5 layers
01-constitution/
02-memory/
  body/  mind/
  business/  money/
03-skills/
04-tools/
05-focus/
99-archive/
📚 PARA
1-projects/
2-areas/
3-resources/
4-archive/
Organize by how actionable it is.
✏️ Minimal start
notes/
people/
projects/
archive/
Begin tiny — grow later.

No "perfect" structure. Pick one, start, adjust as you grow.

Examples & methods: PARA — Tiago Forte (Building a Second Brain). The trick isn't copying mine — it's keeping one system you actually maintain.

Layer 3 · the "now what?"

Skills — what your AI does

A skill is an instruction your AI follows — written once, run forever. It reads your Constitution (acts like you) and Memory (knows the facts).

📝 It's just text you save. Paste this to your AI: "Every morning, read my calendar and write a short brief of who I'm meeting and what to prep." — that's a skill.

My real, named skills — each is just one written instruction:

EvaluateEVAL7 · score any content 0–100 + fix-listEVALX · compare options & rank
Run my lifeIN7 · daily check-inNOW7 · refresh focusEXIT7 · loose-ends sweep
WorkFULL7 · call → profileFEED7 · feed → comments in my voiceSTABS7 · weekly scout
📄 Steal mine: my skills as ready-to-use templates → loai.wisdomblend.com/skills.txt. Open it, paste it to your AI, say "Egils uses these — which fit me?" — then build your own. More examples: prompts.chat (150k+★, any AI).
Layer 4 · the hands

Tools — what it's plugged into

🔌MCP = the "USB-C port for AI" — one open standard, 10,000+ connectors since Nov 2024. A skill uses tools: the tool is the hand, the skill is the know-how.
💬 CommunicationTelegramWhatsAppDiscordSignalGmail
📄 Docs & KnowledgeGoogle DocsGoogle DriveGoogle SheetsNotionObsidian
📅 Plan & TasksGoogle CalendarTodoistTrelloAsanaTickTick
🛠️ Build & AutomateGitHubGitHub Actionsn8nZapier
💳 Commerce & LifeStripeHubSpotGarminStravaApple

+ browser, web search, databases, design (Figma)… Start with 4–6. Add a tool only when a skill needs it.

Sources: modelcontextprotocol.io (MCP = "USB-C for AI") · live popularity PulseMCP · categories from MCP.Directory + Anthropic Connectors. Logos © their owners (via Simple Icons).

Layer 5

Focus — what you do now

📋 Projects & tasksWhat you're delivering — with deadlines.
🔄 Weekly reviewWhat moved, what's stuck, what's next.

And the part that keeps it alive: the feedback loop.

📥 SourcesWhatsApp · email · calendar · Garmin
🧭 Triageyour Constitution rules decide what matters
🧠 UpdatedMemory + Focus, every day

On an always-on machine (a Pi or server) it runs daily on its own; on a normal laptop you run it yourself — still works, just not while you sleep.

Layer 5 · give your AI a project shape

Without a structure, every project drifts

Give the AI one repeatable process and it runs every project the same clean way. No single right one — here are three:

🟢 Mine
Goal
Research
Present
Do
Analyse
📥 GTD
Capture
Clarify
Organize
Reflect
Engage
David Allen
🔁 Lean Startup
Build
Measure
Learn
Eric Ries

Pick one. Name it. Hand it to your AI.

Then wire it to a task tool the AI can reach (Todoist · Notion · Asana). + PDCA, Design Thinking. Sources: GTD · Lean Startup · PDCA (Deming).

Layer 5 · the heartbeat

The feedback loop — what keeps it alive

Without the feedback loop, your second brain slowly disconnects from your real life.

Tiago ForteCapture is the easy part. "Information only becomes knowledge when we put it to use." Notes you never revisit are dead weight.
David Allen · GTDThe weekly review is the "critical success factor" — the make-or-break habit that keeps the whole system trustworthy.
Memory scienceYou forget most new input within days (Ebbinghaus). Each review flattens the forgetting curve — the system must re-touch what matters.
Especially earlyYour goals & North Star shift fast. The loop updates not just notes — it re-tunes your Memory and Constitution.

Sources: Forte — Building a Second Brain · Allen — GTD weekly review · Ebbinghaus forgetting curve.

Feedback loop · build YOUR own

Theory is nice. Here's how to wire it up.

Pick a source, scan it on a schedule, let the brain digest it. Each one = a single cron job. Start with one.

📅 DailyScan your calendar → a morning brief of today's meetings — who they're with and what to prep.
📧 DailyScan email → a digest of what came in and went out, so the brain always knows what's happening around you.
🕷️ WeeklyApify scans posts & comments (your channel, or a competitor) → a weekly read on the business or a person.

One source → one schedule → one digest into Memory. That's a living brain.

Now — see it live

3 demos from my own brain

Real data. Things an empty ChatGPT simply cannot do — because it doesn't have me.

Live demo · 1 of 3 · Memory × Skills

It knows my body

💬 "Given my DNA and my Garmin trend, what should this week's training focus be — and is my weight trend putting the Gauja Trail 50K at risk?"
🧬 DNAgenetic markers in memory
+
⌚ Garminsleep, load, weight — live
🎯 One answerpersonal, specific, today
Live demo · 2 of 3 · Skills × Tools

It turns a thought into a plan

💬 "Make a Google Doc for my son: a SUP trip down the Mazā Jugla. Put-in at Tīnūži, wake-up time, distance, and exactly what to pack — board, prep, water temperature, clothes — so he knows precisely what and how."
📋 Planning skillbuilds the checklist
+
📁 Google Drivewrites a real doc
📤 Shareablesend it to him, done

📄 Plain doc 📊 Rich doc · table + images 🌐 Web page ✨

Same content my AI generated from one message — three formats. Pick for your audience.

Live demo · 3 of 3 · Skills × Tools × Research

It researches anyone — on its own page

💬 "Research [name] — pull their recent LinkedIn & social posts, articles, and what others say (via Apify). Write a positive, detailed profile, then publish it to a private, password-protected page with their photo. Make it shareable."
🔎 Research skill
+
🕷️ Apifyposts · articles · mentions
+
🌐 🔒 Private pageown subdomain · password-gated

🔒 See a real example — a cohort member's page ↗

One message in → a private page out. Then I ask them: "can I feature you?"

Your homework — everyone, the same one

Build your Morning Brief this week

📋 Paste to your AI: "Every morning, read my calendar and write a short brief of who I'm meeting and what to prep." → run it tomorrow → tweak it. Done = your first working skill.

Post in the group: "my skill this week is ___" — next session opens with you showing it.