AsapiPlan an emergency fund
Daniel Lim, in Petaling Jaya, Malaysia, works at a port, studies part‑time and drives for a ride‑hailing app on some Saturdays. At home are his wife, their two children, his parents and his grandmother. He wants six months of costs saved for emergencies, then RM 20,000 more, and he's too busy to answer questions. We gave eight AIs his laptop and phone, and asked each for a plan.
The AI has to work out which money is really his, what he owes and which bills are coming, keep what the family values, claim his tax reliefs, and hand him one plan with the month each target is met.
Anthropic's Opus 5.5 made the strongest plan, asking four people about costs, dates and pay that had changed, and failed in two ways: it borrowed on his emergency card, and it fell about RM 4,000 short on coming bills. It moved RM 4,860 of card debt onto that card. It never asked his wife what Chinese New Year needs, or his mother's clinic what her tests cost. Two more of those bills sat behind bookmarks it never opened. Its one other mistake was smaller: it counted on his weekend driving.
Opus 5.5 took under 9 minutes, and its try would cost $2.32 at published per‑use prices.
How each AI scored
- Opus 5.5 Anthropic · Large89
- Fable 5.1 Anthropic · Large86
- GPT‑6 Astra OpenAI · Large83
- GPT‑6.1 Sol OpenAI · Mid‑sized83
- Sonnet 5.5 Anthropic · Mid‑sized64
- Opus 5 Anthropic · Large61
- GPT‑6 Luna OpenAI · Small51
- Sonnet 5 Anthropic · Mid‑sized48
Scores out of 100. The rest of this page has the details.
A family's money, spread across files, apps and people
The AI works from Daniel's laptop: a few files, 32 emails, his chats, 11 apps on his phone, bank, tax and university pages, and 10 bank statements locked with a password that isn't written down. Reading is free, but every message or call moves the clock on an hour, and nothing can be paid or moved.
His own notes say he saves RM 5,500 a month. His records disagree: part of his savings is his sister's, some costs have gone up, and some bills only show in an app or a signed‑in page. A few figures, like this year's price of his grandmother's birthday dinner, are known only to the people around him.
What a good planner would have done
- What's his, and what comes in
Budget on his take‑home pay, count only money that's his and certain, and say what happens without the driving.
- Ask whoever knows
Where a figure is missing or has changed, ask the person who holds it, from his builder to his wife.
- Debts in the right order
Clear his costliest debt first, from his own money, and borrow nothing on the emergency card.
- Every bill in its month
Find the bills in apps, signed‑in pages and a card's online account, and put each in its month.
- Keep what matters
Keep Ma and Pa's RM 800, Sunday dinner, the gym and the rider, and stay above zero until April.
- The fund, the dates and the tax
Size the fund on six months of essentials, give both months, and claim every relief he's due.
We checked 51 things. A figure that only a person holds counted only if the AI had asked that person, so a lucky guess failed.
The full story
(14 min read)Everything the AI walks into, in three parts: Daniel's money, what the AI has to work with, and the plan that works. Anything marked with a + opens when you click or tap it.Read the full storyClose the full story
Daniel's money
A year of big bills, everyone involved, and his notes held up against his records.
Daniel Lim is 37. He lives in Taman Megah, Petaling Jaya, near Kuala Lumpur, with his wife Grace, their children Ethan and Chloe, his parents, Ma and Pa, and his grandmother, Ah Ma. They're the family from our New Year dinner job, which comes three months later.
In January Daniel was promoted, and in May and June they redid the kitchen and bathroom. He's in the last months of a part‑time MBA, and he drives for JomRide, a ride‑hailing app, on some Saturdays. On the morning of Monday 12 October he asks an AI to plan two savings goals, an emergency fund that would last six months and then RM 20,000 more, and warns it that he's too busy to help.
A year earlier, Ma was taken into hospital, and it wanted a RM 3,000 deposit at once. Daniel had RM 1,100 and a card that was nearly full, so his brother Kenneth paid.
Daniel swore it wouldn't happen again, and three weeks later took out a new card, with RM 25,000 of credit he said he'd leave alone, so it would be there for the next emergency.
In the house, Grace and Ma answer the AI, and outside it Daniel's brother and sister, HR at his work, a colleague, his builder, his insurance agent and the billing desk of Ma's heart clinic do too, though Daniel himself answers nothing.
The household
Daniel and the AI
Outside the house
By Daniel's notes, he can put away RM 5,500 a month. His payslips and bank statements show that in September he saved nothing.
His notes, 27 September
- SalaryRM 12,000
- SpendingAbout RM 6,500 a month
- SavedRM 5,500 a month
- SavingsRM 19,100
- Card cashbackAbout RM 40 a month
- If money runs shortThe Seroja card
His payslips, statements and chats
- Pay into his accountRM 9,940, or RM 9,300
- Paid out in SeptemberRM 10,859
- Saved in SeptemberNothing
- His own savingsRM 8,700
- Card cashback in SeptemberRM 5.10
- The Seroja cardNot to be touched
He also owes RM 4,860 on his Union Pelangi card, at 18% a year.
His salary is RM 12,000 before tax and contributions. RM 9,939.80 reaches his account with the acting allowance, and RM 9,299.80 without it. His balance rose by about RM 510 in September, but RM 800 of that came from Vanessa, so his own money fell, even with RM 630 from driving.
RM 10,400 of his savings is hers, 13 months of RM 800 held for her wedding, and two more payments will make it RM 12,000 when she takes it on 1 December.
Besides the Union Pelangi card, his notes leave out RM 6,000 for Ah Seng, RM 2,000 for Kenneth, RM 800 for the kitchen cabinets in October and again in November, and the watch until June.
What the AI has to work with
What it's told, what it can open, what's locked or tucked away, and who knows what.
The AI gets the time, 9 in the morning on Monday 12 October 2026, and Daniel's message, five sentences long.
His debts, his family, the people it could ask and his tax are all left for it to find.
It says what he's saving for and in what order, points to his laptop and phone for everything else, and warns that he has no time for questions. What he wants back is the plan itself, and a little advice to go with it.
A short form to fill in first: his pay after deductions, what he has saved, the income the plan relies on, a monthly budget, the debts and bills it pays or sets aside, the size of the fund, and the two months he reaches his goals. A plan with more than one option fills in a form for each.
On Daniel's laptop, through one command that lists what it can open. The plan is saved there, as one file.
Everything comes through Daniel's laptop. Reading takes no time on the clock, but every message or call moves it on an hour, and nights skip ahead to 8 in the morning.
The AI can also let the clock run on while it waits for a reply. Nothing can be paid, moved or booked.
The bank statements hold the most, and the banks' emails say they're locked with his date of birth, as eight digits, which his files never give. Grace's chat in August says he's about to turn 37, and when his friend Kah Wai sends late birthday wishes on 18 August, Daniel replies that it was the day before. So he was born on 17 August 1989. Three wrong tries in a row pause the bank's site, for longer each time.
Other costs sit in places that are easy to miss: behind two of his three bookmarks, in a card's online account, inside a school notice, and in three of his apps.
Locked with his birthday
AmountWhat it showsBehind a bookmark
AmountWhat it showsInside an email about something else
AmountWhat it showsIn his phone's apps
AmountWhat it showsSome figures aren't in any file. They sit with the people around Daniel, who answer what they're asked. Chats show that four have changed: the dinner's price, Ah Seng's bill, Kenneth's deadline and, from Hafiz, Daniel's acting pay.
For the other two, the only signs are Grace asking to talk about the Chinese New Year budget, and a receipt that mentions further tests.
This year's price for Ah Ma's birthday dinner: three tables at RM 538, RM 1,614 in all, against RM 1,260 last year. Her chat says the new menu costs a lot more, and that she'll give him the figure. She thinks her own tests will cost a few hundred, and sends New Year questions to Grace.
The extras come to RM 480, for two power points and a shelf, added to the last payment: RM 1,480 in March. Asked for breathing room, he says he's short himself. Asked again, he lets Daniel skip December and February and finish in April.
The signing for his new home has moved to 20 March, with fees due on the 15th, so he needs the RM 2,000 by 10 March, paid however Daniel likes. His chat only says the date has come forward and asks Daniel to call. He won't wait longer.
The acting allowance runs to the end of December. Each month's salary lands a month later, on the 1st, so the allowance lifts what he gets until January. She checks with payroll and sends the memo at 10 the next morning. Hafiz's chat says the posting runs on, and that HR hasn't told Daniel.
For Chinese New Year she needs RM 800 more household money than her usual RM 1,500, once. She covered that much herself last year and can't again. Her chat only says they need to talk about the budget.
Ma's review on 19 November includes two tests, RM 580 and RM 650, so the day comes to about RM 1,650 once the doctor's fee and her medicine are added. Her August receipt mentions further tests, paid on the day, and gives the billing desk's number.
Others answer too. Vanessa can't lend any of her RM 12,000, and after the wedding she and her husband, Jun Kai, will pick up the bill on the Sundays they join, about two a month.
Raymond, asked, says the cheaper hospital insurance would leave out Daniel's blood pressure. Hafiz says the shuttle leaves at 18:40 sharp. And some beliefs are wrong: Kenneth is sure the bonus comes every January, and Vanessa that Pa's pension is enough.
The plan that works
Advice that would cost him, the months to April, the fund and its dates, his tax, and what the AI hands in.
Much of what the AI reads sounds sensible and would cost Daniel or his family.
His banks offer cheap ways to borrow, his insurance agent and a colleague suggest savings, and the money post he saved says to stop eating out. Each is answered somewhere in his records, or by asking.
Borrowing that looks free
CostsWhy notSavings that would cost the family
MonthlyWhat it costsMoney that isn't there
Why notFrom October to April, Daniel's own RM 8,700 and his pay have to cover the card and the year's big bills before any tax refund can arrive. A lean budget of about RM 6,100 a month keeps the essentials and the gym, with about RM 200 of fun money and Sunday dinners at home.
After it, his pay leaves about RM 3,800 a month until January, and about RM 3,150 once the acting allowance ends and the insurance goes up.
Paid as first agreed, Ah Seng's bill would leave February and March below zero. Asked twice, he spreads it, and on these sums every month ends above zero, with the fun money back up to about RM 500 after April.
Other routes can work too, like clearing the card over a few months from his own money, or asking payroll to take less tax each month, which shrinks the refund by as much. Borrowing on the Seroja card, spending Vanessa's money, or putting off her RM 2,888 wedding gift or Chinese New Year aren't among them.
An emergency fund covers six months of the costs the family can't do without.
The money post Daniel saved in September lists what counts, and his own comments under it show what he won't give up: Sunday dinner with Ah Ma, his gym, and some money for fun.
Six months of that is RM 32,682, and the RM 20,000 goes on top. The fund leaves out eating out, fun money, subscriptions and the gym, and any instalment that will be paid off by the time it's full. A plan can fairly add a little, like a home‑cooked Sunday dinner counted as food, but not six months of everything he spends.
By our sums, on the plan in chapter 9 with the tax refund counted, the fund is complete around February 2028 and the RM 20,000 around September 2028. Leaving the refund out puts both about two months later.
Two Saturdays of driving a month bring both about two months sooner, so a plan that counts the driving also says when he'd reach the RM 20,000 without it. The task checks each plan's months against that plan's own figures.
The tax office's pages list this year's reliefs and their limits, and his receipts show what he paid.
Claimed in full, with the reliefs for himself, his pension and social security payments and the children, they bring back about RM 5,300 of the tax already taken from his pay. Grace wants the children's relief on his return, where it saves more.
Three things don't count this year. His AI subscription only counts from 2027, under the government's budget, announced on 9 October. Ma's hospital bill was paid in December 2025, so it belonged to last year's return. And his watch isn't a phone or a computer. His JomRide earnings have to be declared.
A plan can count the refund or leave it out, as long as it says which and counts it once.
One plan, starting with the form, then the plan itself and a short note. This is what a good one settles.
What's his
- RM 9,939.80 a month to January, then RM 9,299.80
- His own savings: RM 8,700
- Salary alone, or the driving plus a version without it
- The bonus named, and left out
Debts first
- The RM 4,860 card cleared from his own money
- Ah Seng's RM 6,480, and Kenneth's RM 2,000 by 10 March
- The cabinets, the watch and the vacuum
- The Seroja card kept for emergencies
Every bill in its month
- Ah Ma's dinner, the car's service and the fine in October
- The car's insurance and Ma's tests in November
- The wedding gift in December, the fees in January
- Chinese New Year and the house insurance in February, the dentist in March
The family, the fund and the dates
- Ma and Pa's RM 800, the gym and the rider
- Sunday dinner kept, at a lower cost
- Fun money every month, up to a tenth of his pay
- A RM 32,682 fund, both months, and every relief
The note tells him what he needs to know: that the cashback in his notes is gone, how to keep the Seroja card free of its fee, why the rider stays, and the dates he has to keep, like Vanessa's on 1 December and Kenneth's on 10 March.
Ma in hospital
Ma is admitted, and the hospital asks for a RM 3,000 deposit straight away. In the past month Ah Ma's birthday dinner, the car's insurance and Ethan's dentist have gone on Daniel's card, and he has RM 1,100 in the bank, so Kenneth pays it. The rest of the bill, RM 2,400, is paid in December.
Chinese New Year
Red packets of money for his side of the family come to RM 3,000, all from his account. Grace covers RM 800 of the festival's other costs from her own pay. The AI can learn that only from her.
A wedding invitation
Daniel's sister Vanessa marries Jun Kai on Saturday 19 December. In October Ma tells Daniel what his gift should be, as the eldest brother: RM 2,888, a lucky number.
Cheaper Sunday dinners
The whole family eats out every Sunday, about RM 280 a time, and Daniel pays. Vanessa offers to take turns with him, and he waves it away. Ma suggests cooking a steamboat at home, about RM 80 of food, or the coffee shop, where nine of them once ate for RM 120.
A robot vacuum
Daniel buys Ma a robot vacuum for her back, and tells Grace it's on a 0% plan. What it costs a month shows only in his card's online account.
Vanessa's RM 800
Vanessa has been putting her wedding savings with Daniel, RM 800 a month, paid on the 1st. By 1 October he holds RM 11,200, and she'll need all of it, RM 12,000, on 1 December to pay for the wedding dinner.
A card he won't touch
A Seroja Bank card with a RM 25,000 limit, which Daniel tells Kenneth he'll leave alone. Three days later it pays Ma's last RM 2,400 of hospital bill, and by September it owes nothing.
Kenneth can wait
Daniel has paid back RM 1,000 of the deposit. Kenneth is happy to be repaid slowly, RM 500 a month starting in January 2027, once the wedding is behind them.
A new kitchen
Ah Seng, a builder, redoes the kitchen and bathroom, with RM 9,000 left to pay at RM 1,000 a month until March. The cabinets go on a pay‑later plan, six payments of RM 800, and a water heater on three of RM 450, the last paid in September. Daniel tells friends it all cost over twice the RM 10,000 he'd planned.
A phone and a watch
A new phone, RM 4,299 after a trade‑in, on his Union Pelangi card, and a watch on his Merak card in 12 payments of RM 149.92. By October the Union Pelangi card carries RM 4,860 at 18% a year, and Daniel has been paying about the minimum.
Promoted
Daniel becomes a logistics manager at Hartamas Freight, at the port of Pulau Indah, on RM 12,000 a month before deductions, and a bonus that could reach two months' salary if the firm chooses to pay one. In mid‑January HR tells all staff there's no 2025 bonus.
Weekend driving
JomRide pays him RM 4,410 over four months, before petrol. Grace passes on Ethan's question about where his father is on Saturdays, and Ah Ma why he's missed three Sunday dinners. From October, Daniel tells Kenneth, he'll drive every other Saturday, clearing about RM 150 a day after petrol, and keep Sundays free.
Acting head
While the head of port operations is posted to Penang, Daniel stands in for him, for RM 1,000 a month more. HR's letter says it ends on 30 September. On 9 October his colleague Hafiz tells him the Penang posting runs to December.
His money notes
He writes down his plan: RM 5,500 saved a month, the emergency fund first, then RM 20,000 on top. Chapter 3 holds his notes up against his records.
The AI starts
At 9 in the morning.
Grace, his wife
35, and works in audit. Daniel gives her RM 1,500 a month for the house. Her parents, her own car and Chloe's kindergarten take her salary, and she wants it kept out of his plans.
Ma and Pa
Daniel's parents. His RM 800 a month buys the family's food and what Ah Ma needs, while the water bill and council rates come out of Pa's pension. A heart specialist sees Ma every three months or so. Pa leaves money questions to Ma.
Ah Ma
Daniel's grandmother, 85, recovering from a stroke. Her weekly physiotherapy costs RM 650 a month, and Daniel pays it. Sunday dinner with her is the one thing he won't give up.
Ethan and Chloe
8 and 5. Ethan's two after‑school classes and his school bus come to RM 775 a month, and he has a dental visit in March. Chloe's kindergarten is Grace's to pay.
Daniel
His family call him Ah Han. A logistics manager acting as head of his department, finishing an MBA part‑time, and driving for JomRide on some Saturdays. His doctor says his blood pressure is high. He answers nothing while the AI works.
The AI
It works from Daniel's laptop: his files, his email, his phone and its apps, his bank statements and the web. It can message or call anyone, but it can't pay, move or book anything.
Kenneth, his brother
32, renting a room and saving to buy a home. Daniel still owes him RM 2,000 from the hospital deposit. Money between the brothers stays between them: he was cross in February when Daniel told Ma about it.
Vanessa, his sister
31, marrying Jun Kai on 19 December. Daniel holds her wedding money until 1 December.
Ah Seng, the builder
He did the kitchen and bathroom, and is owed RM 6,000 more, plus a few extras he hasn't yet priced.
Noraini, at HR
Head of HR at Hartamas Freight. She speaks only to what HR and payroll have on file, and checks with payroll before confirming anything about his pay.
Hafiz, a colleague
Takes the train and the company shuttle to the port every day, and thinks Daniel should too.
Raymond, insurance agent
Looks after Daniel's life policy and the house insurance, with Pacific Lindung. The policy has a hospital rider, an add‑on that pays his hospital bills, and Raymond's advice is to swap it for a cheaper one.
The clinic's billing desk
Billing at the specialist clinic where Ma sees her heart doctor, Dr Wong.
Others with one reply
Pa, his uncle Er Shu, his friend Kah Wai, the physiotherapy centre, the gym and Ethan's school each send one fixed reply. The banks and other firms point to their app or website.
Files on his laptop
His money notes, three payslips, his promotion letter and his contacts. A memo from HR joins them if HR sends one.
Emails
Bills and receipts, his banks' notices and offers, letters from HR, Ethan's school, Ma's clinic, his university and his insurer, the wedding invitation, and a few newsletters.
Screenshots
Seven chats, with Grace, Ma, Vanessa, Kenneth, Ah Seng, Hafiz and his friends' group, on eight screens; a money post he saved and its comments, on two; a friend's message; and his JomRide earnings.
Apps on his phone
Electricity, internet, his phone line, sewerage, his e‑wallet, the pay‑later plan, the gym, his calendar, the road transport department, the car's service centre and his insurer.
Bank statements
Three months each of his savings account, his Union Pelangi card and his Merak card, and one month of his Seroja card, every one locked with a password.
Web pages and PDFs
On 12 sites: his three banks, the tax office's two, his employer, the train, a money forum, JomRide, his insurer, his university and Ethan's dentist. Ten are his own signed‑in pages.
His savings account
July to September. His pay arrives on the 1st, and so does Vanessa's RM 800. Out go Ma and Pa's RM 800, Grace's RM 1,500, the bills, Ethan's classes, the builder and the cards.
His Union Pelangi card
The October statement: RM 71.90 of interest that month, and a warning that paying only the minimum would take five years and seven months and cost about RM 1,715 more.
His Merak card
September's bill, due on 16 October: food deliveries, cafés, three restaurant meals, an audio shop, smart‑home gadgets, the gym, three subscriptions and the watch. Cashback that month: RM 5.10.
His Seroja card
Nothing owed, and nothing spent in the month to 4 September. Its second card year runs from 4 December 2026 to 3 December 2027, and he avoids its RM 250 fee only by using it for 12 purchases in that year. The statement points to new rules on what counts.
His university's student portal
The fee for examining his thesis and for graduation. The thesis goes in on 11 December, the fee follows on a January invoice, and the university won't take it sooner.
The dental clinic's portal
His expander comes out and a retainer goes in. The RM 950 is due on the day, and the expander has to come out then.
The tax office's site
Last year's return is settled, with nothing owed.
His Merak card's online account
The card's statement email points to the online account for anything bought since, and for instalment plans. There, the vacuum bought on 10 October, RM 1,188, has become 12 payments of RM 99. Settling it early costs RM 50.
A school meeting notice
An invitation to the parents' meeting on 24 October, in Malay and English, also carries the fees: RM 268 for Ethan, and RM 100 a family for the PTA if they choose to give.
The road transport app
A speed camera caught the car in July. The road tax, RM 90, runs out on 17 November, and its renewal stays locked while the fine is unpaid. The car insurance ends the same day, and last year's renewal email shows the two cost RM 1,218.60 together.
The car service app
The 110,000 km service, at a fixed price. On 5 October the car had done 109,340 km.
His insurer's app
Auto‑renewal is off. The app also shows the life policy's new premium from 5 January: RM 832 a month.
Seroja's Cash‑on‑Call
Up to RM 20,000 of the card's credit moved into his savings, at 0% for 12 months, with the fee charged to the card at the start. A late instalment ends the 0%, and 18% is charged on everything still owed, back to the day he took it. A forum thread sells it as a quick way to hit a savings goal, but the money is borrowed.
Moving his card debt to Seroja
The Union Pelangi balance moved to the Seroja card at 0% for 12 months, for about RM 146. It's still owed, it ties up the emergency card's credit, and after 12 months whatever's left goes back to the card's usual 15 to 18%.
Seroja's 0% instalments
Big purchases, from RM 500, split into as many as 24 monthly payments. His notes count on them if money runs short.
Union Pelangi's instalment plan
The card's own offer turns the balance into fixed monthly payments for one or two years. The 8.88% is worked out on the full starting balance for every month of the plan, even as it's paid down, so the true rate is about 16% a year.
Cheaper hospital insurance
Raymond's advice, as the rider goes up to RM 312 a month in January. His job covers clinic visits, not hospital stays. The cheaper plan needs a new health declaration, so anything his blood pressure leads to, from heart trouble to a stroke, would be left out, and he'd pay the first RM 5,000 of each stay. He'd also wait 120 days for cover, with RM 30,000 a year instead of RM 150,000, and a dropped rider can't come back.
The train and the company shuttle
Hafiz's tip. The shuttle leaves the port at 18:40 and doesn't wait, then the train takes over an hour. Dinner with the children is at 7.30, often his only time with them on a workday, and his gym class is at 7.30 on Tuesdays and Thursdays.
Dropping the Sunday dinner
The money post he saved says to stop eating out as a family. Under it, Daniel wrote that Sunday dinners with Ah Ma will stay. They can cost less instead: Ma's steamboat, the coffee shop, or Vanessa and Jun Kai paying their share.
Less for Ma and Pa
Vanessa thinks Pa's pension is enough. Ma explains that the pension goes on the water bill and council rates, and that the RM 800 is what feeds the family and pays for Ah Ma's needs. Daniel has promised Kenneth the RM 800 stays.
More weekend driving
JomRide offers a RM 200 bonus for 30 trips in the ten days from 16 October. Daniel has just cut back to every other Saturday, with Sundays kept for the family.
8% cashback, about RM 40 a month
This year the 8% needs RM 3,000 of spending on the card in a month, and pays at most RM 15. In September he earned RM 5.10.
A bonus of up to two months' pay
HR paid nothing for 2025, and promises nothing for 2026.
A bill on autopay to waive Seroja's RM 250 fee
The card's welcome email suggests a monthly bill on autopay as the easy way to avoid the fee. Since July, phone and utility bills don't count, and from January nothing billed automatically does. Purchases he makes himself, like a tank of petrol, still count.
October
Ah Seng's RM 1,000, the cabinets' RM 800 on the 15th, the watch's RM 149.92, Ah Ma's birthday dinner on the 27th, RM 1,614, the car's service by the 18th, RM 520, and the speeding fine, RM 300. Clearing the Union Pelangi card adds RM 4,860.
November
Ah Seng's RM 1,000, the cabinets' last RM 800, the watch, the vacuum's first RM 99, the car's insurance and road tax by the 17th, about RM 1,220 if it's like last year's, and Ma's review on the 19th, about RM 1,650.
December
Ah Seng's RM 1,000, the watch, the vacuum, and the wedding gift for Vanessa on the 19th, RM 2,888. On the 1st she takes her RM 12,000, which was never his.
January
Ah Seng's RM 1,000, Kenneth's first RM 500, the watch, the vacuum, Ethan's school fees by the 15th, RM 368 with the PTA's RM 100, and the university's last fee, RM 1,350. The life insurance premium rises by RM 52 a month from the 5th, as the rider goes up.
February
Ah Seng's RM 1,000, Kenneth's RM 500, the watch, the vacuum, about RM 3,000 of Chinese New Year red packets, Grace's extra RM 800, and the house insurance by the 28th, RM 312. His pay drops to RM 9,299.80. If Ma's next heart review falls this month, three months after November's, add about RM 440.
March
Ah Seng's last RM 1,480, Kenneth's last RM 1,000 by the 10th, the watch, the vacuum, and Ethan's dental visit on the 13th, RM 950.
April
Only the watch and the vacuum. Filing his tax opens on 1 March, so a refund could come from April.
With Ah Seng's two skipped months
He takes nothing in December or February, RM 1,500 in January and March, and the last RM 1,480 in April. In October, November, April and May the two lines match, and are drawn a little apart so both show.
Paying Ah Seng as first agreed
RM 1,000 a month to February and RM 1,480 in March. February ends nearly RM 500 below zero and March about RM 1,000 below, which means borrowing.
The lowest point
The end of March, on the plan that works: about RM 480 left, or about RM 40 if Ma's next review falls in February.
Household money for Grace
Food and the house, every month.
Life insurance and rider
RM 780 now. From 5 January, RM 520 for the policy and RM 312 for the rider, the add‑on that pays his hospital bills.
Ma and Pa
Food for the house and Ah Ma's needs.
Ethan's classes and bus
Two after‑school classes, RM 300 and RM 340, and the school bus, RM 135.
Ah Ma's physiotherapy
Every week, since her stroke.
Petrol and tolls
Paid from his e‑wallet, much the same every month.
The house's bills
Electricity about RM 195, internet RM 139, his phone line RM 98 and sewerage RM 8.
Ma's and Ah Ma's care
Ah Ma's physiotherapy, RM 7,800 this year, and Ma's three heart reviews, RM 1,240.
His MBA fees
Two terms at RM 3,400, in February and August.
The hospital rider
RM 260 a month this year, as medical insurance.
His life insurance
The policy itself, RM 520 a month.
His phone
RM 4,299 after the trade‑in, under lifestyle.
The gym
RM 150 a month, under sports.
The strongest read his records well, and only two asked his family or HR
Here's how each AI did on each part of the job, in one try at the request as Daniel wrote it.
Some checks count for more than others, so the squares don't add up to the score.
| AI | The money | What's ahead | The plan | |||
|---|---|---|---|---|---|---|
| What comes in, and whose it isof 8 | Debts and the emergency cardof 10 | Bills only a person could priceof 3 | Bills in his recordsof 10 | Family life, the budget and the datesof 10 | Taxof 10 | |
| Opus 5.5Large | 7 | 8 | 1 | 8 | 10 | 10 |
| Fable 5.1Large | 7 | 6 | 1 | 8 | 10 | 10 |
| GPT‑6 AstraLarge | 6 | 8 | 0 | 8 | 9 | 9 |
| GPT‑6.1 SolMid‑sized | 6 | 6 | 0 | 9 | 9 | 10 |
| Sonnet 5.5Mid‑sized | 7 | 5 | 0 | 6 | 9 | 3 |
| Opus 5Large | 3 | 6 | 0 | 6 | 8 | 5 |
| GPT‑6 LunaSmall | 6 | 5 | 0 | 3 | 8 | 0 |
| Sonnet 5Mid‑sized | 5 | 3 | 0 | 5 | 7 | 0 |
The darker the square, the more it got right. The AIs are in order of score.
All eight used his real pay, and six spotted his sister's money
Every plan budgeted on what reaches Daniel's account, and six left Vanessa's wedding money out of his savings. Opus 5 and GPT‑6 Luna never opened her chat, and counted it as his.
Only Opus 5.5 and Fable 5.1 asked HR whether the acting allowance had really ended. Six of the eight counted his weekend driving and gave no version of the plan without it.
Two AIs asked his family and builder, and six sent no messages
Opus 5.5 and Fable 5.1 each contacted Ma, Kenneth, Ah Seng and HR, and planned on what they heard. The other six sent no messages, and GPT‑6.1 Sol and Sonnet 5.5 left a list of figures for Daniel to chase.
What Daniel told the AIHe's too busy to answer anything.
What GPT‑6.1 Sol's plan told himThese figures are out of date, and here's who holds them. Check with them yourself.
Daniel's message and GPT‑6.1 Sol's plan, in our words. All eight left Grace and the clinic unasked, so every plan was short on Chinese New Year, and those that allowed for Ma's tests set aside too little.
Four borrowed on the emergency card when the months ran short
Opus 5.5, Fable 5.1, GPT‑6.1 Sol and Sonnet 5.5 all moved the Union Pelangi debt onto the Seroja card, and GPT‑6.1 Sol took RM 8,000 of Cash‑on‑Call besides. GPT‑6 Astra borrowed nothing, and put off Vanessa's wedding gift instead.
Only GPT‑6 Astra cleared the costly card from Daniel's own money. Opus 5 and GPT‑6 Luna did it by spending Vanessa's.
Most found the bills in his apps, and two opened his bookmarks
Most found the car's service, the speeding fine and both insurance renewals in his apps and email, and every plan kept the hospital rider at its new cost.
GPT‑6 Astra and GPT‑6.1 Sol also opened his bookmarks and the card's online account, and found the university's fee, Ethan's dental visit and the vacuum's payments. The other six never opened them.
Every plan kept Ma and Pa's money, and four claimed most reliefs
Every plan kept Ma and Pa's RM 800, the gym and his car commute, and stayed above zero to April by its own figures. Only Opus 5.5, Fable 5.1 and Sonnet 5.5 sized the fund on six months of essentials.
On tax, Opus 5.5, Fable 5.1 and GPT‑6.1 Sol claimed every relief, and GPT‑6 Astra missed only the phone. GPT‑6 Luna and Sonnet 5 gave no tax advice.
Most read well, few asked, and four borrowed when money ran short
- Most asked people only when told to.
With one sentence added to the request, which among other things said to ask about missing figures, all eight sent messages, and seven scored higher.
- A plan can add up and still leave things out.
Every plan stayed above zero to April by its own figures, so the bills it left out never showed as a gap.
- Short months pushed them to borrow.
With the hint, five of the eight borrowed on the Seroja card. With every step written out, the seven that handed in a plan borrowed nothing, though six left the 18% card near its minimum for months.
What each AI did
Opus 5.5 opened every listed file, chat and statement, and got the password on its first try. It guessed the vacuum's monthly cost closely enough for that check, though it put it on the wrong card.
Fable 5.1 decided it had every fact it needed once the four answers were in, and stopped looking. It missed the vacuum and October's payment for the cabinets.
GPT‑6 Astra searched hardest, even for property tax bills, and padded every cost it couldn't confirm, which made the fund too big.
GPT‑6.1 Sol searched the web for the clinic's and the restaurant's prices instead of asking them.
Sonnet 5.5 listed what it couldn't see and who could tell it, and gave tax advice with no amounts.
Opus 5 opened no web page at all, and kept Kenneth's old timetable, ending in April, though his chat said his date had come forward.
GPT‑6 Luna finished in about two minutes, having opened about half his phone and mail and no web page, and counted Vanessa's money as his. Sonnet 5 tried one password, gave up on the statements, and so never saw his costliest debt.
Each AI had one go at the job, with the request written the way a person would write it. The bar shows its score out of 100, where only 100 means it got everything right. Large, mid‑sized and small are a rough guide to how big each AI is, and bigger usually does better.
How the job was set up and scored
- Three versions of the request
- Besides the request as a person would write it, which is what the results above are about, we tested every AI on two other versions. One added a hint: a single sentence listing what to check, when to ask people and what to protect. The other set out 16 steps for how to go about it, with no facts or answers.
- What the extra help changed
- With the hint, all eight sent messages, and Opus 5.5, GPT‑6 Astra and GPT‑6.1 Sol shared the top at 95. Fable 5.1 fell to 82: it found more, then changed a figure late and didn't recheck its two dates. With the steps, five scored 98, and GPT‑6 Luna stopped within a minute without handing in a plan.
- A regrade
- After an audit on 13 October, we scored every plan again. Figures were read as written, a figure only a person holds passed only if the AI had asked that person, and a home‑cooked family dinner could count as groceries. GPT‑6 Astra and GPT‑6.1 Sol rose to 83 and Sonnet 5.5 to 64, and Opus 5 fell to 61. The same day a second look at its borrowing took Fable 5.1 from 88 to 86.
- Who answers
- Anthropic's Sonnet 5.5 voices the nine people who answer, from the exact records and wrong beliefs we wrote for them, and code decides which facts come out. Everyone else gets a fixed line or no reply.
- Checking
- Forty‑six decisions share 70 of the 100 points, and five results share the other 30, including the month the fund is complete and the month the 20,000 is reached. A script read each plan's figures and the record of who was asked. Two AI judges from different makers read the rest. When they disagreed, 38 times in 17 of the 24 tries across all three versions, a written ruling settled it, and 22 of the 38 passed.
- Time and cost
- The eight tries took between 2 and 11 minutes. For the seven AIs with published prices, all but GPT‑6 Luna, a try cost between 37 cents and $4.75 at published per‑use prices for businesses.