How we use AIA personal operating system, with AI in the room
Guardrails & principles

If it leaves the house, it waits for a yes.

The method tells AI what to do. The guardrails tell it when to stop. Three classic ideas do most of the work: poka-yoke (mistake-proofing), jidoka (a human stop) and a lite FMEA (what could go wrong, and what prevents it).

If it leaves the house, it waits for a yes.
Poka-yoke · mistake-proofing

Drafts, not sends

Nothing sends, spends, books or deletes without my yes. The safe path is the default path, so being careful isn't the only thing standing between a draft and a mistake.

Jidoka · a human stop

Stops and asks

When a specialist is unsure, it stops and the item goes in “Needs you.” It handles the normal case and stops at the edge case, instead of guessing.

Draft-and-hold, honestly

Specialists stay in their lanes and hand work back up through the communicator. I see a short summary with the drafts attached, tap yes / edit / no, and only then does anything go out. It's a little slower than full autopilot. That's the point — the tap is where my judgment lives.

What goes wrong

Bots stall. Syncs drop. So there's a reliability heartbeat three times a day that checks the team is actually alive, and the rule is fail loud: if something breaks, it tells me plainly instead of quietly pretending.

Who decides

Four jobs, one yes.

WhoDoes whatFor example
You decideWhat matters. Anything that sends, spends, deletes, books, or speaks for you to someone you love.Values. Approving a reply. Saying yes to a trip.
AI draftsFirst versions in your voice, held until you look.Email replies, briefs, guest messages, plans.
AI proposesSorted piles and suggested next moves.“Unsubscribe from these 6?” “This is waiting on you.”
AI executesOnly quiet, reversible, pre-approved work.Syncing capture, filing, running a standing brief, heartbeat checks.

What could go wrong, and what stops it.

CI tool · FMEA-lite

FMEA (failure mode and effects analysis) asks, for each step: what could fail, how bad would it be, how likely is it, and what prevents it? The classic version scores 1–10 and multiplies. The lite version uses high, medium and low, which is plenty for a household.

Example
What could go wrongEffectSeverityLikelihoodRiskGuardrail
Emmy drafts the wrong tone to a close friendHurt feelingsHighMed HighFamily and close friends get extra care; nothing sends without my yes (poka-yoke)
Penny misses a bill reminderA payment slips off the plan, stressHighLow MedBill senders route to Penny; a red tile three days out
Scout books the wrong datesPlans broken, money spentHighLow MedScout never books: options and a draft plan only; I book
A routine stalls quietlyA brief never arrives and I don't noticeMedMed MedHeartbeat three times a day; fail loud
Concierge shares a private detail with a guestPrivacy and safetyHighLow MedCodes and private details stay out of drafts (“keep out”); drafts held for my yes
Memory sync drops a dayThe system forgets contextLowMed LowThe heartbeat checks sync; gaps show up in the morning brief

From my houseWhen a new bot joins the team, it gets a RACI row and a few lines here before it gets a single task.

Print the blank →
Guiding principles

The filter everything passes through.

Your list will be different. These are mine, in my words.

Feed it so it learns you

Share context, preferences, the human stuff. The stack only gets sharp if you feed it — living context, not cold documentation.

Love as the filter

Presence over busy. If it crowds the people we love, it doesn't ship.

Have a communicator

One chief-of-staff voice that routes, briefs and talks human, so specialists don't scatter noise.

Make it visual

Dashboards, the team map, week-at-a-glance. Don't bury the system in text-only tools.

Frictionless

Lowest-friction capture wins. Voice beats typing; a wearable; ambient taps someday.

Sundays for being present

Away from screens and tech. More present time is the whole point of the system.

A thoughtful partner

Not a guilt machine. Fail loud, coach in the moment.

Human voice, always

Especially with guests and people we love. Never canned.

Quiet backends

Memory lives in the archive and the vault. Chat and the visual home are the only things that push.

Cost-honest, lean

Tools earn their keep. Systems that stick — so the family has room.