I run this OS on myself.
Before I install this at $65M companies, it runs on a sample size of one: me.
The same architecture I'd build for your business — training corpus, decision-making process, departmental dashboards, brain on top — operates daily across my own work, finances, learning, and accountability. I am the testbed. The mistakes happen on me first.
What it looks like
A functioning brain. Trained on:
- My values, beliefs, and personality traits — codified from years of post-it notes on the back of a door, refined into a corpus the system can reason against
- My financial transactions — every statement, fed through code, files into Notion. CFO function.
- My receipts — scraped from personal and business email, auto-matched to transactions in Google Drive
- My calendar — scraped monthly, tracked to 15-minute increments, fed into Notion as the time-tracking layer
- My weekly reflections — structured prompts, filtered through values + beliefs + core concepts, generating reflections that highlight my blind spots
- My news feed — three times a week, the brain pulls world / US / California / LA / AI news plus specific people I'm tracking, into a daily digest
- A bot trained on Jim's body of work — my Still Life co-founder. Edits my final reflection round through his lens.
Each of these is a "departmental head" in miniature: CFO, COO, board, advisor. Each feeds into one place.
The brain holds it all simultaneously.
Why this matters
Building the brain on myself surfaced the bottlenecks before any client paid for them.
Three lessons that made it into every install I do:
1. The robot can't work on what it can't see. Until everything lived in one Notion data lake, the brain was guessing. The migration discipline you see in the Manukora case study comes directly from this — I learned it building on myself first.
2. Hygiene is the moat, not the model. Logging transactions retroactively poisons the data. Logging at point-of-event compounds. The workflow-standardisation half of Phase 3 exists because I watched my own brain degrade when I let logging slip for two weeks.
3. The brain surfaces things I would have never asked. Cross-correlations between my time tracking and my expenses. Patterns between my reflections and my reading habits. The brain pulls signal from corners of my own life I'd never have queried directly. That's the lived experience that informs how I architect the cognitive layer in your business.
What's running today
- Daily digest — three times a week, populates my dashboard with curated news, a rotating quotation pulled from my own quote bank, and topic-specific updates
- Expense bot — statements in, categorised transactions out, filed in Notion
- Receipt-matcher bot — emails scraped, receipts paired with transactions
- Time-tracking bot — calendar scraped to 15-min increments, populated in Notion
- Weekly accountability bot — prompts → reflections → revisions → final report
- Accountability agent — monthly synthesis pulling daily accountability, weekly reflections, transactions, and time tracking into one report with the sharpest questions surfaced
- Jim bot — edits the final round through Jim's lens
What's next
The brain is currently dependent on Notion as the filing system and uses code for the agents. The next step — already in flight — is making it headless: pure dependency on my own LLM with a custom filing system, no interface required.
The reason I'm building it that way: it strips out every assumption about which tool the system needs. When I build for clients, I want zero attachment to any specific software vendor. If the right answer is Notion, fine. If the right answer changes in 18 months, the architecture moves with it.
That's the model-agnostic discipline. I'm not selling Notion. I'm not selling any vendor. I'm installing a system whose substrate can survive a tooling generation.