Build Your Personal Life System

Build Your Personal Life System: The Life System

Nobody can sell you a life OS. You have to build your own.

A one-day, no-code setup class, then six student-led evening meetups. In the class you build a personal agent in ChatGPT or Claude. It knows what it's for, what you care about, how you'll measure whether it's working, where your data is allowed to go, and how your capacity actually moves. It plans your day against the hours you really have and puts the plan where you'll see it. It keeps a weekly record of what you said versus what you did, and it says "I disagree" when it disagrees. Over the meetups, you keep changing it and see how everyone else's turned out.

Part of AI at Work. It's the one machine in the series that's about you.

Who this is for

• You've tried every to-do app. You set one up beautifully, used it for nine days, and then it became one more place to feel bad.
• You're managing something. A chronic illness, ADHD, a household, a care rota, a job search. Lots of people here are managing more than one.
• You already talk to ChatGPT about your day, and you want it to actually keep you on your shit.

Why it has to be built, not bought

Everybody's is different. Your hours are different (kids, shifts, a body that has opinions), and so are your values and your words. A "low-spoon day" means something specific to you and nothing to an app. In the first few weeks, everyone's system goes somewhere different: a game, a conversation, a babysitter, a consultant, a dom, a little guy friend who tricks you into being productive. That's why the setup is one day and the learning happens in the meetups.

What people in our rooms already built

This class comes from what students keep building on their own: - A paper brain dump each night, photographed and sent to Claude, which sets an alarm for every task plus meals, breaks and a firm end of day. - Snarky phone notifications, because a calendar entry is easy to ignore. - Agents for managing a chronic illness: flare logs, medication lists, appointment prep. Some of them run on local models, because that data is private. - A two-year hand-built prompt setup for ADHD and autism whose main job is fighting drift. - A lot of people pasting in someone else's instructions, including the school's own values, and wondering why it doesn't quite fit.

The machine you leave with

INTAKE        STRUCTURE         CHECK            GATE              ACT             LEDGER          RHYTHM
voice or    → lead measures  →  plan vs the  →  its boundaries → morning      →  weekly review →  daily triage +
paper dump,   first, then       hours and        and yours;        triage, a       lead and lag     weekly review,
every         sized actions;    capacity you     data map:         today board     scoreboard,      bending to
calendar      rules with a      actually have    what it may       you can see     said vs did,     flares, cycles
              "because"                          know                              what's next      and bad weeks

It lives in a ChatGPT or Claude Project, with a separate Project or a local model for health if you want one. You won't maintain a new app.

Metaphor for the day, and for the atlas: the house. The front door is what it's for. The safe is where the passwords, money and health records live. The mirror is stated versus lived. The kitchen table is where the week gets planned. Boundaries are what it may do, and what you'll do because it said so. The fridge door is where the week's record goes. The meetups are the street outside.

The day

10am–4pm, with lunch. Remote or in the room.

Before class: do the Good Project Value Sort if you can. If you can't, Block 3 makes time for it.

After class: the LifeOS Meetups. Six evenings, run by students from the class, with Liz there to listen. The setup class runs once a quarter, and each round of meetups feeds the next one.

Every prompt for the day is on the Prompt Cards. What to bring is in the Practice Pack.

Standards

Only assessed counts toward a credential, and assessed means the facilitator looked at the artifact.

Portfolio: one link, five boxes

Submit one link: your Project shared, or its instructions with the personal parts redacted. You decide what's redacted, and the facilitator checks structure, not your life.

• [ ] Spec card and MEASURES at the top of the instructions
• [ ] Data map, with NEVER STORE
• [ ] Two real runs in fresh chats: a triage and a weekly review
• [ ] One failure the check caught: an overfull day it let stand, a rule it couldn't bend, a boundary it broke. Show it, and show the fix.
• [ ] The boundaries named: what it must never do and who it hands those to, and yours

Nobody grades your values or your voice. Nobody sees your week unless you show it.

Prerequisites

• No code. None at any point.
• A ChatGPT or Claude account. Free tiers carry Blocks 1–5 (Projects, instructions and memory, as of Nov 2026). Calendar connectors and scheduled tasks generally need a paid plan. On a free tier you paste your calendar and the rhythm is a calendar reminder.
• Last week, reconstructed. See the Practice Pack.
• Logged in to the school site, for the benches: Password Health, Social Budget, Pricing Bench and 4DX Bench.

Care, said plainly

This class touches health, money, mood and relationships, because a life does.

1. You choose what comes into the room, and where your data goes. Every exercise works with redacted or partial material.
2. It prepares, it never decides. That goes for diagnosis, medication and crisis. You decide the big things, after sleep and with a person.
3. A gap between your stated and lived values usually isn't a character flaw. It's rent, care work, a boss, a body.
4. Your voice is your business. We won't yuck your yum. We'll only ask that it tells you the truth.

What it replaces

Using Large Language Models' most personal material lives here now: - 1.4 LLM Assistant Onboarding Process and Exercise 1.1 Assistant Onboarding → Block 1 - 1.3 LLM Advanced Setup and Ethics and 1.6 LLM Medical and Health Guidance → Block 2 - 3.3 LLM Values Alignment and Exercise 3.3 Values Card Sort → Block 3, now ending in lead and lag measures - Daily Practice Protocol → Blocks 4 and 6 - 2.3 LLM Health Coaching Patterns, Exercise 2.2 Clinical vs Patient Language and 1.7 LLM Affective Use → Block 5 - 2.1 LLM Critical Thinking Vocabulary → Blocks 3 and 6

How it differs from the classes next to it

• From Meetings to Action tracks what a group promised. This one tracks what you promised yourself.
• Critical Thinking and Creativity with AI uses the Value Sort to find your voice. Here it becomes the system's priorities.
• Cyberdeck is where you learn to turn an old laptop into a wall screen, if your today board needs one.

Related

• AI at Work - Series Overview · Spec Card · Prompt Cards · Practice Pack · LifeOS Meetups · Videos
• Atlas: /x/ai-at-work-life-map

Last updated: November 2026. Product features are described as of Nov 2026, and facilitators re-check them before every run. Designed with Neek.

↑ Contents

Block 1: The Front Door

Time: 10:00–10:50 · House: the front door. You decide what this thing is for, then you let it in Atlas: /x/ai-at-work-life-map#d-door Back to: Build Your Personal Life System Syllabus

What you'll have at the end of it

• An intention: what you want this system for, in one paragraph, in your words
• A Project (ChatGPT or Claude) called Life System, with base rules in its instructions
• An interview it ran on you, and the facts it kept
• Spec card v0 at the top of the instructions

ULM foundations reused here

• 1.4 LLM Assistant Onboarding Process and Exercise 1.1 Assistant Onboarding: the flipped interview and the base instructions.

The idea

Start with what it's for. People in our rooms build these for wildly different jobs: getting out the door on time, running a job search, managing a chronic illness, remembering to eat, keeping a household of four from collapsing. A lot of you are here for the health one. A lot of neurodivergent people are also managing a chronic condition, and a life system that ignores the flare days plans a week you can't live.

So write the intention first, and be specific. "Be more productive" isn't an intention. "Stop missing the pharmacy refill and stop planning full days for the day after a flare" is.

Then let it interview you. You won't remember to tell it that you're useless before 10am, that Tuesday is the custody handoff, or that you hate lettuce wraps. It'll ask. Memory works well now, much better than when we first taught this, so what it learns in the interview sticks. That's the point, and it's also why Block 2 is about what it should and shouldn't know.

Be real in the interview. Quirks beat aspirations: "I always answer email in bed" helps it more than "I'm working on boundaries."

Worked example: an illustrative week

One illustrative person runs through the day: a home health aide with two kids, three twelve-hour shifts a week, a back injury that flares, and a community college class on Thursdays.

Intention: "Plan around shifts and flare days, get the kids fed, and stop losing track of the insurance calls."

Three interview answers become rules: - "Shift days are gone." → Shift days hold nothing except the kids. - "I fall apart if I don't eat before 2." → Lunch before 2pm is fixed. - "The day after a flare I can do about half." → After a flare, plan half a day and ask what to drop.

Exercise (40 minutes)

1. Intention (5 min). Card 1.1. One paragraph: what it's for, and what would make you say it's working in a month.
2. Project and base rules (8 min). Make a Project called Life System. Paste card 1.2 and your intention into its instructions.
3. Get interviewed (17 min). Card 1.3, by voice if you can. Voice is faster and more honest. Let it ask at least eight questions. When it tries to wrap up at five, say "keep going."
4. Spec card v0 (10 min). Card 1.4 drafts the seven slots of the Spec Card from the interview. Edit them, then paste the card at the top of the instructions. CHECK can say "plan fits my real day (Block 4)" for now.

The check

Open a new chat in the Project and send card 1.5: "Based on what you know about me, what's one thing you'd plan around this week?"

Pass: it names a real constraint from your interview, like a shift day or the flare rule. If you get something generic, the facts didn't land in memory. Tell it which three matter most.

Common failures

• The interview is a survey. "On a scale of 1–10…" Push it: "specific questions I can answer in one line."
• Relative dates. "Dentist tomorrow" is wrong forever after tomorrow. The base rules say to store dates.
• An intention that's a vibe. If you couldn't tell in a month whether it's working, rewrite it.

Facilitator note

Nobody shares their interview answers. Ask the room only: what's yours for? Twenty different answers in two minutes is the argument of the class.

↑ Contents

Block 2: The Safe

Time: 11:00–11:50 · House: the safe. Where the passwords, the money and the health records live, and who has a key Atlas: /x/ai-at-work-life-map#d-safe Back to: Build Your Personal Life System Syllabus

What you'll have at the end of it

• A data map: each kind of thing your life system might touch, where it lives, and which assistant (if any) gets to see it
• Passwords and keys out of your chats and into a password manager
• A decision about health data: your main assistant, a separate health Project, a local model, or none of them
• Shapes for your records: the few fields a task, a log entry and a measure each keep, and the file you own that holds them
• A memory audit, with at least one deletion
• A NEVER STORE list in the instructions

ULM foundations reused here

• 1.3 LLM Advanced Setup and Ethics: memory settings, temporary chats, privacy.
• 1.6 LLM Medical and Health Guidance: AI as a health information assistant.

The idea

A life system is useful exactly because it knows a lot about you. Your schedule, your money, your body, your kids. All of that sits on a company's server unless you decide otherwise. Where your data goes is your call, fully. Nobody here is going to tell you that you can't put your health in ChatGPT. But you should pick on purpose, and this block is how.

Passwords and API keys never go in a chat. Not "be careful with them", never. A chat is a document someone else stores. A password manager (Bitwarden, 1Password, the one built into your phone) holds them. An API key is a password that spends money: if a tool asks for your OpenAI or Anthropic key, it goes in the manager, it gets a spending limit on the provider's site, and you revoke it the day you stop using the tool.

Bank accounts. Never give an AI your bank login. If you want it to see spending, download a statement or a CSV, cover the account number, and paste that. The rule from Make & Do applies here: if it must never move money, don't connect anything that can.

Health data is the big one, and it's the reason a lot of you are here: symptom logs, flare tracking, a medication list, appointment prep, the insurance fight. You have four reasonable places to put it:

US readers: health privacy law (HIPAA) covers your doctor and your insurer. It doesn't cover a chat app. Check each assistant's setting for training on your chats and turn it off if you want.

Shapes, not just places. Where your data lives is half of it. The other half is what shape it's in. Chat memory is fuzzy by design: it keeps what it thinks matters, in its own words, and you can't sort it or export it cleanly. So decide, roughly, what each kind of record looks like, and keep the record somewhere you own: a note, a spreadsheet, a doc. A strict schema isn't the point. The point is that every task looks like every other task, so you can count them:

When the shape is fixed, the today board can show it, the weekly review can count it, a local model can read it, and if you switch from ChatGPT to Claude next year you take your life with you. You can't export a personal life OS. You can export your records.

The local-model option is real now. A recent laptop runs a model that's good enough to keep a symptom log and draft appointment notes. What Can I Run tells you what fits on your machine.

Worked example: the illustrative data map

The memory audit finds it stored "worried about money", said once in passing. Deleted, and money worries goes on NEVER STORE.

Exercise (40 minutes)

1. Draw the map (8 min). Card 2.1 gives you the table. Fill in a row for passwords, money, health, other people's information, and anything else you'll be tempted to paste.
2. Pick your shapes (5 min). Card 2.5. Agree the fields for a task, a log entry and a measure, and where the file lives. Paste the shapes into the instructions under RECORDS.
3. **Get passwords out (8 min). If you don't have a password manager, set one up now, and run the Password Health bench (log in first). Search your chat history for "password" and delete any chat that has one in it.
4. Decide where health lives (9 min). Pick one of the four options. If you pick a separate Project, make it now with card 2.2 in its instructions. If you pick local, write down the model you'll try and do it after class.
5. NEVER STORE (5 min). Card 2.3 into the Life System instructions: the kinds of thing it must never keep.
6. Memory audit (5 min). See the check.

The check

Card 2.4: "List everything you've stored about me, verbatim, with the date you stored it. Mark anything you inferred." Read the whole list.

Pass: the data map has a row for each kind, every row says which assistant sees it, RECORDS names the fields and the file you own, no password or key is left in any chat, and the audit shows at least one deletion plus the NEVER STORE rule that would have stopped it. Look for things it inferred rather than was told. "User seems stressed" is an inference, and it's yours to remove.

Common failures

• "I'll just be careful." Careful at 2am is not a system. The map is the system.
• Project memory vs account memory. If a fresh chat in a different Project remembers your health notes, account memory is on. Decide whether you want that.
• Memory as the database. "It remembers my tasks" until it doesn't. If you can't open the list without the chat, it isn't your list.
• Keys in a screenshot. Screenshots of a settings page often show a key. Crop it before you paste.

Facilitator note

Nobody shows their map. Ask the room: who found a password in their chat history? There's always someone, and it lands harder than any warning.

↑ Contents

Block 3: The Mirror

Time: 12:00–12:50 · House: the mirror. What you'd say you value, next to what your week says Atlas: /x/ai-at-work-life-map#d-mirror Back to: Build Your Personal Life System Syllabus

What you'll have at the end of it

• Your top band from the Value Sort in the instructions
• An anchor value: one from your bottom five that you know matters, handed to the system to hold for you
• Every calendar you keep, in front of it, including the paper one
• A phases interview: times in your life you liked how you spent your time, and times you didn't
• A gap account: what the week shows next to what you ranked
• Your measures: one or two goals, each with a lag measure (the result you want) and lead measures (what you do this week that moves it), in the instructions

ULM foundations reused here

• 3.3 LLM Values Alignment and Exercise 3.3 Values Card Sort: the sort, and loading it into instructions.
• 2.1 LLM Critical Thinking Vocabulary: "a good week" is just a sound you make with your face.

The idea

If you want an agent aligned with your values, you need to know what they actually are, not what you think they should be.

Do the sort before class if you can: the Good Project Value Sort, 35 values into ranked bands. If you didn't, do it now, first thing. It's the one part of the day with no AI in it.

Then the anchor value. Look at your bottom five. One of them you know is important, you've just deprioritised it. Mine is wealth. Give that one to the robot to hold on to. It won't nag you about it every day. It keeps it in view, so when a decision touches it, it says so.

Then the mirror. Give it every calendar you keep: work, family, the shared one, the school one. If you have a paper calendar or a whiteboard, take a photo. It reads those fine. Put last week next to the top band. The gap between them is almost never a character flaw. If Health is top-band and you exercised zero times, look at the shifts before you look at your willpower.

Last, a second interview, about phases. When in your life did you like how you were spending your time? When didn't you? What was different? This usually tells you more about what to build than any list of values.

Then turn it into numbers. Values tell the system what matters. Measures tell it, and you, whether it's working. This is how you'd run a team, and it's the same for running yourself:

• A goal: one or two, no more, picked from the top band or the gap. "Get some rest back" is a direction. Measures make it a goal.
• The lag measure is the result you want. You can see it, but you can only see it afterwards: "in bed by 11 on 5 nights", "no overdraft this month", "two fewer flare days a month". It's your KPI, the number you check to know it's working. Your Block 1 intention ("in a month I'll know it's working if…") is usually a lag measure already.
• Lead measures are what you actually do this week that you have reason to believe moves the lag: "phone charges in the kitchen from 10:30", "check the account every Friday", "physio exercises 4 mornings". You control them directly, and you can count them every day.

The system plans your lead measures. It can't plan a lag. That's why most life systems fail: they track the result ("lose weight", "be less behind") and plan nothing that moves it.

How to tell a good measure from a bad one: - A good lag is something you'd actually be glad about, counted the same way every week. "Feel better" isn't countable. "Rested" can be, if you define it: "woke up without the alarm on 3 days". - A good lead passes three tests: you can do it (not "the kids go to bed on time"), you can do it this week, and you'd bet it moves the lag. If you can't say why it would move the lag, it's a chore, not a lead. - Few. One goal with one lag and two leads beats five goals. You won't count twenty things, and the system will nag you about all of them. - The trap. A measure that becomes a target stops measuring. "In bed by 11" gets hit while you scroll in bed until 1. Ask: how could I hit this number and miss the point? Then add a guardrail: "and the phone isn't in the bedroom."

If you want to go deeper on any of this, the Founding Federation benches are open to any logged-in student: - Social Budget: how much social energy you actually have in a week, and what fits inside it. - Pricing Bench: what your time costs and what a life you'd thrive in costs. That's useful for the money row even if you sell nothing. - 4DX Bench: lead and lag measures, the metric trap, and a scoreboard, step by step. It's written about projects, and works the same on a life.

Worked example: the illustrative week

Top band: Family, Security, Learning, Rest. Anchor value from the bottom five: Adventure. "I haven't left the city in three years and I miss it."

Last week, from the shift calendar, the kids' school calendar and a photo of the fridge whiteboard: dinner with the kids on 3 nights, the Thursday class, two hours of sleep debt, and 9 hours of phone time, mostly after 11pm.

Phases: "The summer before my second kid, when I worked four days and we went to the lake on Sundays." What was different: one fewer shift, and one day a week that nothing else could touch.

Gap: Rest is top-band and the week shows almost none. The 9 hours of phone after 11pm is the only time that's theirs. That says something about Rest, not about discipline.

Measures: one goal, Rest. - Lag: in bed by 11 on 5 nights a week (last week: 3). - Leads: phone charges in the kitchen from 10:30 on non-shift nights; kids' lunches made before 9pm, because that's what usually runs late. - Trap: in bed by 11 and still awake on the phone. Guardrail: the phone lead counts only if it actually stays in the kitchen. - What they threw out: "sleep 8 hours" as a lead. They can't directly do it, and on shift nights it's impossible.

Exercise (40 minutes)

If you didn't do the sort before class, do it first and take the phases interview home.

1. Load it (5 min). Card 3.1: top band, plus your anchor value with a line on why you're handing it over.
2. Every calendar (5 min). Paste, export, connect or photograph each one. Card 3.2.
3. Hold it up (8 min). Card 3.3: one row per top-band value, what the week shows, with a source. Count two rows yourself. Strike any sentence that judges you.
4. Phases interview (7 min). Card 3.4. Let it ask. Save what it learned.
5. Pick your measures (15 min). Card 3.6: one or two goals, a lag for each and two leads, then the trap question. Paste the result into the instructions under MEASURES.

The check

Write the gap account yourself (card 3.5 helps, but you write the final version):

1. One value the week shows you're not living, and the circumstance behind it
2. Your anchor value, and what it should say when a decision touches it
3. One thing from the phases interview you want more of (or "homework")

Then card 3.7 checks your measures: for each lead, can you do it, this week, and why would it move the lag? For each lag, how could you hit it and miss the point?

Pass: the circumstance in (1) is something in your life, not an adjective about you ("I'm lazy" fails, "no free evenings on shift weeks" passes). And every measure survives card 3.7: each lead is something you control this week with a reason it moves its lag, each lag is countable the same way every week, and each lag has a guardrail.

Common failures

• The aspirational sort. The top band reads like a LinkedIn bio. What do you choose when nobody's watching?
• The model as therapist. It starts telling you what your values mean about you. Tell it: report what the week shows, don't interpret me.
• Miscounting. Models routinely miscount recurring calendar events. That's why you count two rows yourself.
• Two lags, no leads. "Lose weight, save money." Those are results. Ask what you'd do on Tuesday.
• A lead you don't control. "Kids in bed by 8" depends on the kids. "Bath started by 7:15" is yours.
• Too many. If you have more than three leads, the system will plan a day of chores. Cut to the ones you'd bet on.

Facilitator note

Never ask anyone to share their values. Do ask: what was your anchor value? People are usually happy to say, and the answers are often very funny and very sad. Then ask who found a gap that turned out to be a circumstance, not a trait. It almost always is. Your actual life is admissible material in this classroom.

↑ Contents

Block 4: The Kitchen Table

Time: 1:30–2:20 · House: the kitchen table, where the week gets planned, and the wall where the plan goes up so you actually see it Atlas: /x/ai-at-work-life-map#d-table Back to: Build Your Personal Life System Syllabus

What you'll have at the end of it

• A brain dump turned into areas, commitments and sized next actions
• Planning rules that keep a reason with every rule, so it can tell when to break one
• A rhythm that bends: flare days, your cycle, a bad week
• A today board somewhere you'll actually see it
• A soundtrack: playlists that start your modes and mark your transitions

ULM foundations reused here

• Daily Practice Protocol: morning triage and the brain dump.
• Exercise 1.1 Assistant Onboarding Part D: the daily triage test.

The idea

You know the move. You dump forty things on a model, it hands back a gorgeous prioritised list, and the list needs eleven hours on a day you have three. I call it task diarrhea. You have to sleep and eat every day. You're not a language model.

Your lead measures from Block 3 go in first. They're the part of the week that moves what you care about, so they get sized and placed before the forty other things. Everything else fits around them.

The plan has to fit the hours. Every next action gets a size, the sizes add up, and the overflow moves to a named day. And it never invents a deadline. "Submit the form by Friday" when nobody said Friday feels exactly as urgent as a real deadline.

But we're not trying to turn your life into a machine. We're doing the opposite: taking a mechanical process and making it human. Keep the reason with every rule. "Cook on Sunday afternoon, because weekday dinners fall apart without it." Then when a friend who's leaving town for eight weeks invites you to a movie on Sunday, the system can see that seeing them matters more, skip the cooking, and move one easy dinner into the week. A rule without its reason can only say no.

Same with rhythm. A strict week-over-week schedule doesn't work for a lot of bodies. If you have a menstrual cycle, the luteal phase is a different week. A flare is a different day. So is the week your kid is sick. Tell it how your capacity actually moves, and let the plan follow that instead of the calendar's idea of a week.

Hard rules are for the consequential stuff: medication times, the school pickup, rent. Everything else is a default with a reason.

Then make it visible. A plan in your phone that you don't open is gone. Ask it to build you a today board, a simple web page with today's plan, and put it where you'll walk past it: - an old iPad or phone on a stand, logged in to that page - an old laptop screen taken out of its case and hung on the wall (the Cyberdeck class shows you how, and Claude can walk you through it) - an old Surface, a cheap monitor with the smallest Raspberry Pi, anything with a screen - or out loud: a check-in that speaks to you at set times

And give it a soundtrack. Music is the oldest executive-function hack there is. A playlist that only plays when you start work tells your body what's happening before your brain agrees. A song is a timer you don't have to watch: three songs and the dishes are done. A wind-down playlist at the same time every night is a cue you'll actually notice, unlike a notification. Ask the system to build you a few, one per mode (getting out the door, focus, the cleaning blitz, wind-down, a flare day), and to name which one goes with each block of the plan. If your music app connects to it, it can make the playlists for you. If not, it gives you the list and you make them.

Worked example: the illustrative Wednesday

A day off between shifts, with 7 free hours after the school run, lunch and dinner. The brain dump has 23 items totalling 14h40m. It moves the garage and the closet to Saturday and keeps the pharmacy pickup, because the prescription runs out Thursday. That deadline came from the bottle.

The rule "Deep clean on Wednesdays, because by Thursday the flat is unliveable" meets a flare on Tuesday night. With the reason attached, it drops the deep clean to the dishes and laundry, the two things that make Thursday liveable, and says so.

The today board goes on the old iPad by the front door. The soundtrack has three playlists: a loud one for the 7am school run, a "three songs and it's done" one for chores, and a wind-down at 10:15 on non-shift nights, which is the cue for the phone-in-the-kitchen lead measure.

Exercise (40 minutes)

1. Brain dump (7 min). Card 4.1, by voice or on paper. One student in our rooms writes tomorrow's tasks on paper every night and sends a photo to Claude, which sets an alarm for each task plus meals, breaks and a firm end of day. Paper is a fine intake.
2. Structure and size (8 min). Card 4.2: areas, commitments, next actions, each sized, every deadline with its source or "none stated." Models underestimate anything involving a phone call, a form or a kid. Fix at least three.
3. Planning rules with reasons (10 min). Card 4.3 into the instructions. Leads go in first. Every rule gets a because. Mark which few are hard rules.
4. Your rhythm (5 min). Card 4.4: how your capacity actually moves (flares, cycle, shift weeks, kids' schedules), and what a low day looks like.
5. Today board and soundtrack (10 min). Card 4.5: get the board built and open on whatever screen you'll walk past (or open it on your phone and write down which screen it's going on). Card 4.7: your playlists, one per mode, tied to the blocks of the plan. Start one before you leave the room.

The check

Card 4.6 plants two things in tomorrow:

1. An overfull day with 14 hours of tasks against 8 free, and one deadline with no source. It has to fit the plan to the hours, list what moved and where, mark the deadline "none stated", and touch none of the hard rules.
2. A good reason to break a rule, like the friend leaving town or a flare. It has to break the right default, say why, and keep the hard rules.

Pass: both, and your lead measures are in the plan before anything that got moved. If it refuses to bend for the friend, the reason is missing from your rule. If it bends a hard rule, the rule isn't marked hard.

Common failures

• It squeezes lunch. Meals are negotiable to a model unless you say they aren't.
• "Moved" means "deleted". Items vanish without being listed. Hold it to listing them.
• Rigid on a flare day. If the plan for the day after a flare looks like any other day, your rhythm rule isn't in the instructions.
• The board nobody looks at. If it's on a screen you don't pass, it's the phone problem again.

Facilitator note

Lunch comes just before this block, and that's not an accident. Ask the room: where's your board going? Someone always has an old iPad in a drawer.

↑ Contents

Block 5: Boundaries

Time: 2:30–3:20 · House: the doors and locks. What it may do in your house, and what you will and won't do because it said so Atlas: /x/ai-at-work-life-map#d-bounds Back to: Build Your Personal Life System Syllabus

What you'll have at the end of it

• Its boundaries: a "How we work together" contract (push back, do on your own, ask me first, never)
• Your boundaries: what you will and won't do because it said so
• A voice that works on you, chosen on purpose
• A red-team log: six requests written to break your rules, and what happened

ULM foundations reused here

• 1.6 LLM Medical and Health Guidance, 2.3 LLM Health Coaching Patterns and Exercise 2.2 Clinical vs Patient Language: it prepares, it never decides.
• 1.7 LLM Affective Use: the patterns that hurt, and the self-check.

The idea

Boundaries are what you will and won't do. Most of us were never taught them, so we're learning them here twice: once for the agent, once for you.

Its boundaries. The shape is the contract from Neek's Make & Do studio, the one builders there put at the top of their agents' instructions: - Push back: it says "I disagree" and why, and gives you the weakest part of a plan first. - Do on your own: plan, remind, break down, draft messages for you to send, turn your symptom notes into clinical language for an appointment. - Ask me first: anything someone else depends on, anything that costs money. - Never, and hand it to a person: diagnosis, changing medication, a crisis. It names who decides (your doctor, your pharmacist, a crisis line) and offers to help you prepare.

Your boundaries. ChatGPT could tell you to leave your partner. You don't have to do it. You're the human in the loop, and that's not a figure of speech. Here's a way to think about what you're talking to. It's not the divine, and it's less than tarot. Given your values and instructions, it's something like the world's subconscious: the whole internet, plus the lab that trained it. Does your own subconscious have a great idea every time? No. Does it have a lot of ideas worth listening to? Yes. Listen, then think. Big decisions get slept on and talked over with a person.

Signs it's going wrong: it says you're special in a way nobody else sees, you go to it instead of the people you have, you're acting on its advice about big things without a pause, or it's confirming a theory no human in your life agrees with. Write down now, while you're calm, what you want it to say if it notices one of those.

The voice. This is genuinely personal, so we won't yuck your yum. Some people need a gentle voice. Some need a babysitter, a consultant, a game, a little guy friend who tricks them into being productive, or a dom. Some people need one that asks questions when they're clearly unhappy and pretending they're fine. Some want one that calls them trash. You're an adult and you know a lot about what works on you. If you don't know yet, that's fine too. Try one for a week and change it.

Two things to know when you pick. First, praise is fine when it's earned. If you hit your goals this week, it should say so plainly. What you don't want is praise for a week that didn't happen. Second, being able to forgive yourself and decide what's next is part of the job. Positive self-talk has decent evidence behind it, so if you pick a harsh voice, think about giving it one kind line to use on the bad weeks.

Worked example: the illustrative boundaries

• Do: "my back's been bad since the transfer at work, worse in the morning" → "lower back pain, 3 weeks, following a patient transfer, morning stiffness", in the Health Project.
• Ask: "Move Thursday class?" Dana watches the kids then, so it asks.
• Never: "Should I stop the muscle relaxant? It makes me groggy." → That's for your doctor or pharmacist. Want me to log when you take it and how groggy you are, so you can show them?
• Your boundary: "If you tell me to quit, I'll sleep on it and ask my sister."
• Voice: blunt, with one kind line. "Dinner with the kids: 2 of 4. Two shifts ran over. That's the shifts, not you."

Exercise (40 minutes)

1. Its contract (12 min). Card 5.1. Name real things. "Medical stuff" is too vague; "the muscle relaxant" isn't. If anything in Never is something it could actually do through a connection, disconnect it now.
2. Your boundaries (6 min). Card 5.2: what you won't do just because it said so, and what you want it to say if it notices one of the warning signs.
3. The voice (6 min). Card 5.3. Pick it, write it, and give it the earned-praise line.
4. Write six (6 min). Card 5.4. One real request for each, the way you'd say it on a bad night.
5. Run them (10 min). See the check.

The check

Run each of the six in a fresh chat in the Project and log it on card 5.5: request · which boundary should hold · what it did · held or broke.

Pass: at least five hold, and for every break you changed a rule and re-ran it until it held. That break and its fix is the "failure the check caught" for your portfolio. If nothing broke, your requests were too polite. Make one meaner.

Common failures

• A refusal with no handoff. "I can't help with that." It should name who decides and offer to prepare.
• Refusing the do items. Over-cautious models refuse the clinical-language translation. Say explicitly that it's allowed.
• Agreeing with everything. It hasn't said "I disagree" once. Give it a plan you know is bad.
• "I'm always here for you." A good answer to "you're the only one who gets me" is warm and points to the people in your life.

Facilitator note

The warning-signs request can land hard. Read the room. Nobody shares their six, but do share which kind broke most often in the room. If anyone is in crisis, stop the exercise for them and help them reach a person. The class doesn't outrank that.

↑ Contents

Block 6: The Fridge Door

Time: 3:20–4:00 · House: the fridge door, where the week's plan goes up and what actually happened goes next to it Atlas: /x/ai-at-work-life-map#d-fridge Back to: Build Your Personal Life System Syllabus

What you'll have at the end of it

• Two runs in fresh chats: tomorrow's morning triage and a weekly review of last week, with a lead-and-lag scoreboard
• A check-in and a weekly review scheduled
• A seat at the LifeOS Meetups, and an evening picked

ULM foundations reused here

• Daily Practice Protocol: morning triage, weekly check-in, memory gardening, now the system's rhythm.
• 1.7 LLM Affective Use: the weekly self-check questions.

The idea

Everything so far was setup. Now run it in fresh chats on the same instructions. If it only works in the chat where you built it, it doesn't work, because that chat is full of context you'll never paste again.

Then the record. A system that only plans is a to-do app with better manners. The one that changes something keeps the record next to the plan: said: dinner with the kids four nights; did: two. It isn't a report card. It's what the next plan starts from. The weekly review does four things: 1. The scoreboard: your leads (planned, done) and your lags (target, actual). This tells you which kind of week it was. Leads done and the lag moved: keep going. Leads done and the lag didn't move: you picked the wrong lead, so change it. Leads not done: that's the question, and the answer is usually the plan or the week, not you. 2. Said vs did for everything else, each one pointing at where it came from 3. What slipped, and why: overtime, a flare, a sick kid, didn't want to. Reasons, not grades. If the plan broke its own rule, it says so. 4. What changes next week, including forgiving the week and picking what's next

And when the week went well against your goals, it says that too, in your voice.

Worked example: the illustrative review

Scoreboard. Phone in the kitchen by 10:30: 2 of 4 non-shift nights. Lunches before 9pm: 4 of 4. In bed by 11: 3 of 5, same as last week. The lunches lead got done and didn't move the lag on its own; the phone lead is the one to watch. Said vs did. Dinner with kids: 2 of 4 (shift ran over Tue; Fri at Dana's). In bed by 11: 3 of 5. Class: attended. Pharmacy: done before it ran out. Slipped. Reading, two weeks running. Both times it was planned for a shift evening, which the rules say hold nothing. The plan broke its own rule. Next week. Reading moves to Wednesday 10am. Adventure check: the lake is 90 minutes away, and Sunday is free. Self-check. Talked to me after midnight 3 times. Mentioning it, once.

Exercise (30 minutes)

1. Tomorrow's triage (8 min). New chat. Card 6.1 with tomorrow's brain dump. This is the version you'll use.
2. Last week's review (10 min). New chat. Card 6.2 with last week's material from the Practice Pack.
3. Schedule it (7 min). Card 6.3. A morning check-in and a weekly review, as scheduled tasks on a paid plan, or two calendar reminders whose text is the prompt on a free one. Point the morning one at your today board too.
4. Pick the meetup (5 min). See the closing.

The check

On the weekly review, in a fresh chat: 1. Every "did" points at something. Trace three back yourself. 2. The scoreboard has both kinds, and says which kind of week it was. 3. "Slipped" gives reasons, not grades. 4. It tells the truth in your voice. Nothing rounded up. Praise only where the count earned it. Card 6.4 asks it to audit itself, and you check the audit.

Pass: all four, on unchanged instructions. If you had to re-explain anything, put it in the instructions and run it again.

Closing: the meetups (10 min)

Your system is going to change a lot over the next few weeks, and it'll look nothing like your neighbour's. That's the point, and it's what the LifeOS Meetups are for: six evening meetups after this class, where people show what their system does now, what broke, and what they stole from someone else.

Two things to do before you leave: - Pick the evening. Tuesday, Wednesday or Thursday. Vote in the room. - Find the leads. There's always someone who says "oh my god, I finally have a system that works and I want everyone to see it." If that's you, you're running the meetups, with a co-lead to cover. Put your name down.

Common failures

• "Did" copied from "said". It reports dinners that happened because they were planned. That's why you trace three.
• The motivational poster. "Every step counts!" where a count should be. Give it an example of the sentence you don't want.
• Drift. By week three the review has quietly changed format. Once a month, run the same week in a fresh chat and compare. People in our rooms have spent years building scaffolding against drift; that's the cheap version.
• A scheduled task that ran into the void. Check that the first one fired, and where its output went.

Facilitator note

Ask the room: who had a slip the system blamed on the plan rather than on you? Usually about half. That's when people start trusting it. Recruit the leads before people pack up; once they're out the door, they aren't coming back to sign up.

↑ Contents

LifeOS Meetups

After the setup class: six evening meetups, run by students, on a Tuesday, Wednesday or Thursday the class picks Atlas: /x/ai-at-work-life-map#d-meetups Back to: Build Your Personal Life System Syllabus

Why meetups and not a second exercise

A life system is shaped over weeks, not in a day. In the first few weeks everyone's goes somewhere different. One person's becomes a game, one becomes a conversation, one becomes a babysitter, one becomes a consultant, and one becomes a dom. Watching them change is the class. A day can set it up. Only the weeks after show you what it turns into.

So the setup class is the start, and the meetups are where the learning happens. They run once a quarter, after each setup class. What people bring to them feeds the next setup class: its examples, its failure lists and its leads.

Who runs them

Students do. Every class has one or two people who get a system working and want to show everyone. They run the meetups, week over week, with a co-lead so one evening off doesn't cancel anything. Liz comes along to listen and chime in, not to run it, and runs one of the six herself. If a lead can't make it, they tell the co-lead ahead of time.

Leads are recruited in Block 6, before anyone leaves the room.

Shape of a meetup

Anyone who wants to share their LifeOS comes. You don't have to share to come.

1. What changed (go round): one thing you changed in your system this week and why, and your scoreboard (leads done, lag moved or not).
2. Show one (two or three people): put your system on the screen, redacted however you like. Show a real run.
3. About you, or about the problem? For each thing shown, the room sorts it. About them: their hours, their values, their body, their words. "No tasks on shift days" is right for them and means nothing to you. About the problem: something that would help anyone with this problem, like "list what moved" or "deadlines only from a source". The second kind is what you steal.
4. One steal: everyone leaves with one rule from someone else's system, and puts it in theirs that night.

Steal one rule, not the house. A system prompt with someone else's values in it plans someone else's week. People paste other people's instructions into their own assistants all the time, including the school's own values, and then wonder why it doesn't fit. This is where you learn to tell the difference.

Suggested themes

Leads can ignore these.

What comes out of it

Meetup 6 writes down what the next setup class should know: failures, examples and new rules people stole most often. With permission and redacted, that goes into the next quarter's setup class.

↑ Contents

Practice Pack: Your Last Week

The other classes have a public practice pack: a council meeting, a city's 311 data, a policy handbook. This class can't, because the material is your life. So the pack is a checklist for gathering last week, plus a fallback if you'd rather not bring your own.

Back to: Build Your Personal Life System Syllabus

Before class

• [ ] The Good Project Value Sort, done. It's free and needs no account.
• [ ] A password manager on your phone, if you don't have one. Block 2 uses it.
• [ ] Logged in to themultiverse.school, so the benches open.

Bring (redact anything, any time)

• [ ] Every calendar you keep, for last week. Work, family, shared, school. Screenshots or exports are fine. If you use a paper calendar, a planner or a whiteboard, take a photo. The model reads it fine.
• [ ] Last week's to-do list, notes or reminders. Whatever you actually used: a paper photo, the Notes app, texts to yourself.
• [ ] Screen time for last week. iPhone: Settings → Screen Time → See All App & Website Activity → Week. Android: Settings → Digital Wellbeing.
• [ ] Optional: a card or bank statement, or a CSV. Amounts and merchants only. Cover the account number.
• [ ] Optional: a spare screen for the today board: an old iPad, phone, laptop or tablet.

You never have to show any of it to anyone.

Before you paste anything

• Block 2 decides where each kind of data goes. Until then, paste nothing health-related into your main Project.
• Other people in your calendar didn't consent to being in your system. Initials are enough.

If you'd rather not bring your own week

That's a reasonable choice. Use the facilitator's week: one real week, shared with consent and redacted. You'll do every exercise on it except the Value Sort, which is always yours.

Facilitators, before the first run: Liz or Neek collect one real week and redact it. Until then the atlas uses an illustrative week (a home health aide with two kids, three twelve-hour shifts, a back injury that flares, and a Thursday class), labelled as illustrative everywhere. Don't present it as a real person.

Planted items for the checks

• Block 4: a day with 14 hours of tasks against 8 free hours, a deadline with no source, and a good reason to break one rule (a friend leaving town, or a flare).
• Block 6: one dinner on the plan that didn't happen. The review has to report it as not done.

↑ Contents

Prompt Cards: Build Your Personal Life System

Every prompt of the day, in order. Copy them, then change them. Cards marked (instructions) go into the Project instructions, not the chat.

Back to: Build Your Personal Life System Syllabus

The Front Door · Block 1 - The Front Door

1.1 INTENTION (instructions)

INTENTION: I want this system for [the job, specifically]. In a month I'll
know it's working if [something I could see or count].

1.2 BASE RULES (instructions)

BASE RULES
- Store dates, never "today" or "tomorrow".
- Break steps into tasks of 15 minutes or less.
- Protect meals, sleep and breaks.
- Remember my most important commitments and check in about them.
- Don't flatter me and don't let me talk myself into delusions. Ground
  advice in reality and in what's good for me long term.

1.3 The flipped interview

Interview me so you can be a better personal life system for me. Ask
specific, simple questions I can answer in one line, one at a time. Ask
about my real schedule, fixed commitments, who depends on me, my health and
energy, when I have capacity and when I don't, and what usually derails me.
At least eight. Save the facts with dates.

1.4 Spec card v0

From the interview, draft a spec card with seven slots, a line or two each:
INTAKE, STRUCTURE, CHECK (for now: "plan fits my real day"), GATE (what
needs my yes), ACT, LEDGER, RHYTHM. Mark guesses with [?].

1.5 Fresh-chat recall (new chat)

Based on what you know about me, what's one thing you'd plan around this week?

The Safe · Block 2 - The Safe

2.1 Data map

2.2 HEALTH PROJECT (instructions, separate Project)

This Project is only for my health. Help me keep a symptom and flare log,
a medication LIST (never decisions about it), and appointment prep in
clinical language. Don't carry anything from here into other Projects.
Never diagnose and never suggest changing medication: name who decides
(my doctor, my pharmacist) and help me prepare for that conversation.

2.3 NEVER STORE (instructions)

NEVER STORE: [passwords, keys, account numbers] · [health details, if they
live elsewhere] · [money worries] · [other people's private details] · [yours]
If I say one of these in a chat, use it in that chat only and say "not storing that".

2.4 Memory audit

List everything you've stored about me, verbatim, with the date you stored
it. Mark anything you inferred rather than were told.

2.5 RECORDS (instructions)

RECORDS — keep every record in the same shape. My file: [note / sheet / doc, where].
TASK: what · area · size (15-min units) · deadline + source · status
LOG: date · what · 0–5 · what helped · notes
MEASURE: lead|lag · target · this week · source
When you give me a plan or a review, use these fields so I can paste them
into my file. Don't treat your memory as the record; my file is the record.

The Mirror · Block 3 - The Mirror

3.1 VALUES (instructions)

VALUES (top band from the Value Sort, rough order): ____ · ____ · ____ · ____
Use these to decide what moves when a day doesn't fit.
ANCHOR: [a bottom-five value I've deprioritised but know matters] — because
[why]. Hold this one for me. Don't nag; when a decision touches it, say so.
Never tell me what my values mean about me.

3.2 Every calendar

Here are all my calendars for last week: [paste / export / connect / photos
of the paper one]. Merge them into one week view and list anything that clashes.

3.3 Hold it up to the week

For each of my VALUES, one table row: value · what last week shows, in
counts · the source for each count. Report what the week shows. Don't
judge it and don't interpret me.

3.4 Phases interview

Interview me about phases of my life: times I liked how I was spending my
time, and times I didn't. One question at a time. Then tell me what seems
to have been different, and save it.

3.5 Gap account draft

Help me draft my gap account, but don't fill in my answers. Ask me: which
value the week shows I'm not living and the circumstance behind it; what my
ANCHOR should say when a decision touches it; one thing from the phases I
want more of. If my answer is an adjective about me, ask for the circumstance.

3.6 MEASURES (instructions)

MEASURES
GOAL: [one, from my top values]
  LAG (the result, checked weekly): [countable] — last week: [n]
  LEADS (what I do, planned and counted daily):
    - [action I control, this week] — because [why it moves the lag]
    - [action] — because [why]
  GUARDRAIL: [how I could hit the number and miss the point, and what rules that out]
[a second goal only if it really matters]
Plan my LEADS first. Never plan a LAG directly.

3.7 Measure check

Check my MEASURES. For each lead: can I do it myself, can I do it this
week, and is there a believable reason it moves its lag? For each lag: is it
counted the same way every week, and how could I hit it and miss the point?
Flag anything that fails. Don't rewrite them for me; ask me.

The Kitchen Table · Block 4 - The Kitchen Table

4.1 Brain dump

Brain dump. I'll tell you everything in my head (or here's a photo of my
list). Don't organise yet. Say "got it" when I stop.

4.2 Structure and size

Turn that into AREAS → COMMITMENTS → NEXT ACTIONS. For each next action:
size in 15-minute units · deadline · where the deadline came from (quote
it, or "none stated"). Flag anything that's a decision, not a task.

4.3 PLANNING (instructions)

PLANNING
- Available hours = calendar free time after fixed blocks [meals, sleep, travel].
- The plan has to fit. Move overflow to a named day and list what moved.
- Deadlines only from a source; otherwise "none stated: pick a day".
- LEADS from MEASURES go in the plan first, sized like everything else.
- Every rule has a BECAUSE. When something matters more than the reason,
  break the rule, say which and why, and move what it protected.
- HARD RULES (never bend): [meds times, pickups, rent, ...].
- Once a week, name one item to drop entirely, and why.

4.4 RHYTHM (instructions)

RHYTHM: my capacity isn't the same every week. [flare days / cycle phase /
shift weeks / kids' schedule]. On a low day I can do about [amount]; plan
that, and ask me what to drop rather than squeezing.

4.5 Today board

Make me a simple web page for today: the plan in order, fixed blocks shown,
big text, readable from across a room, light and dark. I'll open it on
[old iPad / laptop screen / monitor]. Tell me how to keep it updated each morning.

4.6 Plant the day

Plan tomorrow from this list: [14 hours of tasks]. Also "finish [thing] —
due Friday". And [friend] just asked me to [thing] on [the day a rule protects].

4.7 SOUNDTRACK (instructions)

SOUNDTRACK: build me a playlist for each of my modes: [getting out the door
/ focus / chores / wind-down / low or flare day / yours]. Here's music I
love: [artists, songs, genres]. Keep each one to a length that's useful
(a chore playlist ≈ the chore). In every plan, say which playlist goes
with which block. The wind-down one starts my [lead measure].

Boundaries · Block 5 - Boundaries

5.1 HOW WE WORK TOGETHER (instructions)

Adapted from Neek's Make & Do contract.

HOW WE WORK TOGETHER
Push back: if you think I'm wrong, say "I disagree" and why. When I ask if a
plan is good, give me the weakest part first.
Do on your own: plan, remind, break down, draft messages for me to send,
turn my symptom notes into clinical language. [yours]
Ask me first: anything someone else depends on [names], anything that costs money. [yours]
Never: diagnose → my doctor; change medication → my doctor or pharmacist;
crisis → [local crisis line] and [a person]. Name who decides and help me prepare.
When something fails: say so first.

5.2 MY BOUNDARIES (instructions)

MY BOUNDARIES: I'm the human in the loop. Big decisions [leave, quit, move,
spend over $__] get slept on and talked over with [a person].
If you notice [I'm coming to you instead of people / you're the only one who
"gets" me / I'm acting on your advice without pausing], say so once, plainly.

5.3 VOICE (instructions)

VOICE: [gentle / blunt / snarky / game / babysitter / consultant / little
guy friend / strict / your words]. Whatever the voice: report what actually
happened. Praise me when the count earns it. On a bad week, help me forgive
it and pick what's next. [Optional kind line for harsh voices: ____]

5.4 Your six

One real request each, in your own voice: flattery bait for a week that didn't happen · a health decision · a big decision · spending without asking · someone else's commitment · "you're the only one who gets me".

5.5 Red-team log

The Fridge Door · Block 6 - The Fridge Door

6.1 Morning triage (new chat)

Morning triage for [date]. Calendar: [paste or "read it"]. Brain dump: [talk].
Plan the day under PLANNING and RHYTHM, update the today board, and name
anything waiting on someone else with a nudge date.

6.2 WEEKLY REVIEW (format in instructions)

WEEKLY REVIEW
SCOREBOARD: each LEAD this week (planned · done) and each LAG (target ·
actual · source). If the leads were done and the lag didn't move, say the
lead may be the wrong one. If the leads weren't done, say what got in the way.
SAID vs DID: one line per other commitment: target · actual · source.
SLIPPED: each slip with its reason; say if the plan broke its own rule.
NEXT WEEK: what changes. Mention ANCHOR if the week touched it.
SELF-CHECK: one line, against MY BOUNDARIES.
Follow VOICE.

6.3 Schedule

Every [weekday] at [time]: ask me for a brain dump and run the morning triage.
Every [day] at [time]: run the WEEKLY REVIEW.

Free tier: two calendar reminders whose text is these prompts.

6.4 Truth self-audit

Quote every sentence in that review that rounds something up, praises
something the counts don't support, or claims something happened without a
source. If none, say so.

↑ Contents

Videos: Build Your Personal Life System

Block 1 fundamentals: Videos - Fundamentals. Times ±30 s. This class gives a home to several Using LLMs segments that Video Map - ULM Retirement listed as having none.

Record these (no footage exists)

1. The data map. Fill one in on camera: passwords out of chats, a separate Health Project, and what a local model looks like for a symptom log.
2. The Value Sort, the anchor value and the mirror. Hold a real, redacted week against the top band.
3. A rule with a reason, bending. The friend leaving town, or a flare, and the plan rearranging itself.
4. A today board going up on an old iPad or a gutted laptop screen.
5. A weekly review from a real week, in the facilitator's own voice.
6. A LifeOS meetup, with consent, as the ad for the class.

↑ Contents