This AI System Will Make You So Smart It’s Almost Unfair

Author: Dan Martell (12:02)

Step-by-step build of his “AI brain” — a persistent, self-compounding knowledge base any AI can read.

1. Go Pro

Pay for an AI subscription. Free tiers = older models, slow. Paying alone puts you in the ~0.3% of users.

2. Install the Brain (pick the storage)

Conversations forget everything when closed; you need files in your structure that any AI can read. Three options:

Option	Verdict
Google Drive	Where it started; limited, not process-oriented
Notion	”AI-native data store” — files + context + structure
Obsidian (his pick)	Free, markdown = readable by every AI, graph view shows connections forming

3. Give It an Identity — 3 Files

• user.md — who YOU are: roles, communication style, frameworks you live by.
• soul.md — how the AI acts: voice, tone, values (“founder to founder, challenge me with love, no hedgy words — no try/hope/maybe; be resourceful before asking”).
• identity.md — who the AI is and does (his is “Kai”: part coach, part chief of staff, part accountability partner).

Pro tip: don’t write them yourself — have an AI interview you about how you work and draft the three agent files.

4. Wire the Brain — 7 (+1) Folders

Structure raised his answer accuracy from 60% → 85%:

1. people — one file per person, full context
2. projects — all projects, timestamped, with context
3. decisions — what was decided, how, alternatives
4. companies — research, competitors, index of companies you deal with
5. meetings — decisions and preferences come from here
6. daily — daily dump: 3–5 lines on what happened
7. knowledge — insights, frameworks, quotes worth reusing
8. (pro tip) MOC folder — Maps of Content: summary files linking bodies of work across folders (e.g. youtube.md linking script frameworks + hook libraries + thumbnail tasks). Build these lazily, when a topic gets messy.

5. Feed the Brain

Connect your systems; have AI extract into the folder structure daily. The brain only keeps what’s decision-relevant — filters noise like your own brain does.

His stack: Granola auto-transcribes meetings, then extracts via a saved template prompt:

Extract from this meeting: 1) Decisions — what did I decide, by whom, why? 2) Commitments — who promised what, by when? 3) Preferences — how people work/communicate. 4) Key insights — frameworks, strategic shifts, non-obvious observations. Output as markdown, skip small talk.

Drops output as <date> <meeting name>.md into meetings/.

6. Compound the Brain Overnight

The real brain rewires itself while you sleep; so should this:

• Manual: nightly prompt — read everything added today; find orphan notes (people/projects/companies mentioned without files); create those files; consolidate duplicates; update MOCs with new links; flag anything strategic for review.
• Automated (his setup): Claude scheduled task / cron runs that same prompt every night at 11pm.

More links = higher signal for context retrieval → better answers. End result: “send the invite to John” just works, because the brain knows which John, his email, his number.