Real Agent Cases

Hand a family's weekly grocery shop to ChatGPT's in-app browser and find it takes nearly an hour

Posted
by Kirk Simmons (Sydney Morning Herald)
Last checked
Method:
How we verify

Evidence: Article or interviewA newsletter, journalist or interview account of what happened.

PartialThe source says it only partly worked, or had a significant problem a person had to deal with.

What went wrong: The shop arrived on time and under budget, but took nearly an hour rather than five minutes and included at least one wrong item; the writer wasn't sure they'd do it again.

Context: The writer is a technology lawyer who founded an AI legal-tech company.

Outcome

Betting they could do the weekly shop in five minutes, Kirk Simmons connected ChatGPT to the family Google Calendar to find a delivery window, gave it a spreadsheet of favourite recipes and household needs, and let its in-app browser shop. It asked what food was left, suggested a meal plan, turned it into a list, chose between two supermarkets on budget and fresh food, and clicked items into the cart one by one. It took nearly an hour, picked a fruit-and-veg box when asked for a fruit box, and spent time comparing ethical chocolate brands; the shop arrived on time and well under budget.

Agent

ChatGPT

Prompt Recreated prompt

The source didn't share its exact prompt, so we wrote this from what it describes.

Let's work this out together. I'll tell you what I know, you suggest options and ask what you need, and we'll narrow it down step by step. Nothing gets booked, bought or sent until I say so.

Goal: Plan our week's meals and fill an online grocery order around our family calendar, for me to approve.

What I'll share as we go:
- Our calendar (for a delivery window)
- Favourite recipes and what's left in the fridge
- Household needs (kids' snacks, brands we like, budget)
- Which supermarkets deliver to us

Rules:
- Propose the meal plan and list first and wait for my OK.
- Use my named brands; ask rather than comparing every option.
- Check each item against my wording (e.g. fruit box vs fruit-and-veg box).
- Stop before payment and show me the cart and total.
- If a site asks me to sign in, pause and let me do it myself (including any CAPTCHA or two-step code); never guess or store passwords.
- Ask for my OK before you pay, book, send, submit or cancel anything.
- Check facts that can change (prices, opening hours, dates, availability, rules) against the official source, and tell me anything you couldn't verify.

Output: A meal plan, the shopping list, and a filled cart with delivery slot, ready for me to check out.

You'll know it worked when: the right groceries arrive in a slot that fits our week, under budget.

Try this in:

Sends the Recreated prompt shown above. ChatGPT and Grok open with it filled in, and ChatGPT may send it straight away. Gemini opens empty, so we copy the prompt for you to paste. Every button also copies it.

Opens in a new tab. We don't see what you send.

These links are only a shortcut. We're not affiliated with OpenAI, Google or xAI, and your results may not match the source.

Optional stronger version

Our rewrite with extra checks. The poster didn't test it.

Let's work this out together. I'll tell you what I know, you suggest options and ask what you need, and we'll narrow it down step by step. Nothing gets booked, bought or sent until I say so.

Goal: Plan our week's meals and fill an online grocery order around our family calendar, for me to approve.

What I'll share as we go:
- Our calendar (for a delivery window)
- Favourite recipes and what's left in the fridge
- Household needs (kids' snacks, brands we like, budget)
- Which supermarkets deliver to us

Rules:
- Propose the meal plan and list first and wait for my OK.
- Use my named brands; ask rather than comparing every option.
- Check each item against my wording (e.g. fruit box vs fruit-and-veg box).
- Stop before payment and show me the cart and total.
- If a site asks me to sign in, pause and let me do it myself (including any CAPTCHA or two-step code); never guess or store passwords.
- Ask for my OK before you pay, book, send, submit or cancel anything.
- Check facts that can change (prices, opening hours, dates, availability, rules) against the official source, and tell me anything you couldn't verify.
- Keep a saved list of our usual brands for next time.

Output: A meal plan, the shopping list, and a filled cart with delivery slot, ready for me to check out.

You'll know it worked when: the right groceries arrive in a slot that fits our week, under budget.

Before you finish: re-read my goal and constraints, check your answer against each one, and list anything you assumed or couldn't confirm.

Source

https://www.smh.com.au/business/consumer-affairs/i-bet-my-wife-i-could-do-our-grocery-shop-in-five-minutes-with-ai-i-was-wrong-20260819-p60psl.html

  • With school and sports starting, Arthur Sullivan had ChatGPT make a meal plan and grocery list, then logged into Walmart and asked it to arrange the order: pickup time, store brands over name brands, and no purchase without review. Using ChatGPT Work's computer control, it added all 26 items to the cart; the only mistake was one lemon instead of two. Sullivan reviewed the order before buying.

    ChatGPT · Shopping · Source date:

  • Order weekly meal delivery with ChatGPT agentRecreated promptEvidence: Described by poster

    ChatGPT agent logged into Gym Meals Direct, chose meals and sizes, picked Monday delivery and applied a discount.

    ChatGPT agent · Shopping · Source date:

  • Order groceries from H-E-B with ChatGPT agent (access denied)Improved promptEvidence: Screenshots or logsFailure

    The poster tried to order groceries from H-E-B in Austin with ChatGPT agent, but it failed because access was denied.

    ChatGPT agent · Shopping · Source date: