When should you use it?
- You are new to computer-using agents
- You want failure modes on a low-stakes surface first
- You compare prices across two delivery sites weekly
Not for: Letting the bot place the order without you; Using errands as cover to skip limits on customer tools; Practicing on live customer inboxes or production deploys
What do you need, and what do you get?
You need
- Item list with quantities
- Two stores or apps to compare
- Day and time for the weekly prompt
You get
- Side-by-side carts with delivery costs
- A dollar difference and a recommendation
- A wait state until you approve any order
How does it work, step by step?
- Pick an errandGroceries, group-order watch, subscription list audit — low blast radius.
- Compare, do not buyBuild carts or drafts and wait.
- Watch the failuresNote mismatches, missing items and confused UI moments.
- Only then widenAfter two clean weeks, consider higher-stakes permissions.
The prompt
Our own prompt, inspired by Carnage (@0xCarnagee) and written to work in Grok Bot, ChatGPT agent, Claude and others. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
Weekly errand practice — compare only. Day/time: {schedule}. Items and quantities: {items}. Store A: {store_a}. Store B: {store_b}. Compare items across both. Assemble the basket on whichever option has the lower all-in cost with delivery. Show the two baskets next to each other with the dollar gap and any substitutes. Leave checkout alone. Bring the comparison to me and wait. If a price or item is unclear, ask instead of guessing.
Inspired by Carnage (@0xCarnagee): X Article: The Grok Bot Team's Own Workflow (11 steps). The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Worst case should be a wrong bag, not a wrong customer email.
- Never place the order in the practice loop.
- Missed newsletter unsubscribes are useful lessons; missed invoice cancels are not.
- Distinct from deep bill audits; this is practice, not recovery.
Our take
Carnage puts grocery and delivery bots beside work bots on purpose: cheap errands reveal how an agent fails when the stakes are a missed carton, not a customer ticket. Practice there before you connect mail that carries your name. Errands make good practice because the worst outcome is a wrong item in a cart, not a mistaken message to a client. Pick something with a small blast radius, such as comparing a weekly grocery cart across two delivery sites or listing your subscriptions, and have the bot build carts or drafts without ever placing the order. Watch where it struggles: mismatched items, missing products, confusing screens. Those habits are the same ones that will show up later with work tools. After two clean weeks, consider giving it something with more consequence, and keep approval on anything outward. Do not use the practice loop as cover to skip limits on customer-facing tools, and never practice on live customer inboxes or production deploys. This is rehearsal, not money recovery.
Where the time goes back: You discover brittle agent behavior on carts instead of during customer sends, so high-stakes permissions come with fewer surprises.
FAQ
Why waste a bot on groceries?
How long should I practice?
Can it order after I approve?
Source and credit
Inspired by Carnage (@0xCarnagee). We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- X Article: The Grok Bot Team's Own Workflow (11 steps)
- Creator
- Carnage (@0xCarnagee)
- Published on
- X
- Original date
- Aug 19, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Grok Bot
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. Compare-and-wait; no purchase.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 24, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.