A practical guide to my AI Operating System

Working with AI Agents

A system for organizing AI-assisted work, from the first instruction to the final handoff.

v2.0 — September 15, 2026

In brief

The AI OS is a library of plain-text instructions that tells an AI agent how to work on your research.

01

Two guiding principles

Researcher control

The researcher retains authority over substantive decisions throughout the work.

Radical transparency

The AI-assisted process is documented so the user and/or others can inspect its inputs, intermediate work, decisions, and outputs.

For the fuller argument:With Great Powers ↗Read the chapter draft ↗

02

Best practices by design

03

The AI OS, at a glance

Across your work

The active task

Inside each project

Events + agent submissions

Coordination & bookkeeping

04

How a project is structured

Context

Brief, instructions, sources, data.

Plan

Tasks, dependencies, gates, deadlines.

Record

Progress, decisions, outputs.

Example: multi-source data collection
Starting point: the user's research questionIllustrative schedule · 2026
Task5–9 Oct12–16 Oct19–23 Oct26–30 Oct2–6 Nov9–13 Nov
Operationalize
Source A
Source B
Source C
Merge sources
Replication package
User decisions◆ Select◆ Approve merge◆ Accept
05

The lifecycle of a task

One session, one task

Example: update a literature table.

06

Make it your own

Start with the library

  • Choose one project and a recurring task.
  • Adapt the personas and skills to your work.
  • Add your identity, preferences, and project context.
Explore the public library

Add native bookkeeping

  • Install locally for a selected project.
  • Connect the Claude Code or Codex hooks.
  • Verify registration, a checkpoint, and closeout.
Open the installation guide

Install level by level

  • Start with one instructions file.
  • Add the library, a project, bookkeeping, and portfolio routines as you need them.
  • Your agent installs each level and stops for your decision.
Open the level-by-level guide