Design Your Personal AI Operating System

Over the previous modules you have collected a lot: a mental model of how language models work, a Work Loop, prompt templates, verification routines, privacy rules, study methods, coding habits and a sense of when to automate. If you leave them as scattered advice, most of it will fade within weeks.

This article gathers them into one place, a personal AI operating system. The phrase sounds grand. The reality is modest: a few pages of decisions you have written down, plus a routine for keeping them current. Its value is that you decide in advance what you will do, so you do not have to decide while tired, hurried or tempted.

• What belongs in a personal AI system.
• How to write rules you will actually follow.
• How to choose tools by criteria, without depending on purchases.
• How to structure and maintain a playbook.
• A review routine, and a one-page manual template.

Your system covers the areas where you use AI, and the cross-cutting rules that keep it safe and honest.

AREAS OF USE  1. Learning      2. Research     3. Work         4. Coding  5. Writing       6. Career       7. AutomationCROSS-CUTTING LAYERS (apply to every area)  A. Principles    B. Privacy boundaries    C. Verification rules  D. Tool criteria E. Playbook              F. Review routine

You will not use AI in every area. Include the ones that matter to you and ignore the rest. The layers are what make the system a system.

Principles are a few sentences about what you are trying to do and what you will not do. They guide decisions that no rule anticipated. A good set is short. For example:

1. I use AI to think and work better, not to avoid thinking.
2. I am accountable for everything I send, submit or ship.
3. I am honest about what I did and what the tool did.
4. I protect other people's data as I would my own.
5. I match my checking to the stakes.

Write your own, in your own words. Principles that you can remember are the ones that will guide you.

Turn article 28 into a personal table of what goes where.

Add three standing habits: redact before pasting (placeholders, synthetic data), check the account (personal or work) before starting, and review connected apps every month.

Convert the stakes ladder from article 1 and the routine from article 26 into rules that fire on a trigger.

Rules phrased as if this, then that are easy to follow, and easy to notice when you break them.

Rather than listing products, which change, write down the criteria you use to choose. From article 5:

1. What does it see? (Only what I give it, or my files, mail and code?)
2. Where do facts come from? (Training, the web, my sources? Can I see them?)
3. What happens to my data? (Retention, improvement use, human review, deletion, business terms.)
4. What can it do on its own? (Talk only, or also send, change and delete?)
5. How do I check its work?
6. What does it cost, and what are the limits?
7. Can I leave? (Can I export my prompts and notes?)

Then pick tools by category, not by brand:

You can complete everything in this course without paying. Many products have free tiers, and limits and features change. Add a paid plan only when you can answer yes to all three: I have a specific, repeated task that the free option cannot do. I have measured the benefit, for example by timing the task with and without. The terms suit the data I will use it for. If a purchase is driven by fear of missing out, wait a month and revisit.

The playbook is where your reusable prompts and workflows live. The Work Loop's last step, Systemise, is about adding to it.

Name: [what job it does]When to use: [situation]Template: (the prompt, with named blanks like [AUDIENCE] and [PASTE TEXT])Inputs I must supply: [facts, source text]Output check: [what I verify every time]Risk level: [low / medium / high] and data level allowed: [public / personal / work / restricted]Tools it works in: [any / specific]Last tested: [date]   Known weak spots: [notes]

• Organise by job, for example Learn, Research, Write, Code, Career, Automate, with plain names.
• Keep it in plain text or Markdown so it works in any tool and can be exported. Do not trap your playbook in one product.
• Version small changes: a date and a line on what you changed.
• Add canary tasks: three to five tasks with answers you know, which you rerun after major tool changes to see whether behaviour shifted.
• Never store secrets or other people's data in it.
• Prune. If you have not used an entry in two months, archive it.

A system that is never reviewed drifts. A light schedule:

Here is a template. Fill it in, print it or save it where you will see it.

MY AI MANUAL        Updated: [date]PRINCIPLES (3 to 5)1. ...2. ...PRIVACYNever paste: [your list]Work content only in: [approved tool and account]Before pasting: redact, check the account.VERIFICATION RULES (if this, then that)- If numbers, dates, names or quotes: trace to a source.- If citations: open them.- If high stakes (health, law, money, safety): prepare questions only, verify with experts.- If code: read it, run it, test it, scan it.- If it affects someone else: a human decides.TOOLS (by role)General: [ ]   Search: [ ]   Documents: [ ]   Code: [ ]   Work: [ ]Criteria I use: sees what, facts from where, data terms, actions, checking, cost, exit.PLAYBOOKLocation: [file or folder]. Top 5 entries: [names].Canary tasks: [3 to 5, with known answers].LEARNING RULES (if I am studying)Attempt first for [15] minutes. Hints before answers. Cold test weekly. AI-off blocks: [2] a week.REVIEWWeekly [day], monthly [date], quarterly [month].WHEN THINGS GO WRONGStop. Revoke exposed secrets. Tell [person]. Write down the lesson.

A final-year computer science student's manual might say:

• Principles: "I use AI to learn faster, not to skip learning. I can explain everything I submit or commit. I tell people when AI helped."
• Privacy: "Never paste keys, passwords, other people's data or my college's exam material. Internship code only in the tool my company approved."
• Verification: "Every citation opened. Every number traced. All code run, tested and read."
• Tools: one general assistant on a free plan, a source-grounded notebook for lecture notes, the coding assistant in the editor with approved settings.
• Playbook: ten entries: concept explainer, hint ladder, flashcard maker, mock interviewer, debug hypotheses, code reviewer, resume bullet rewriter, research question framer, email drafter, weekly planner.
• Learning rules: "Attempt first for 20 minutes. Two AI-off study blocks a week."
• Review: "Sunday evening, 15 minutes."

That is one page, and it is enough.

The most common failure is building an elaborate system and abandoning it. Start with:

• Five rules that would have prevented your last AI mistake.
• Five playbook entries for tasks you actually repeat.
• One weekly review slot.

Grow it only when you feel a need. A system that you use at 70 percent of its ambition beats a perfect one that sits unread.

Products will be renamed, merged, repriced and redesigned. Your system should survive that:

• Your principles, privacy rules and verification rules are tool-independent. They stay.
• Your playbook is plain text and portable.
• Your tool list is by category. When a product changes, you swap one row.
• Your quarterly review catches term changes and new features, and tests whether anything worth adopting is real.

• Writing vague rules ("be careful") instead of triggers and actions.
• Building a system around one product.
• Buying tools before knowing what you need.
• Never reviewing.
• Storing prompts with secrets or private data in them.
• Making the manual so long that nobody reads it.
• Skipping the privacy table.

• Next: Workflow Recipe Cards for Study, Code, Career, Research and Work gives you ready-made entries to start your playbook.
• Then: Your 30-Day Plan and Long-Term AI Skill Roadmap turns the system into a schedule.
• For note-taking foundations, see Meetings, Notes and Long Documents: Summaries You Can Trust.