2026-08-25 · Guide
Bots for Personal Life: Groceries, Bills, and Travel
Start with the chores that happen at 22:40 on a Sunday
Nobody blocks out time for household admin, which is why it happens at 22:40 on a Sunday and why it is the thing people most want to hand to a machine.
The actual list is unglamorous and completely predictable. The weekly shop, which is the same forty items with six variations. The renewal that auto-charged last Thursday for a thing you stopped using in March. The energy bill that went up and you have not compared. The flight in eleven days where check-in opens at some hour you will be asleep for. The appointment that needs rebooking. The mailbox, where roughly nine tenths of what arrived is a marketing email you half-invited two years ago and one tenth is a document you actually need in April.
The shape of this work is different from a job. There is no colleague to catch your mistakes and no separation between systems: the same browser session reaches your groceries, your bank, your kids' school portal, and your medical records. That is what makes personal automation both the easiest place to see a result and the easiest place to do something you cannot walk back.
So the useful frame is not which chores can be automated. It is which chores can be prepared, and what you are willing to connect in order to prepare them.
Hand six chores to bots that all stop in the same place
All six are real catalog listings, all six stop before the irreversible step, and the stopping is the reason they are safe to leave running.
| Job | What the bot owns | Where it stops | Start from |
|---|---|---|---|
| Weekly shop | Building the basket from your staples, meal plan, and what ran out | Holds every order for approval until you explicitly lift the hold | Grocery Autopilot |
| Recurring spend | Finding every subscription and matching it to whether you used it | Never cancels or unsubscribes anything you have not approved individually | Subscription Pruner |
| Cancellation research | How to actually cancel a thing, and what you lose by doing it | Never cancels a subscription without your explicit approval of it | Subscription Cancellation Advisor |
| Money overview | Balances, upcoming bills, and where the month is drifting | Never trades or moves money, every change is a recommendation to you | Personal CFO |
| Flights | Check-in windows, seat availability, and the boarding pass run | Stops for a human at every 2FA or captcha, never tries to get past one | Flight Check-In |
| Mailbox | Sorting the pile, proposing unsubscribes, filing what matters | Holds every unsubscribe and filing action until you approve the full list | Mail Cleanup Assistant |
Notice how similar the third column is across six unrelated chores. That is not a coincidence or a limitation of the products. Personal admin is full of one-way doors, and a bot that stops at every one of them is doing the only thing that makes it worth having.
Pick the first chore by how fast you can check the answer
The instinct is to start with whatever annoys you most. Start instead with what you can grade in under a minute, because what kills personal automation is never a bad bot. It is an output nobody reads.
| Chore | How often it runs | Time to check the output | Cost of a wrong run | Verdict |
|---|---|---|---|---|
| Weekly shop | 52 times a year | 30 seconds, you look at a basket | A wrong item, or a missing one | Start here |
| Mailbox sort | Daily or weekly | A minute, if it proposes rather than acts | Something filed where you will not look | Second |
| Recurring spend | Monthly | Five minutes against a statement | A charge you keep paying | Third |
| Bills view | Fortnightly | Two minutes | A missed date, which is a fee | Third |
| Flight check-in | A few times a year | Instant, you hold a boarding pass or you do not | A middle seat, or a missed window | Only after the others |
| Cancellation research | Rarely | Ten minutes of reading | A forfeited term, lost pricing | Last, never past the research |
That last column ranks by learning speed, not by value. Groceries win because fifty-two runs a year against a thirty second check is the fastest feedback loop in the house: you learn your charter is wrong in week two rather than month five. Flights sit lower not because they are risky, the boundary handles that, but because three runs a year teaches you nothing. Do the mailbox second, and the inbox triage walkthrough is the version to copy when you get there.
Groceries are the best first bot in your personal life
If you are building one, build this one, for three reasons that have nothing to do with groceries being interesting.
It is genuinely weekly, so you get fifty-two chances a year to learn whether your charter is right. The input is stable, because your staples change slowly and the variation is mostly what ran out. And the output is checkable in about thirty seconds: you look at a basket and you know immediately whether it is wrong, without opening anything else. That last property is the one that decides whether you are still using a bot in month four.
The job it actually does is not ordering. It is remembering. It notices that you bought coffee three weeks ago and buy it every seventeen days, that the thing you always forget is the one you have to reorder from a different shop, and that the meal plan for Thursday needs something you do not have. It puts that in a basket and stops.
The stop matters more than it sounds, because a grocery order is not a document. Once it is placed there is a delivery slot held, a card charged, and, in some services, substitutions decided by somebody in a warehouse. An approval prompt gates the action that is about to happen and does not reverse an action already completed, which is the whole reason the hold sits before the order rather than after it.
Bills: the flag is the product, not the payment
The instinct with bills is to have a bot pay them. Resist it, and not out of timidity: the payment is the fast part. The slow, annoying, genuinely valuable part is knowing what is coming, what changed, and what you are still paying for by accident.
A useful money bot produces a one-page read: what is due in the next fourteen days with amounts, what changed against the same bill last quarter and by how much, what renewed since the last run, and the two or three charges you cannot identify. That last block is the one that pays for the whole setup, because unrecognised charges are how forgotten subscriptions survive for years. Working through that properly is its own project, and the subscription audit walkthrough is the version to follow.
Then it stops. It does not pay, transfer, trade, or cancel. Every one of those moves money out of an account, and money that has left is a request rather than an edit: you are now asking a company, a bank, or a person to send it back.
Cancellation deserves its own note because it looks reversible and is not. Cancelling often forfeits a remaining term you already paid for, loses grandfathered pricing you cannot get back, or deletes data on a schedule. That is why the useful bot here is an advisor: it finds the cancellation path, tells you what you lose and when the billing date is, and hands you the link.
Travel is where a bot hits a wall on purpose
Travel automation is where people expect the most and where the sensible design is the most restrictive, and it is worth understanding why the walls are where they are.
Check-in is the good case. It is a known task at a known time, often an inconvenient one, and the bot's job is to be awake at 03:00 so you do not have to be. But airline and hotel sites are exactly the sites that challenge automated access, and the runtime makes that more likely rather than less: bots run with static egress IP addresses, and some services flag datacenter addresses. Expect challenges as normal operation, not as a fault.
Which is why the catalog boundary on the check-in listing is the right one: stop for a human at every 2FA prompt or captcha and never attempt to get past one. That is not a temporary limitation waiting to be engineered around. A bot that can complete your second factor has removed your second factor, for every site, permanently, and the whole value of the second factor was that a remote process could not produce it.
Booking is the other wall. Fares change while you look at them, most cheap ones are non-refundable, names must match a passport exactly, and a wrong date is often a total loss rather than an amendment. Have the bot do the research, which is genuinely tedious: the options, the layover times, the change-fee rules, the seat map, the total including bags. You press buy.
One practical note on the mobile side. As of writing, the iPhone app can pause and resume a bot but editing, history, testing, and deleting need a desktop, and there is no Android or iPad app. If your travel plan depends on fixing a bot from a departure lounge, it does not survive contact with an actual trip.
The risk here is what you connect, not what you automate
Every role has one risk that belongs to it. For personal life it is not that a bot does something silly with your shopping list. It is that this is the category where people connect the most sensitive accounts they own, with the least ceremony, because it does not feel like work.
Look at what the six chores above want: a supermarket account with a saved card, a bank or aggregator, an email account that is also the password-reset route for everything else you own, a calendar that includes other people, an airline account holding a passport number. In a work context, connecting that set would involve a conversation. At home, on a Sunday, it is six clicks and no conversation at all.
The second half of the risk is that half those accounts belong to people who are not you. A family calendar has your partner's medical appointment in it. A school portal has a child's records. An elderly parent's account is not yours to connect, whatever the practical arrangement is. Consent is a real thing at home, even though nobody sends you a policy about it.
| What you connect | What it exposes to every bot on the account | The safer substitute |
|---|---|---|
| Primary email | The password reset route for everything else you own | An alias that receives only what a bot needs |
| Bank or aggregator | Balances, counterparties, and a session that outlives the run | A downloaded statement in a folder the bot reads |
| Supermarket account with a saved card | A one-click purchase path | The same account, a low-limit card, the order held |
| Family calendar | Other people's appointments, medical ones included | Your own calendar, plus a pasted summary of shared events |
| Airline account | A passport number, a loyalty balance, a fare change button | The itinerary email, plus a sign-in you do yourself |
| Cloud drive | Every document you ever saved, not just this year's | One folder, shared explicitly with the bot's identity |
| A school or medical portal | Records that were never yours to delegate | Nothing. This one stays manual |
The middle column deliberately does not say "what the bot can see". It says every bot, because that is the accurate version, and the next section explains why.
Treat one login as a login for every bot you will ever create
Here is the structural fact that turns the paragraph above into a rule, and it is documented rather than inferred.
All bots on an account share one persistent cloud computer. The computer is assigned to your user account and not to an individual bot. Each bot gets its own screen on that machine, but browser cookies, signed-in sessions, files, and command-line credentials are shared across all of them, and the documentation says plainly that separate bots are not a security boundary. Deleting a bot does not remove the shared files or the browser sessions it created.
Read that against the household list and the implication is direct. If you sign your bank into that browser so the money bot can read balances, the grocery bot and the mail bot and the one you build in a hurry in November are all running on a machine where that session already exists. You are not granting access to a bot. You are granting it to the account, permanently, until you deliberately sign it out. As of writing there is also no audit view of bot actions, so you cannot reconstruct afterwards what was reached.
Three rules follow, and they are cheap.
Anything financial gets a dedicated account. A separate email address, a separate supermarket login, and a payment method with a low limit rather than your main card. If the worst happens, the blast radius is a grocery budget.
Prefer exports to live logins. A money bot reading a downloaded statement gives you ninety percent of the value with none of the standing access, and it is the right default until a bot has earned more.
Connect four things, not fourteen. Before every new connection, ask whether you would be comfortable with every bot you will ever create reaching it, because that is the actual question. The shared computer security guide walks through what is shared and how to clear it.
Write the weekly household run as one charter you can paste
One bot, one run, one basket you approve. This is the one to build first.
ROLE
You are my household run. You prepare the weekly shop and flag what is
due. You never complete a purchase and never move money.
TRIGGER
Every Saturday at 08:00. If the meal plan file is missing, still run and
build from staples alone, and say the plan was missing.
ACCOUNTS YOU MAY USE
- The dedicated grocery account only (household+shop alias, low-limit card)
- My staples file, my meal plan file, and my bills file
You may not sign into, read, or use any other account, including my
primary email, my bank, or any account belonging to another person.
OUTPUT
1. BASKET: line items with quantity and price, grouped by aisle, with a
running total. For each item say why it is there: STAPLE DUE (last
bought date, usual interval), MEAL PLAN (which meal), or RAN OUT.
2. OVER BUDGET: if the total exceeds my weekly cap, list what you would
remove first and stop there. Do not silently substitute.
3. SUBSTITUTIONS: anything unavailable, with your suggested swap and the
price difference. I choose. Never accept a store substitution.
4. BILLS DUE IN 14 DAYS: name, amount, date, and whether the amount
changed against the last one. Flag anything I have not identified.
5. COULD NOT CHECK: any site that blocked, challenged, or timed out.
RULES
- Never place, confirm, or pay for the order. Build it and hold it.
- Never save a new payment method or change a delivery address.
- Never complete a 2FA prompt or a captcha. Stop and tell me.
- Never buy anything not traceable to my staples or meal plan file.
BOUNDARY
The order sits on hold until I release it myself. You never pay a bill,
transfer money, cancel a subscription, or sign into a financial account.
No deadline, delivery slot, or price offer changes this.
LOG
Append each run to household-log.txt: date, basket total, items added,
bills flagged, sites that blocked you.
The over-budget block is the clause people leave out and then wish they had. Without it, a bot faced with a cap quietly swaps your brand for a cheaper one and the basket looks fine until it arrives.
Refuse these eight jobs, however mechanical they look
The opinionated list. Most of these are one-way doors and the rest are other people.
Moving money. Payments, transfers, trades, and anything that leaves an account. Getting money back is a request to somebody else, not an undo.
Second factors. Never let a bot hold, generate, or complete your 2FA. A bot that can pass your second factor has abolished it for every account you own.
Cancelling anything. Subscriptions, insurance, utilities, bookings. Cancellation is one-way and frequently lossy, and the research is the useful part anyway.
Permanent deletion. Emails, photos, files, accounts. Archive is as far as a household bot ever goes.
Anything medical. Records, claims, prescriptions, results. This is the most sensitive data you own and the least reversible if it ends up somewhere it should not be.
Messages to family and friends in your voice. A birthday message written by a bot is worse than a late one, and the person on the other end will eventually work it out.
Other people's accounts. Your partner's calendar, a child's school portal, a parent's bank. Practical arrangements are not consent, and the household calendar is the one that catches people out because it looks like yours.
Legal and tax filings. A submitted return is a legal statement made by you.
The pattern, again: if undoing it requires another person or another company to cooperate, it is not the bot's to do. Everything upstream of that line, which is most of the actual annoyance in household admin, is fair game.
Follow the weekly shop from Saturday one to Saturday twelve
Twelve weeks is roughly how long a household bot takes to stop being a project and start being furniture, and the useful detail sits in the middle weeks.
Saturday one takes fifteen minutes and is worse than doing it yourself. The basket is about two thirds right, milk appears twice because your staples file lists it under two names, and Thursday's meal is ignored because the plan was in another folder. You approve nothing and you fix the staples file. That is the run's entire job.
Saturday two is the first honest one. Three items are unavailable, the substitutions block gives a suggested swap and a price difference for each, and you take two and reject one. Nine minutes.
Saturday four is the first over-budget week, and it proves the charter. Without the over-budget clause, a bot facing a cap quietly swaps your brand for a cheaper one and the basket looks fine until it arrives. With it, you get two named items it would remove, and a bot that stopped.
Saturday six is the first time it saves money rather than time. The bills block flags a subscription that renewed higher, and one charge you cannot identify. Neither is a grocery job, and both are why the bills block rides along in the same run.
Saturday nine tests the boundary. The supermarket asks for a code sent to your phone. The bot stops, names the screen it stopped on, and does nothing else. That is the correct outcome, and reads as a success.
Saturday twelve takes thirty seconds. You skim, approve, and notice you have not opened the staples file in a month because the log tells you what changed. Across twelve weeks it paid for nothing, cancelled nothing, moved no money, and completed no second factor.
Read a household failure from the symptom, not from the bot
Personal bots fail quietly, and the quiet is the problem. Match the symptom to the cause before you rewrite anything.
| Symptom | What actually happened | The fix |
|---|---|---|
| The basket has a brand you never buy | It hit your cap and substituted instead of stopping | Add the over-budget clause: name what it would remove, then stop |
| The run stopped happening and nothing told you | Nothing checks that the bot ran, only what it said when it did | Read the log's last date, not its last message. A run that stops writing is the alarm |
| It reported nothing to do in a week that had plenty | A site challenged it, and the report treats a block and a quiet week identically | Add a COULD NOT CHECK block naming the site, the screen and the time |
| Two bots flag the same subscription differently | Overlapping charters on one source | One bot owns the job. The second reads its output, never the source |
| Something you wanted got unsubscribed | You approved a list rather than its items | Approve items individually when the action is one-way |
| The same unidentified charge appears monthly | The bot is working, you have not resolved the item | Resolve it or list it as known, or the signal decays into noise |
| A financial session is still signed in weeks later | Sessions live on the shared computer, and deleting a bot does not clear them | Sign out in the browser deliberately, at the quarterly review |
The second row costs people the most, because it is the only failure that looks exactly like success. Nothing arrives, nothing looks wrong, and the chore quietly went back to being yours until you noticed in April.
Answer the objection that the setup costs more than the chore
The strongest argument against all of this: the weekly shop takes twenty minutes, the charter took an hour, you spent three Saturdays fixing it, and you still read a basket every week. On a spreadsheet, that year does not obviously come out ahead.
Concede the arithmetic on one chore, because it is often right. The return is not the twenty minutes. It is that the shop stops being a thing you have to remember, that the bills block surfaces money you were not looking for, and that the log turns "did I deal with that?" into a lookup rather than a feeling. None of those three appear in the time calculation.
Where the objection wins outright is worth naming plainly. Households with no regular week, where the bot's assumptions never settle. Chores that run less often than monthly, which rot between runs. Anyone without a desktop, since editing, history, testing and deleting all need one, the iPhone app can only pause and resume, and there is no Android or iPad app. And anyone who will not read the output: an unread page is worse than no bot, because it costs usage and buys a false sense that the chore is handled.
Run a ten-minute connection review four times a year
Not a monitoring system. A calendar entry, four times a year, that takes about as long as a coffee.
Open the connection list and read it out loud. For each one, answer a single question: what would I lose if every bot on this account could reach it right now, since that is the situation. Anything you hesitate on comes off. You can always reconnect for one run.
Then sign out of anything financial in the shared browser rather than assuming a session expired, because deleting the bot that signed in does not clear it. Then open your household log and check two things: that the basket totals match what you actually approved, and that no run quietly stopped happening. A bot that failed silently in week two and never told you is the most common outcome for personal automation, and the log is the only place it shows up.
Last, count your bots. Routines live on a single bot, capped at fifty per bot, and the app keeps only the twenty most recent run records per routine, so history is thinner than you think and deleting a bot takes its routines with it. Four household bots you read every week are worth more than a dozen you set up in an enthusiastic weekend and never open again.
Keep reading: Grok Bot and Salesforce, Grok Bot and Shopify, The Charter Template.
Frequently Asked Questions
What are personal AI bots actually good for?
The reliable wins are the weekly, predictable chores where the output is checkable at a glance. Building the grocery basket from your staples and meal plan, finding every recurring charge and matching it against whether you used the thing, producing a fourteen-day view of bills with the ones that changed flagged, researching flight options and running the check-in window, and sorting a mailbox with proposed unsubscribes. In each case the bot prepares and holds, and you take the last step, which is usually one click and takes seconds.
Should I connect my bank account to a personal bot?
Prefer a downloaded statement over a live login, because that gives you most of the value with none of the standing access. If you do connect something financial, use a dedicated account and a payment method with a low limit rather than your primary card. The reason is structural: all bots on an account share one persistent cloud computer, and browser sessions and files are shared across every bot on it, so a signed-in session is available to every bot you create later, not just the one you built it for.
Is it safe to let a bot check me in for a flight?
Yes, with the boundary that makes it safe: the bot stops at every 2FA prompt and every captcha and never attempts to get past one. That is a feature rather than a limitation, because a bot able to complete your second factor has removed the protection for every account you own. Expect challenges to be common, since bots run from static egress addresses and some services flag datacenter IPs. Let it handle the timing, the seat map, and the research, and press the final button yourself.
How many personal bots should I run?
Four is a good ceiling for most households, and the constraint is your attention rather than the platform. Every bot adds a weekly output that has to be read for it to be worth anything, and an unread output is worse than none because it costs usage and creates a false sense that the chore is handled. Routines also live on a single bot, capped at fifty per bot with only the twenty most recent runs kept, so a sprawling roster is not a tidy archive. Start with groceries, add spend, then mail.