A Cake Studio Went Live on a Claude Code-Built Order System in 12 Days
A cake shop got one order system, and no email goes out until the owner says yes.
What happened A product executive built his wife's San Rafael cake studio a system for orders, inventory, recipes, payments and owner-approved customer emails in 12 days with Claude Code, and in its first week 8 of 9 customers who got a review request left a review.
Who this is about: Custom cake studio (Marin Cake Studio, San Rafael, California)
What it takes
- Cost
- Free planClaude: Free plan. Pro is $20 a month billed monthly ($17 a month billed yearly), and Max starts at $100 a month. Team seats are $25 a month billed monthly.Price checked Oct 24, 2026 on claude.com. Prices change; check before you buy.
- Setup time
- A few weeksOur estimate from the source, for one business.
- Skill needed
- Developer skillsYou build or run your own code, even if AI writes it.
Explain it simply
A cake shop tracked orders on paper, in several apps, and on the owner's phone. Her husband used Claude Code, an AI that writes software, to build one system. It covers orders, recipes, ingredients, payments, and customer email. It can draft a reminder or a review request, but nothing goes out until the owner says yes.
What was done
The bakery ran on Intuit invoicing, MailChimp, a paper calendar and the owner's phone, so a 9 pm text asking for a Saturday cake meant hand-making a quote, invoice, deposit, calendar entry, shopping list and later a review request.
Chad Holdorf built "Bakery Buddy" against the real business: orders, customers, inventory, recipes, calendar and campaigns, plus an assistant that answers questions from live business data. A daily engine drafts review requests and reminders, and nothing sends until the owner approves it.
How it went
The studio was running its real orders on the system 12 days after the first commit. In the first week the review automation went to nine customers and eight left a review. The builder found that the owner preferred asking the assistant (or being told) over clicking through dashboard pages, so the assistant became the core of the product.
- Limits: a nightly sync quietly marked paid customers as unpaid, and an agent ran out of tokens mid-task and silently created nothing. Both were found and fixed. A voice "Counter Mode" for taking orders hands-free can't be shown to have produced a sale, because orders don't record where they came from.
- He says he would add source tracking before building a feature next time. The builder is an experienced product executive who ships code, so this isn't a no-code project.
Prompt
1. The assistant calls real tools (orders, inventory, calendar) and returns live data, so numbers are never generated.
2. Daily engine: draft review requests and reminders; queue them for the owner's approval. Nothing sends without a yes.
3. Record the order source (text, web, counter, voice) on every order before building new intake channels.
4. Make failures loud: alert when a sync changes payment status or an agent run ends without creating anything.
The design rules the post describes; no prompts were published.
Business fit
Who this helps, and how:
Retail store owners: have drafted review requests go out after pickup, but only after you approve each batch.
Local & serviceStartup founders and e-commerce sellers: ask an assistant "what am I low on?" from your live inventory instead of checking dashboards.
Shopify & ecommerceLocal service businesses: turn a late-night text inquiry into a drafted quote, deposit link and calendar entry for you to approve.
Local & serviceContent creators: add source tracking to every order from day one so you can tell which channel or feature brings in sales.
Creators
Your turn
Ideas to try on your own work. These are starters, not results anyone has measured.
- Which order detail still lives on paper or in a text on your phone?
- What customer email should wait for your yes before it is sent?
- Which part of an order should one system hold so you do not type it twice?
Related
- Evidence grade DA Machine Shop Owner Built an Offline Job Scheduler for 22 Machines With ClaudeLight evidence, using Claude, Claude Code and Git
- Evidence grade BA Small Agency Cut Monthly Client Reports From 45 Minutes to 10 With ClaudeStrong evidence, using Claude, ChatGPT and Canva
- Evidence grade DA Charleston Wedding Planner Cut Timeline Writing From 6–8 Hours to 2 With ClaudeLight evidence, using Claude
- AI use cases for business operationsDepartment hub: AI for operations, 57 examples
Similar tools
Other tools for this job, and how they differ. Chosen by our editors, never paid for.
“Strong evidence” (grade B) means: The person who did the work describes the setup and results in detail. How we grade. Spotted a mistake? Suggest a correction.
