Shared memory beats long prompts
Writing a decision down once, where every agent reads it, is more reliable than repeating it in every instruction.
In progress
My AI operating system
MoMo OS: a team of AI agents that runs my desk work.
I run my work with a team of AI agents on a shared operating system. Each agent has one job, they share one memory of decisions and policies, and nothing that commits a business goes out without my approval.
The problem
Leading two businesses and building new ones meant too much desk work for one person.
The approach
A team of specialised AI agents on one shared memory of decisions and policies. Every external commitment needs my approval.
How it works
A coordinating agent receives a request and routes it to a specialist: research, strategy, content, or operations. The specialist works from the shared memory (the decision log, policies and approved facts), prepares the output, and hands it back. If the output would commit a business to anything external, it waits in an approval queue for me.
The rules agents follow
- No invented facts: a claim without a source is marked as assumed, or answered “unknown”.
- No commitment without me: offers, discounts, contracts and anything new stay drafts until I approve them. A support agent may only send answers from a library I've approved.
- Decisions are written down once, in the decision log, and every agent reads from it.
What I learned
Lessons so far
“Unknown” is a good answer
Letting agents say they don't know turns hidden guesses into visible gaps.
Approval is one decision
Approving a prepared draft is one decision. The work is in the desk work before it.
Next
What happens next
Next: documenting the agent roles and approval rules in more detail, and writing up what works and what doesn't in Insights.
Let's talk
Let's map what your business can hand to agents.
30 minutes. Your workflows, what to automate first, and where a person must stay in charge.
Book a call