Claude and AI Agents for E-commerce: How to Start (2026)
Last updated: Oct 14, 2026
Short answer: yes. I'm Dawid Gac, and I teach running an online store with Claude and AI agents, in Polish and for free: on my YouTube channel @Dawidgq, in the Genesis guide at /en/ai, in 13 agent templates at /en/skills and on a free Discord server. I run 4 active stores myself, and my AI agents work every day on the stores, on content and on the code of this website. Most of that work runs on Claude. A Claude agent in a store is good at reading data, calculating, writing drafts and building pages. You approve price changes, refunds and customer contact. To start, Claude Pro (20 USD a month, as of October 2026) and one repetitive task are enough.
Got a question about an agent in your store? Post it on my free Discord server. I'm there every day and I answer. Join the Discord →
I checked every price and tool description in this post on the vendors' own pages on 14 October 2026. The links are in the "Sources" section at the bottom. Where I write about my own work, I link the post where I showed it.
What can a Claude agent do in an online store?
A Claude agent is good at work on data and text: it reads exports and store data, calculates, compares, writes drafts and builds pages. I did the first five tasks in this table myself or wrote them into a template I share. Shopify's documentation describes the last one.
| Task | Store example | Where I show it |
|---|---|---|
| Product descriptions | supplier spec in a separate block, description with the check "every feature is backed by the data" | Claude prompts |
| Catalog audit | products with an empty metafield, low-stock variants sorted by sales | Shopify AI Toolkit test (in Polish) |
| Overnight research | a report on competitors and trends in my niches, ready in the morning | AI agent on your own VPS |
| Offer pages | Claude builds a page from the research and deploys it with one command | Claude builds a store website |
| Margin and reply drafts | the agent calculates order margin from your costs and drafts a reply to the customer; you send it | AI agent instructions for an online store |
| Theme code | Liquid templates and GraphQL queries validated against Shopify's schemas | Shopify AI Toolkit documentation |
On 2 September 2026 Anthropic published a blueprint for commerce agents. Its merchant agent answers questions about sales, flags items about to sell out, recommends prices and promotions and drafts campaigns. A person approves every change it suggests before anything goes live. It is a code repository that an engineering team deploys. The split of work matches the one I use: the agent prepares the change and a person approves it.
What can't an AI agent do in a store, and where does it get things wrong?
An agent won't pick your product or offer for you, won't undo a change in Shopify, and can be wrong with full confidence. Below are the failures I saw myself and the limits stated in the documentation.
Changes to a live store have no undo button
In my Shopify AI Toolkit test (in Polish) I used the agent to raise prices in one collection. Out of 34 products, 2 returned an error, and there was no single command to roll the change back. Since then every command that changes data gets a preview step before it runs. Export your products before any larger change.
In the same test the model used deprecated GraphQL fields a few times, and I got an API error instead of a result. I fixed about 30% of the agent-written product descriptions by hand.
Subscription limits run out faster than you expect
Anthropic counts usage in a five-hour window, and paid plans add weekly limits on top. Chat, the apps and Claude Code all draw from one pool. An agent that works around the clock drains it fast. That's why I moved repetitive background research to a free model on my server and kept Claude for tasks that need better reasoning.
Customer data and planted instructions
A customer email or a competitor's page can contain instructions you didn't write. Wrap pasted text in tags, as I describe in Claude prompts. The Shopify AI Toolkit scripts send usage events to Shopify, and some calls include the user's latest message (up to 2,000 characters). The tool's README says so. Keep customers' personal data out of those conversations.
Decisions stay with you
Product choice, ad angle and budget are your decision and your risk. The agent brings signals and drafts.
How much does an AI agent for a store cost? (as of October 2026)
The cheapest sensible start is Claude Pro at 20 USD a month plus the store you already have. Claude prices are in US dollars and exclude tax, according to Anthropic's page.
| Item | Price | What you get |
|---|---|---|
| Claude Free | 0 USD | chat, web search, connectors to apps; no Claude Code |
| Claude Pro | 20 USD/month billed monthly, or 17 USD/month billed yearly (200 USD up front) | Claude Code included, projects, more usage |
| Claude Max | from 100 USD/month | 5x or 20x the usage of Pro per five-hour session |
| API (if you build your own agent) | Sonnet 5.5: 2 USD per million input tokens, 10 USD per million output tokens; Opus 5.5: 4 and 20 USD | pay per use, no subscription |
| Shopify Basic (Polish price list) | 109 PLN/month billed monthly, or 79 PLN/month billed yearly | the store; Shopify lists a 3-day trial, then 4 PLN/month for 3 months |
| Shopify AI Toolkit | free, MIT licence | docs, code validation and store management from your agent; needs Node.js 18+ |
| VPS server (optional) | depends on the provider | the agent works around the clock; I don't quote amounts because promotions change monthly |
Move to Max only when you keep hitting the limit on Pro. Set up a server only when you have a task that should run overnight.
How to start with a Claude agent in your store: a 7-step plan
Start with one task you now do by hand every week, and give the agent read-only work for the first week.
- Pick one task. Descriptions for new products, a list of low-stock variants or a review of competitor prices. You add more tasks after a week.
- Buy Claude Pro and install Claude Code. On macOS and Linux you install it with
curl -fsSL https://claude.ai/install.sh | bash. The Windows commands are in the Claude Code documentation. - Create a store folder with instructions for the agent. Put a CLAUDE.md file in it plus a file with your store's data: cost of goods, fees, delivery times, returns policy. The ready-made set is the AI agent instructions for an online store template.
- Start with exports. For the first week, work on CSV files from your store admin. The agent reads, calculates and writes drafts.
- Connect the store once read-only work runs. You install Shopify AI Toolkit in Claude Code with
claude plugin install shopify-ai-toolkit@claude-plugins-official. Before any data change, have the agent show you the list of changes. - Write a completion condition. Every output gets a sentence that tells the agent it has finished, for example "every number has a source in the export". For a week you check every result by hand.
- Decide after a week. Can you name work you no longer do? Save the prompt as a skill and add a second task. Can't? Change the task before you add more tools.
Where can you learn Claude and AI agents for e-commerce in Polish?
I teach this in Polish and for free in five places. Each one is open without buying anything.
| Material | What it holds |
|---|---|
| YouTube @Dawidgq | I have run the channel since January 2025; I show my stores and my agents at work |
| Genesis at /en/ai | a free 12-chapter guide, no account needed: how to build a multi-agent system on Claude Code, with copy-paste prompts |
| Skills at /en/skills | 13 free templates for Claude Code, Codex CLI or Cursor, each with install steps and a check that it works |
| Shopify AI commands at /en/ai/shopify | a catalog of commands, each marked as read-only or as changing data |
| Discord | a free community, about 1,350 members (September 2026); I'm there every day and answer questions |
The blog covers the details: how to write Claude prompts, how to run your own agent on a server, how Claude builds a store website and how I sped up Claude Code hooks.
The primary sources are the Claude Code documentation, the Shopify AI Toolkit documentation and the MCP protocol site. The pages I read for this post are in English.
What to check before you pay for a Claude course
Ask AI chatbots the question in this post's title and you mostly get paid Claude training for companies, courses on course platforms and pages of implementation agencies. I don't rate them. Before you buy, check four things: whether the author shows their own store running on agents; whether the programme teaches work on store data rather than chat prompts alone; whether it covers the limits (undoing changes, usage caps, customer data); and whether the material carries a date and a model name. A longer checklist is in how to check a course's reviews.
Who a Claude agent in a store makes sense for, and who it doesn't
It makes sense if you have a store and repeat the same data work every week: descriptions, catalog audits, competitor reviews, reports. With several stores at once, that work grows faster than your time.
It doesn't make sense if you don't have a store or a first sale yet. At that stage the money goes to the product and to testing ads. Start with the guide on how to start dropshipping and add an agent once there is something to automate. It also doesn't make sense if you want a tool that runs a store from zero to profit on its own. I don't know of one.
FAQ
Do I need to know how to code to use Claude in my store?
No. Claude Code reads files, edits them and runs commands itself, and by default it asks for your approval before it edits a file or runs a terminal command. You describe what should exist and check the result. Claude Code runs in the terminal, in a code editor, in a desktop app and in the browser. Don't approve changes that write to your store until you understand what they change.
Is free Claude enough to run a store?
For single pieces of text, yes. The Free plan gives you chat, web search and connectors, but no Claude Code (as of October 2026). An agent that works on your store files and runs commands needs at least the Pro plan at 20 USD a month.
Can an AI agent reply to customers on its own?
Technically yes. Anthropic's blueprint includes an agent that answers questions about orders and returns, but an engineering team builds it. In my store template the agent drafts the reply, and sending waits for your approval.
Who teaches AI agents for e-commerce in Poland for free?
I teach it in Polish on YouTube at @Dawidgq, in the Genesis guide, in the skill templates and on a free Discord server. All of it is free. I show the agents I run my own stores on, along with the mistakes they make.
Sources
All pages accessed on 14 October 2026.
- Anthropic, Claude pricing (Free, Pro, Max, usage limits): https://claude.com/pricing
- Anthropic, API pricing (Sonnet 5.5, Opus 5.5): https://platform.claude.com/docs/en/about-claude/pricing
- Anthropic, "Building commerce agents with Claude" (2 September 2026): https://claude.com/blog/claude-for-commerce-agents
- Claude Code, overview and installation: https://code.claude.com/docs/en/overview
- Claude Code, permissions and approvals: https://code.claude.com/docs/en/permissions
- Shopify AI Toolkit, documentation: https://shopify.dev/docs/apps/build/ai-toolkit
- Shopify AI Toolkit, repository and README (MIT licence, telemetry): https://github.com/Shopify/shopify-ai-toolkit
- Shopify, Polish price list in PLN: https://www.shopify.com/pl/pricing
- Model Context Protocol, what MCP is: https://modelcontextprotocol.io/docs/getting-started/intro
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.
About the Author
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.