AI Cofounder OS, a Notion operating system
AI agents don't fail because they're not smart enough. They fail because there's no system governing them. This is the governance layer I run my own business on, packaged as a Notion OS: context handling, approval gates, staged writes, weekly hygiene. Works with Claude, ChatGPT, or whatever you run.
Why systems beat prompts
Every protocol in this OS exists because one of these bit me while running my own company with an AI chief of staff. If you've run an agent for more than a week, you've met at least two.
Every new session starts blank. The AI doesn't know what it did yesterday, what's decided, or what it was told to leave alone. You burn the first ten minutes re-briefing, and things still slip.
Small unreviewed changes pile up. A field updated here, a document rewritten there. It looked right in the moment. Two weeks later you can't tell what's canon and what the AI invented.
You want the AI working without asking about every tiny thing. You also don't want it making irreversible calls alone. Without a gate system you either babysit constantly or accept the risk.
Stale instructions, dead context, and contradictions accumulate. The AI starts optimizing for rules that no longer apply. The system gets slower and more brittle, not faster.
What's inside
The protocols are plain text. Paste them into Claude, ChatGPT, Gemini, or whatever you run. The system is what matters, not the model.
A single page your AI reads at the start of every session and updates at the close. Persistent memory with dated entries: current focus, decisions, work in progress, holds. The AI picks up exactly where it left off.
The AI can read, analyze, draft, and recommend. It cannot write to canon until you review and approve. Every proposed change lands in Staging first. No silent no-ops, no phantom writes.
Explicit approval checkpoints. The AI knows what it can do alone (research, drafts), what needs a notification, what waits for sign-off, and what escalates cold: money, legal, anything irreversible.
A weekly hygiene pass, plus the recursive layer: every routine you run ends with a retrospective and files improvement proposals for its own instructions. You approve, the system compounds. Five minutes a week and it gets sharper every month instead of staler.
Comes pre-loaded with a fictional demo company so you see the whole system running on real-looking data before you touch anything. A one-page RESET guide clears it and makes it yours in under 10 minutes.
Pricing
$39
$99
$199
Not ready to buy? The fire-no-write axiom, the single most important protocol in the OS, is free.
Get the free guideQuestions
Yes. The protocols are plain text and work with any AI. The OS governs your workflow, not the model. The system is what matters, not which AI you run.
A working knowledge: able to navigate pages and databases. This is not a Notion tutorial. If you're new to Notion, spend two hours with their free guides first, then come back.
The RESET guide clears demo data in under 10 minutes. Full customization to your business takes one to three hours depending on how many protocols you adopt.
Built for solo operators. Works with small teams if one person owns the canon.
Email me after purchase and I'll set you up with an upgrade for the difference.
It compounds. The recursive layer has every routine end with a retrospective and file improvement proposals for its own instructions. You approve the good ones at the weekly Remora run. Routines get sharper every week instead of running on stale instructions.
Alex Margolick. I run Margolick AI, and this OS is the governance layer I use to run it. Not affiliated with Anthropic, OpenAI, Google, or Notion.