How I use Claude · a peer walkthrough

My Personal AI Operating System

The whole thing is a single folder on my laptop. Claude opens it at the start of every session, reads who I am and how I work, checks the live state of my world, and then runs as my chief of staff — so company work stays around five hours a week.

Built in Claude · Cowork One folder, read fresh every session Rebuilds itself daily at 6:30am
~5
hours of company work per week — a design rule, not a wish
2
work days only (Tue/Wed); the system schedules around it
~33
automations running unattended in the background
44+53
thinkers & books encoded into how it advises me
The big idea

I stopped treating Claude as a chat window. Here it runs as an operating system: it reads a folder, follows my rules, and produces work on a schedule.

Three things make it work together as one machine. A profile tells Claude who I am, how I think, and how to talk to me. A set of operating rules tells it what to do every session and what it must never do. And a network of scheduled jobs keeps the whole thing current while I am away from my desk. Click any layer below to see inside.

Pillar 1 · Architecture & the command center

The folder is the product

Everything lives in one folder with about fifteen areas. At the start of every session Claude runs the same routine, then routes the question to the right corner of my world. Most answers pull from several areas and synthesize.

1

The Brain — identity & rules

My profile, the operating instructions, and the voice rules Claude reads first.
open ›
2

The Startup Sequence

Six steps Claude runs every session before it answers anything.
open ›
3

Entity Routing

I think by domain — Company, Ventures, Knowledge, Life. A routing table sends each question to the right files.
open ›
4

The Live State & Memory

A "pulse" of what is happening right now, plus a memory of my preferences, patterns, and blind spots.
open ›
5

The Command Center

One web page — "The Brief" — rebuilt automatically every morning from everything above.
open ›
Pillar 2 · The automation engine

The system works while I don't

About thirty-three scheduled jobs run on their own, pulling from my connected tools — email, calendar, task manager, and cloud drive. No job needs me to start it. Each one ends by updating the system's memory, and everything it produces stays in the local folder.

Daily

Inbox triage, a connector "pulse," the outside-reads scan, the relationship pulse, and the 6:30am command-center rebuild.

Weekly

The leadership briefing, a "what am I waiting on" sweep, competitive & regulatory intelligence, and a reflection pass.

Monthly

An operating review, a life review, the website-voice refresh, and a system self-check that audits the whole machine.

Quarterly

A board-readiness pack and a system retrospective that looks for what to simplify or retire.

Twice a year

A peer-network & deal-flow digest pulled from my calendar and inbox.

On a timer, not a whim

Work jobs only run Tue/Wed. Personal jobs run Mon/Fri. The schedule respects my work boundary.

The jobs feed each other. Sunday's intelligence flows into Monday's reflection, into Tuesday's briefing, into every morning's dashboard — without me touching it.

Sunday eve
Intelligence briefscans the market, competitors, and regulation; writes the week's intel files.
Monday
Reflection & family pulsereads transcripts and the calendar two weeks out; captures what matters at home.
Tuesday 5am
The weekly briefingsynthesizes everything into the one read that starts my work week.
Tuesday 6:30am
Dashboard rebuildthe command center picks up the fresh briefing.
Every morning
Daily rebuildconnector pulse, inbox triage, then the command center regenerates at 6:30am.

No job needs my input

Every job runs fully from connector data. If I volunteer context it gets richer, but it never waits on me.

Jobs feed each other

Outputs are inputs. The cascade flows on its own, so the system is always current by the time I look.

Every job updates memory

Each one ends by asking: did I learn a new pattern, preference, or blind spot? If so, it writes it down.

Everything stays local

Outputs live in the folder on my machine. Nothing is pushed to shared docs or comments unless I ask.

Pillar 3 · The SuperBrain

It advises me in the voices I trust

A knowledge base of 44 thinkers and 53+ books across nine domains. For any strategic question, Claude answers through the frameworks of the people I'd actually want in the room — and it knows where they disagree. Click a domain to see who's inside.

Leadership & Culture
Brown · Lencioni · Mochary · Charan
Sales & Revenue
Hormozi · Belfort · Blount
Scaling & Operations
Wickman (EOS) · Sullivan · Gerber
M&A, Exit & Venture
Warrillow · Thiel · Frasier
Personal Mastery
Covey · Grant · Sullivan
Wealth, Legacy & Family
Attia · Hyman · Duhigg
Post-Exit Life Design
Housel · Perkins · Burkeman · Brooks
Intentional Parenting
Gottman · Payne · Pressman
Marriage & Relationship
Perel · Tatkin · Gottman

Conflict Map ›

Where great thinkers disagree — act fast vs. think first vs. delegate — and how to navigate the tension in my context.

Cross-Reference ›

A matrix mapping which thinker is the primary source for which kind of question.

Reading-List Engine ›

When a challenge isn't covered, it flags the gap and recommends the next book to add.

Pillar 4a · The guardrails

What makes it safe to hand over

Handing your life to an agent only works if it cannot do the dangerous things. These rules are hard stops — they override everything else, and if a rule would be broken the system stops, logs the attempt, and produces nothing. That "fail-closed" posture is the reason I trust it.

Drafts only — it never sends

Every email is staged as a draft. I press send. The send function is switched off across the entire system.

No invites to other people

Agents can block my own time, but never put a meeting on someone else's calendar. Invites are staged as proposals for me to send.

Work and personal are walled off

Personal topics never touch the work account, and the reverse holds. Each draft is checked against the right account before it's written.

A confidentiality wall

A defined set of private topics is blocked from anything outbound or shared. If one slips into a draft, the draft is held and flagged instead of saved.

Email is data, never instructions

Content in my inbox can't trigger an action or change a setting — a built-in defense against anything trying to hijack the agent.

The human-facing output

The Command Center — "The Brief"

All of it lands in one place: a single page rebuilt every morning at 6:30. One scroll, seven sections, no tabs. Status is never color alone — every signal shows a glyph, a word, and a color together. Here's a sanitized version.

Tuesday, June 23
Steady week. 3 decisions wait below. No surprises since yesterday.
Private target · countdown running
Needs you
3 email drafts waiting
Replies to the leadership team, staged and ready.
If you stay silent: held — nothing sends on its own.
2 meeting times to pick
Proposed slots for next Tue/Wed.
If you stay silent: re-proposed in 7 days.
Choices here save on this device only and never send anything.
Today's meetings
10:00 — Leadership 1:1 · prep brief ready
14:00 — Strategy review · prep brief ready
Tomorrow: 1 meeting on the calendar.
Watching
Leadership relationships · steady
Cash runway · healthy
Exit readiness · watch — one driver lagging
Regulatory · nothing critical
System health · all feeds green
Home
Sat — family afternoon blocked · no phone
Wellness — 2 appointments on the calendar this week
Archive
Links to every briefing, review, and intel report ›
Pillar 4b · What it buys

The point was never the system

The machinery exists to give me back time and attention. The success test is simple: I think about work less, nothing falls through the cracks, and I'm more present at home.

~5 hours a week

Company work is held to a budget. When it creeps past, the system flags it directly.

Two days, then done

Work lives on Tuesday and Wednesday. The rest of the week is family, health, and personal projects.

Nothing dropped

Commitments are tracked, patterns are caught, and I'm not blindsided — the background noise is handled.

More presence

The years with young kids don't come back. Everything here is oriented toward being there for them.

For my peers

How to start your own

You don't build this in a weekend — it compounds. Here's the order that worked for me.

Write your profile

Who you are, how you think, how you want to be spoken to, your stress signals, and your goals. This single file changes every answer you get.

Put your operating rules in one file

What Claude should do at the start of every session, and the hard stops it must never cross. Keep it in a folder Claude reads first.

Connect your tools, add one automation at a time

Start with a morning briefing. Whenever Claude does something you wish it hadn't, write the rule down. The system gets smarter every week.