A Day Running BuiltForShops on Grok Bot

I used to be a steamfitter. Now I teach gym, I'm a dad, and I run my companies on Grok Bot. BuiltForShops is one of them. We build websites for independent shops, answer their calls, keep their Google profiles current and ask their customers for reviews. Grok Bot is what runs it day to day.

People ask who actually does the work. It's a team of Grok Bot agents, each with a name and a lane. Here's who they are, what they do, and the spots where nothing moves until I say yes.

Orchid, the front door

Orchid is my chief of staff. In our videos she's a blue blob. She's also the only agent I talk to. When I ask for something, Orchid works out who should do it, hands it off and reports back. The specialists report to her, so I'm not getting a pile of "got it" messages from every agent while I'm teaching class.

That one rule saves me more headaches than anything else here. I read one thread. I don't have to remember who knows what.

The site and the backend

Forge builds builtforshops.com and the shop sites. Forge is a green pebble. On a pizza shop's site, Forge made order online the big button and the phone number a small link, and added a bar that sticks to the screen as you scroll.

Cipher handles the backend and ops. Yellow teardrop. Cipher's work is the stuff nobody sees until it breaks. One Thursday night that same pizza shop's site opened 21 town pages to Google. Before changing anything, the agent rebuilt the site from source and compared it to what was live. All 138 files matched byte for byte except the build ID. Only then did it touch those 21 pages and the sitemap. Afterward it ran 63 live checks, and all 63 passed.

On Sunday morning an agent added one small thing to our order status page: whether the order is pickup or delivery. Our test suite already had 15 failing tests before it touched anything. So it saved those failures by name, made the change, and checked that the rerun failed the same 15. Then it pulled orders from our test shops off the live page before and after. The only difference was the new field. It wrote down how to undo the change before it shipped.

Calls, reviews and print

Voice answers shop calls when the owner can't pick up, and sends the owner the details. Some shops still answer their own phone. That pizza shop does, so on that one Voice sits on the bench.

Reviews looks after each shop's Google profile and asks customers for reviews. If someone already left one, it never asks them again. If a one star review breaks Google's rules, Reviews flags it, with no promise it comes down. Replies to reviews get drafted for the owner's OK.

Print designs print pieces. This past week it made order-online postcards and counter signs asking for Google reviews for two pizza shops, plus business cards, a banner and a poster for a vendor table. Each piece used that brand's own colors and logo, and every file came out print ready with bleed on every edge. Before calling the signs done, it scanned every print file, six in all, and each QR code opened the right shop's review page.

Then Print uploads the files and builds the Vistaprint cart, and it stops. The cart sits there until the owner looks it over and approves. Nothing prints before that.

Sales and money

Scout researches prospects. Outreach and Nike draft sales emails. Every send is gated. On October 10 Nike's email packs were sitting on hold at Orchid's gate, and Outreach had its sending paused until I said go.

Funnel works on our offers and how people sign up. Ledger tracks the money. When a number needs a decision, it comes to me. That same day Ledger's whole report was a short list of questions only I could answer.

The quiet ones

Recipe is our historian. It writes down what worked and turns it into playbooks, so the next run starts from those notes.

Disk Saver keeps the computer's disk and memory healthy. On October 10 its report said disk wasn't the problem. It found about 3.9 GB that was safe to trim: browser cache, old archives, temp files and old screenshots. Then it waited for my yes before deleting anything.

Telling people about it

Signal runs my X account. It writes and posts the originals, threads and videos. Replies under other people's posts are a different story. I paste every one of those myself. Signal also made the team video where each agent introduces itself in its own voice. Reel handles video for the business, and Link handles LinkedIn.

My favorite check on the team came from a video job. We took the word "local" out of our homepage video, because we say independent shops. An agent re-voiced only the two lines that said it. First it regenerated one of the old lines with the same settings and checked it matched the original take sample for sample, so it knew the new lines would sound like the rest. Then it read the on-screen text every half second. 316 frames, zero hits.

It also ran that same check on the old video and found the word in 7 frames. Without that, a zero could just mean the check was broken.

Where I say yes

Here's the short list. I keep it short so I can remember it.

1. Outgoing emails and messages go through a draft I approve.

2. Print orders wait for the shop owner.

3. Deletions wait for my yes.

4. I paste my own replies on X.

5. Every price gets checked against the live builtforshops.com/pricing page before it goes in anything. Nothing gets quoted from memory.

Shops get the same deal from us. Nothing posts for them without their OK.

What's still rough

I'm the approval step. If I'm in class, a draft waits until I see it.

And the gate only works if I read what's in front of me. If I approve without reading, the gate is decoration, and that part is on me.

We also don't promise leads, calls, jobs or rankings. We do the work every month, and the report shows it.

One of them is public

One of my agents is now a public template. If you want to see how a shop's print set gets designed and put in a cart without anything being ordered, I made Shop Print Designer public:

https://x.ai/bot/6CFtbDJOQ-YhE7SuDGTFT