Back to Blog
7 min read
E-commerceAI

Claude Builds My Store Website, Hostinger Deploys It: My 2026 System

Last updated: Sep 23, 2026

Short answer: I build store websites with two agents and never touch the code. The first agent, Hermes, runs on a server 24 hours a day (05:16). It looks for products that are starting to climb, picks the five biggest competitors (05:49) and writes their websites into a file for a second agent (04:28). The second agent is Claude. It takes that file, builds the site and deploys it to Hostinger as a static site through a connector (09:41). I show the whole process live in the video „AI zarobiło mi 84 000 zł w 3 dni. Sklep prowadzi się sam” (Polish, "AI made me 84,000 PLN in 3 days. The store runs itself").

Stuck on a prompt? Post it on my free Discord server and ask. Join the Discord →

This post is the written version of the video, with timecodes to the exact moments and clear marks where I don't do the math on camera.

Disclosure: the video is tagged #reklama (Polish for "ad") and was made in cooperation with Hostinger. The link hostinger.com/dawidhc and the code DawidHC come from the video description. The hostinger.com/dawidhc link is an affiliate link: if you buy through it, I earn a commission. That doesn't change anything in this post. In the video I say the code gives 10% off (02:58) and that there is a 30-day money-back guarantee (03:01). I don't give the plan price in the video, so I don't give it here either.

About the number in the title: the 84,000 PLN figure is not said in the video itself, and I don't break it down there. This post doesn't expand on it. This is not a promise of results. Outcomes depend on product, budget, execution, market and experience.

How an AI agent researches competitors before building the site

Hermes does the research. It lives on a Hostinger VPS and works around the clock (05:13). I covered setting it up in the post on running your own AI agent on a VPS.

Every day it scans Google Trends, TikTok, X and other social platforms (05:27). It looks for a product that is starting to rise. When it finds one, it looks for competitors in that niche (05:42).

It works at night. In the morning I log into the agent and get a report on what it did (03:59). The example in the video is stores selling coffee products (04:12). The report describes their websites: what they look like and what they do.

I barely read that report (04:21). It's an MD file one agent writes for another agent (04:28). A human isn't needed in that chain.

How to pick the competitor whose site you model

I pick competitors through the Facebook Ad Library. The agent is set to find the five biggest (05:49). The filter is the number of ads still active after 30 days (05:56). Whoever runs the most makes the list.

Then the agent goes through their websites and records every detail (06:05). That description becomes the build instructions for the second agent.

I'll say plainly what I change: the names. The agent gives me a few names to choose from, and the rest is a copy (06:19). I've done it this way from the start, and that hasn't changed (06:25).

Another agent researches the ads: how they look, how they sell, which hook works best (06:27). That step is already automated, but I don't show it in this video. I wrote up the ad strategy itself in Meta Ads 2026: my strategy step by step.

What prompt to give Claude so it builds the store website

I give the prompt in Polish and English (06:48). Here is its skeleton:

Part of the prompt What I tell the agent Timecode
Source use the Hermes research, pick the best of the five coffee stores 06:50
Task copy and build their site on Hostinger using the MD file 07:00
Design use the Emil Kowalski design skill and all the animations in it 07:21
Questions if you have design questions, always ask me 07:37
Bar make it professional, it can beat the best competitor 07:50
Animations realistic coffee bags, items scroll with the page 08:03
Naming make up the store and product names yourself, research the name 08:23

The design skills I use are called Emil Kowalski Skills (07:15). It's a set of instructions for building websites. I use it a lot because I like what it produces.

I share my own research and site-building skills (04:42) on Discord. The link is in the video description: discord.gg/jnPNUk9qZD.

How to connect Claude to Hostinger and deploy the site

I connect Hostinger to Claude with a connector (02:03). Claude builds the site, and deployment is one command. I type: deploy project coffee to Hostinger as a static site (09:41). "Coffee" is the project name.

In the video I mention what comes in the package: a free domain for a year (02:29) and a daily backup (02:31). I mainly use Hostinger as a VPS, but also for building sites (02:19).

How long an AI site build takes and how much I fix

A build takes half an hour to an hour, depending on the site (08:42).

I fix very little. With older Opus models I made changes; now I barely have to (08:55). The newest model builds even list pages and quizzes well (09:04), so I don't explain as much as I used to.

The result in the video is the homepage of a coffee store (09:52). The agent based it on the five stores from the research (09:59). The colors fit coffee and it looks professional. That is my on-screen judgment, not a conversion test. I don't measure this page's conversion in the video.

Why build separate pages on Hostinger when the store is on Shopify

The main store is on Shopify, and that's where everyone buys the product (03:15). For that one product I build several different offers (03:24). I run three or four pages per store, one per offer (02:43).

The order: Shopify store first, then the offer pages on Hostinger (06:42).

Offer page type Purpose Timecode
product page the main offer 03:31
quiz walks the buyer to the product 03:35
list page a different angle on the same product 03:37
story page e.g. where the coffee comes from, product link at the bottom 11:00

I only build new offers once the store is selling (10:28). Facebook sets the direction: the analytics show which customer it picked, and I build offers and prelanders for that customer (10:41). For the page's role itself, see the glossary entry on landing pages.

I don't show in the video how much such an offer adds to sales.

The contest: design my app

At the end of the video I announce a contest (11:24). I built an app, a habit tracker. The task is to design the look of every page and every option in it (11:44).

The prize is longer-term work with me on app design (12:03). I build the apps; I'm looking for someone who designs them. You can design it yourself or with AI's help (12:27). The requirements and the place to submit are on Discord (12:52).

What's next in the series

This is the first video in a series on automating dropshipping with AI (00:57). Upcoming episodes cover, among other things, automating customer service (01:11).

FAQ

Do I need to code for Claude to build my site?

No. I don't touch code in the video. I give the agent the research and a prompt, and I deploy with one command (09:41). You need to describe what you want and judge the result.

Does the agent copy a competitor's site 1:1?

In my setup it copies the structure and description from the research, and I change the names (06:19). In the prompt I also say the site can be better than the best competitor (07:56).

What does this setup cost?

I don't calculate costs in the video. I only mention 10% off with the code (02:58), the 30-day guarantee (03:01), a free domain for a year (02:29) and a daily backup (02:31). The current price is on Hostinger's site.

Want more practical material?

Tick the consent box in the form and I will send you more playbooks on e-commerce, AI, and scaling a store.

I only send emails if you tick the consent box above and confirm it in the email that follows. The data controller is Dawid Gac. Privacy policy

About the Author

DG

Dawid Gac

E-commerce Educator & Entrepreneur

Dawid Gac is a Polish entrepreneur, e-commerce educator, and co-founder of EcomBrain. His public work includes YouTube education, community resources, and a documented 1:1 mentoring method where feedback is attributed to the named coach.