Light evidence (grade D: anecdotal)Light evidence: a first-person post written about a day into use. Concrete workflow, but early and couldn't be checked.

A Two-Restaurant Owner Taught an AI Agent to Build Distributor Orders From the Chef's Sheet

A chef showed a helper how he builds the food order after a bad first try.

What happened After a failed first try, a Georgia restaurateur used Grok Bot's TEACH button to show an agent how he reads the chef's handwritten sheet and builds the distributor cart. On the next restaurant's order it did it "almost flawlessly", with one or two mistakes.

Who this is about: Two restaurants in Georgia (Suite 200 Grille, Milton; Whiskey & Water, Blue Ridge; chef-owner Anthony Battaglia)

Grok BotPFG (distributor ordering website)7shiftsResyHostieLocal & serviceApprove-to-send

Get the prompt

OperationsAdded Oct 16, 2026Source: x.com

What it takes

Cost: From $30/monthGrok Bot: Premium is $30 a month, Super $100 and Ultra $200, each billed monthly. Grok Bot is included with an eligible Premium, Super, Ultra, Teams or Enterprise plan.Price checked Oct 24, 2026 on x.ai. Prices change; check before you buy.
Setup time: An afternoonOur estimate from the source, for one business.
Skill needed: No codeApps and settings only.

Explain it simply

A chef who owns two restaurants asked a helper to turn the kitchen's handwritten sheet into an order with the food seller. The first try was a mess. He placed the order while the helper watched, and that became the rules. The helper then built the other restaurant's order almost perfectly.

New to AI? See how AI works

What was done

In one day he set up named agents for a chief of staff, 5 inboxes, guest follow-up, Hostie and Resy, 7shifts, marketing, bar variance, inventory-to-PFG ordering and reporting.

The first attempt at turning the chef's handwritten inventory sheet into a distributor cart went wrong: wrong workflow and several agents touching the cart.

He clicked TEACH, placed the order himself and showed how he reads the sheet, finds products, sets quantities and builds the cart; that became rules and a reusable process.

Other agents pull monthly sales reports to fill city and state tax and compliance PDFs, close the loop on guest calls and texts via the reservation book and 7shifts, and triage 5 inboxes.

How it went

Given the second restaurant's order, the agent went into the distributor site on its own and built the cart "almost flawlessly", with one or two mistakes, less than 24 hours after he started. He lists plenty of screwups too: duplicate emails, agents stepping on each other, bad assumptions and browser issues.

Very early: written about a day into use (about 24 hours in), with no time or cost figures.
It isn't stated whether the cart was submitted automatically or checked first.
Not promotional: he runs restaurants and doesn't sell the setup.

Prompt

PromptWritten by our editors from the source

Here's a photo of tonight's handwritten order sheet from [chef].

1. Read each line: product, quantity, unit. List anything you can't read.
2. Log into [distributor site] and find each product using our order guide [link or file]; if there's no exact match, list the closest options instead of choosing.
3. Build the cart but don't place the order.
4. Send me the cart with totals and any lines you weren't sure about.

Follow the rules we saved from the last order I showed you.

Setup notes

Show the agent one order end to end yourself, then save that as written rules before handing it the next one.
Give one agent sole ownership of the cart; he found several agents touching it caused the first failure.

Business fit

Who this helps, and how:

Independent restaurants: turn the chef's order sheet into a distributor cart for you to check.
Local & service
Bars, cafes and caterers: teach an agent your ordering routine once and reuse it every week.
Local & service

Your turn

Ideas to try on your own work. These are starters, not results anyone has measured.

Which handwritten sheet would you rather not type into an order cart?
What would you show a helper, step by step, after a bad first try?
Which mistake would you still look for before the order is sent?

Similar tools

Other tools for this job, and how they differ. Chosen by our editors, never paid for.

ChatGPT
Used in entries here to read shopping lists from photos of inventory sheets; you'd still build the cart yourself.
OpenClaw
Self-hosted, always-on agent framework with memory and tools; fully yours to configure and maintain.
Lindy
No-code agent builder with a large template library; credit-based pricing can make experimenting expensive.

“Light evidence” (grade D) means: A brief, early or in-progress claim. Treat it as an idea to test, not proof. How we grade. Spotted a mistake? Suggest a correction.