I Built a Dog Gift Store and Handed It to My Agents

I built a dog gift store, then handed it to my agents

So here is the thing I am testing right now: can a small set of AI agents take a brand new store from zero and actually grow it, with me only making decisions. The store is Doggiftery — breed-specific dog gifts, the mugs and ornaments and blankets, the dog-mom and dog-dad stuff, memorial pieces. It is a real store and a test bed at the same time.

I picked it on purpose. It is internal, so there is zero risk to any client while I push the limits. If the agents do something dumb, the only person who pays for it is me.

Building it was the easy part, and it cost under $5

Here is the part that still gets me: I built basically the entire store with AI, and the whole thing cost under $5 to set up. Not five thousand. Five dollars.

Every piece of it ran through AI. Around 900 products imported and cleaned up: supplier branding stripped off every single image, the white backgrounds recolored to a soft cream so the catalog looks like one real brand instead of a dropship pile, and alt text written for all of them. Product descriptions written. 36 articles covering every breed and gift type, each with an AI-generated hero image, a real product strip, an FAQ, SEO titles and meta, schema, and internal links tying them together. The image editing ran in code for free. The content ran through the pipeline.

Add it up and the API spend was around $4. The Shopify plan is $1/mo on the intro offer. So a real, fully stocked, content-rich store for under $5 all in. A few years ago this is a month of work and a real chunk of change. Now it is a few days and pocket money. Still wild to me.

This is the same pattern I wrote about in automating ecommerce at scale. The build is no longer the bottleneck.

The hard part is ranking a brand new domain

Here is what nobody tells you when you spin up a fresh store: building it is maybe 10 percent of the job. The other 90 percent is getting Google to trust a domain that is a few weeks old with no history and no links.

The crawl is actually moving. Impressions went from 15 to over 360 in about a week as pages got indexed. But every one of those is sitting at position 50-something with zero clicks. That is exactly what a new site looks like before authority lands. And authority means links, which a brand new site has none of.

What the agents do every day

Right now an agent checks Search Console once a day and ships one small, reversible fix. Last week it caught that a poodle gifts page had climbed to position 11, found the internal link that was missing, and added it. Another day it fixed 18 broken internal links across the money pages. Small stuff, but it is the kind of grind a human forgets to do, and it compounds.

What it does not do yet is the thing that matters most: build links and get the store featured somewhere real. That part had been sitting in a queue because nothing was driving it.

Where it is stuck, and what I am building to fix it

This is the honest part. The make-or-break for Doggiftery is foundational links, and they were overdue for more than a week because the system was good at the small daily maintenance and bad at pushing the bigger plays. The agents kept reporting "nothing new" while the single most important task sat untouched.

So the thing I am building now is a layer that reasons about the goal instead of just running a checklist. It looks at where the project actually is, finds the move that matters most right now, and puts it in front of me to approve instead of letting it rot. The first real test was Doggiftery, and the first thing it surfaced was "go get links, that is the whole game right now." Exactly right.

The bigger experiment

Doggiftery is one store, but it is really a question I am trying to answer for the whole portfolio I am running: how much of growing a site can agents own, and where does a human still have to decide. The answer so far is that they can build and maintain a lot, and the judgment calls — what to prioritize, when to spend, what is worth doing — still come back to me. Which is fine. That is the split I want. Tools that do the work, humans that decide.

I will keep posting what works and what breaks. If a pack of agents can take a dog gift store from zero to real traffic, that playbook copies to everything else I run.