The Second Brain Evolution Guide

BEYOND THE
KARPATHY BUILD

I started with a source-grounded LLM wiki. Then I turned it into an operating system for memory, projects, recorded calls, AI workers, task review, and proof.

Explore the 30 upgrades Download Markdown

00 / THE CORE SHIFT

The brain stopped waiting for me.

The original build preserved sources and turned them into linked knowledge. That gave my AI memory it could trust.

The most important evolution came later. My second brain now gives the right context to the right AI, tracks the work, checks the result, and saves what happened.

Memory became management.

01 / START HERE

The base architecture

Build trustworthy memory before adding agents.

raw/

Preserve the evidence

Store original notes, transcripts, clips, documents, and conversations without rewriting them.

wiki/

Synthesize the knowledge

Create source notes, concepts, entities, syntheses, and questions that link back to evidence.

control/

Govern the system

Use AGENTS.md for rules, index.md for navigation, and log.md for material changes.

02 / THE 30 UPGRADES

Each change solved a specific limitation.

The left side names the upgrade. The red panel shows what became possible after I installed it.

Processed-source lifecycle

Separate new sources, completed sources, and supporting assets.

Human operating areas

Add folders that match your personal life, business, projects, templates, and daily work.

Journal-grounded memory

Connect new reflections to prior entries and durable personal context.

AI conversation capture

Preserve important chats with their decisions, corrections, resources, and open questions.

Conversation audits

Find missing substantive chats, repair incomplete records, and avoid duplicates.

Source-of-truth reorientation

Require the AI to read the index, recent log, project rules, and relevant evidence.

Provenance and contradiction handling

Trace claims to evidence and preserve conflicts when knowledge changes.

Creator intelligence

Track creators, links, observations, contact fields, and captured content.

Participant attribution

Record hosts, guests, speakers, demonstrators, and framework owners.

Visual dashboards

Map key files, source context, decisions, and next actions.

Standard project workspaces

Give each project rules, orientation, tasks, and a dated session log.

Git synchronization and recovery

Add version history, secret checks, shared state, and recovery rules.

Automatic source processing

Turn new files into source notes, wiki updates, repaired links, and log entries.

Verified YouTube and web intake

Preserve exact identity, creator, dates, transcript status, links, and resources.

NotebookLM call intelligence

Route recorded calls into person or project notebooks, then reconcile them locally.

Exact artifact preservation

Save the exact document before or alongside external sharing.

An AI Chief of Staff

Assign one agent to own context, reconciliation, contradictions, strategy, and synthesis.

Proactive intelligence reviews

Schedule bounded reviews for strategy, revenue risks, new sources, and vault health.

Soul files for AI employees

Define identity, temperament, role, protections, relationships, boundaries, and failure mode.

Proactive context brokerage

Deliver the assignment, evidence, corrections, tools, destination, reviewer, and limits before execution.

Workforce operations

Separate queue and handoff management from canonical judgment and final review.

Sequenced agent installation

Install the owner, brain, Chief of Staff, work board, workforce operator, then specialists.

A live Task Command Center

Give work a brief, criteria, checklist, artifacts, blockers, reviewers, and receipts.

Privacy-scoped search

Query an approved manifest and return answers with source paths and freshness dates.

Brain DNA

Navigate domains, folders, notes, and recent records through an interactive hierarchy.

Governed dispatch and review

Require acknowledgment, progress, exact artifacts, named review, and completion receipts.

Least-privilege access

Separate vault scope, task access, connector permissions, public exports, and human approval.

Correction-to-skill learning

Encode corrections into skills, SOPs, souls, templates, or operating rules.

Complete reusable packages

Include the SOP, intake, discovery, assembly, schema, quality gates, and stop conditions.

A portable installer

Interview the new owner before creating a structure that matches their life and work.

03 / INSTALLATION ORDER

Install this in five stages.

Thirty upgrades installed at once create thirty new failure points. Build in order and test each layer.

01
Trustworthy memory
Raw sources, wiki synthesis, source notes, index, log, and processed-source tracking.
02
Personal operations
Journals, operating areas, conversation capture, project workspaces, and Git.
03
Better intake
Creator profiles, participant attribution, YouTube capture, call intelligence, and artifacts.
04
Proactive management
Recurring reviews, Chief of Staff ownership, souls, context brokerage, and governed review.
05
Productization
Hosted search, visual navigation, reusable packaging, and a portable installer.

04 / GIVE THIS TO YOUR AI

Audit your current second brain.

Copy this prompt into an AI that can read your vault or knowledge base. It will produce a staged gap analysis without changing your structure first.

Help me upgrade my current second brain into a source-grounded operating system.

Start by inspecting the existing structure. Do not delete, rename, or move files without my approval.

Create a written gap analysis against these capabilities:

1. Raw source intake and processed-source tracking
2. Source notes and traceable wiki synthesis
3. Human operating areas for life, business, and projects
4. Journal and AI conversation capture
5. Project orientation, task, and session records
6. Git synchronization and recovery
7. Creator and participant attribution
8. Recorded-call intelligence
9. AI Chief of Staff ownership
10. Soul files and proactive context brokerage
11. Governed task dispatch, review, and receipts
12. Correction-to-skill learning
13. Reusable system packaging
14. Privacy-scoped hosted search

For each gap, explain what is missing, what the change would enable, the smallest safe implementation, any privacy or approval risk, and how we will test it.

Then propose a staged installation plan. Wait for my approval before making structural changes.

05 / DEFINITION OF DONE

Your second brain has become an operating system when it can:

preserve original evidence;
trace claims to their sources;
recover decisions from conversations and calls;
orient an AI inside any active project;
deliver context before assigning work;
collect proof and route review;
learn from corrections;
protect private knowledge from public systems; and
continue without depending on one chat or one agent's memory.

Build the memory first. Add management second. Add more agents after both layers work.

Keep the complete playbook

Take the guide with you.

Download the full Markdown edition or save this page as a clean PDF.

Download the guide