AIOS Implementation Playbook

Build Your AI Operating System

A structured implementation guide for building an AIOS that runs your business, frees your bandwidth, and scales without adding headcount. Follow the five parts in order. Each one builds on the last.

1 Mental Model
2 Get Set Up
3 Add Intelligence
4 Build Automation
5 Operate Right
8
Modules
5
Implementation phases
~7h
Total reading time
90 min
To first working setup
Part 01
The Mental Model

Before touching a terminal, understand the vocabulary and the framework. Every decision you make about what to build, in what order, at what level of automation — it all flows from this foundation.

📍 Outcome: You can explain Capabilities, Scripts, and Workflows — and where any task sits on the automation spectrum.
Part 02
Get Set Up

Two paths to the same destination — choose the one that fits your comfort level. Both result in a working workspace with a primed CLAUDE.md and your four context files populated. This is the foundation everything else builds on.

📍 Outcome: A working AIOS workspace with /prime loading your business context on every session.
Part 03
Add Intelligence

Context alone isn't enough. This phase connects your real business data and packages your repeatable processes into reusable skills — transforming the AIOS from a smart assistant into a genuine business intelligence layer.

📍 Outcome: Live business data flowing in daily. Key processes packaged as skills that run consistently every time.
Part 04
Build Automation

With context, data, and skills in place, this phase turns your task audit into a structured automation roadmap. Augmented slash commands take immediate work off your plate. Cron jobs and triggers make work happen while you sleep.

📍 Outcome: Daily brief running before you wake up. Top recurring tasks automated or supervised. Bandwidth reclaimed.
Part 05
Operate Right

Before client data enters the system — and before you scale to a team — these two modules lock the system down and make it efficient. Security without optimisation creates a protected but slow AIOS. Both are required for production use.

📍 Outcome: Six security layers in place. Sessions running 2.5× longer before hitting caps. System ready for client work.
Key Concepts Reference

The core vocabulary used throughout all five parts of this guide.

Capability
A skill with user_invocable: false. Always-on, auto-activates when relevant context is detected. Gives the AIOS access to an API or tool without being explicitly called. See → Module 01.
Automation Primitives
Script
Deterministic code (Python, bash) that does the same thing every time. The reliability backbone — handles the mechanical parts of any workflow. Used inside skills and commands for the parts that must never vary.
Automation Primitives
Workflow / Skill
A user-invocable skill. Combines capabilities, scripts, and natural language prompts into a structured process. Can be run in Augmented mode (you invoke it) or graduated to Supervised/Automated over time.
Automation Primitives
Context OS
The four context files that give every Claude session full business awareness: business-info.md, personal-info.md, strategy.md, current-data.md. Loaded by /prime at the start of every session.
Setup
/prime
The session initialization command. Reads the workspace structure and all context files, summarizes understanding, and confirms readiness. Run at the start of every session without exception.
Setup
Data OS
The layer that connects real business data to the AIOS. Local SQLite database, API connectors, daily snapshot scripts, data audit cron, and a mission control dashboard. Enables the daily brief.
Intelligence
Augmented
You invoke the workflow, AI does ~80% of the work, you steer key decisions. The first level of automation — always the starting point for any new workflow before escalating to Supervised or Automated.
Automation spectrum
Supervised
Runs on a cron or trigger without you initiating it, but you review the batch of outputs before anything ships. ~96% off your plate. The right destination for high-value tasks you wouldn't fully automate.
Automation spectrum
Snapshot methodology
Because most APIs return cumulative totals (not daily deltas), you take a daily snapshot and calculate the difference yourself. Essential for Data OS — start snapshotting on day one, the warm-up period is 7–30 days.
Data OS
Lethal trifecta
Simon Willison's framework: Tools (agent can act) + Untrusted input (reads external content) + Sensitive access (has credentials) = catastrophic prompt injection risk. Every AIOS deployment hits all three by design.
Security
RTK (--hook-only)
A PreToolUse hook that strips raw tool output before it enters context. ~89% of CLI noise tokens eliminated. Always install with --hook-only flag — without it, it adds tokens instead of removing them.
Token Optimisation
Shadow price
The API-equivalent cost of tokens consumed on a flat subscription plan. What tools like ccusage report — not your actual bill, but what those tokens would cost on pay-per-token API pricing. Used to measure optimisation progress.
Token Optimisation