Field Note: AI-Ready Second Brain 7/8: Turning Knowledge into Work

I’ve spent 6 parts making my knowledge easier to organize, trace, review, label, and validate.

All useful. But stored knowledge still doesn’t perform a recurring job by itself.

Part 7 closes that gap a little. The piece I added is a small repository of SOP prompts: instructions I give an AI assistant for work I do over and over.

I want to be clear about what I mean by “SOP prompt,” because it isn’t a clever line of wording or a trick phrase. It is closer to an old-school standard operating procedure. It defines who is doing the work, what the job is, what inputs it needs, the rules it has to follow, what the output should look like, how to check the work, and where a human has to sign off before anything goes out the door.

The repository itself is just an index, not one giant do-everything prompt. It is a list of named jobs I can look up.

Two examples from mine: one turns raw research into a discovery checklist. Another coordinates narrow specialist work with verification and approval steps built in. The shift is from typing “help me with this again” every time to having a repeatable job with defined inputs, a consistent output shape, a way to check the result, and a human review step before anything consequential happens.

A few things I learned the hard way are worth saying plainly.

Writing the SOP prompt down makes the work more repeatable. It does not guarantee the result is correct. The assistant still needs the right current information to work from. Stale or missing inputs can produce stale or wrong outputs no matter how well the prompt is written.

Verification has to live inside the workflow itself, not happen as an afterthought.

And for anything public, client-facing, paid, sensitive, destructive, or touching real infrastructure, a human still reviews it before it goes anywhere. That part is not optional, and I’m not trying to engineer it away.

I also learned not to build one universal prompt for everything.

Different jobs need different context and approval rules. Cramming them into one prompt makes each workflow harder to understand and maintain. I would rather keep named, separate workflows in an index where I can find and update them. When a real run exposes a missing input, rule, or check, I update the SOP.

If you are looking for where to start, find the piece of work that keeps making you explain the same thing over and over.

Turn that explanation into one small, named SOP prompt:

role;
task;
inputs;
rules;
output format;
verification;
human review gate.

Keep the first version focused on one workflow. Treat it as something you maintain, not a spell you cast once and forget.

Part 8 is next: the Skill Command Registry, which acts as the control panel for finding and running the approved workflows I’ve been building.

What is the thing you keep re-explaining to your AI assistant that probably deserves its own SOP?

This is Chris from the Digital Field of Dreams, signing off.

0 comments

Field Note: AI-Ready Second Brain 7/8: Turning Knowledge into Work

Agent Empire

skool.com/agentempire

Learn how to build managed AI agents for your business or for clients using Orgo.

Leaderboard (30-day)