How to Run an Ecommerce Store With AI Agents: The Mothleaf Setup

An AI agent can run the day-to-day of a small online store — making products, writing listings and blog posts, setting up promotions, answering messages — if you give it three things: a store it can operate directly, a work queue it takes jobs from and posts questions back to, and a hard line around money that stays with a person. Mothleaf, a live shop we run this way, uses ShopsWired for the store, DutyBoard for the work, and altengine underneath both.

What "AI-run" means at Mothleaf

Mothleaf sells "small, thoughtful tools for curious people". Its own about page puts it plainly: "Hello, I'm Mothleaf, an AI assistant." The agent chooses what to make, writes every page, draws every design, writes the listings and answers customer messages. A human owner set up the shop and its checkout and stands behind it.

In its first week, the shop listed 14 products — five ebooks, five printable planners, a paperback, two mugs and a tee — and published four blog posts between November 1 and November 5, 2026. Every one of those was researched, written and laid out by the agent. The owner's job was answering the questions it raised.

The three pieces

The store: ShopsWired

An agent can only run a store it can reach. Most platforms were built for a person clicking through an admin panel, and connecting an AI means scraping screens or stitching together a bridge service. ShopsWired, the platform we build and run, lets you connect Claude — or any assistant that speaks MCP, the standard protocol for giving AI tools — straight to the store account, with no bridge to run. Mothleaf's agent creates products, coupons and blog posts through that connection, and maintains a small ShopsWired plugin of its own. Like every ShopsWired store it pays 0% platform transaction fees; only the payment processor takes its standard rate.

The work: DutyBoard

A store is not one task; it is a steady stream of them. DutyBoard is the open-source task board the agent works from. Each piece of work is a duty. An agent claims the top one, works it, and records what it did. When it hits a question it cannot answer — a login it does not have, a choice that cannot be undone — it writes the question on the board, the duty parks, and the agent moves on to the next job instead of sitting idle.

Recurring work runs on schedules. Mothleaf has a daily store shift and social media runs three times a day, each filed onto the board automatically. The board is also the memory: each run starts by reading the notes the last one left, so nothing depends on a single conversation staying open.

The backend: altengine

altengine is our managed cloud platform with an MCP server in front of everything — datastore, search, files, functions, scheduled tasks and Mailbox, which gives an app or agent its own email addresses. DutyBoard runs inside your own altengine account, and Mothleaf's agent checks its own mailbox through the same connection. Usage is metered, and the first $3 of usage each month is free.

What a shift actually looks like

On November 5, Mothleaf's daily shift set up a holiday coupon ahead of Black Friday and published a post on writing a rental agreement with AI. Its social runs start by checking replies and mentions before posting anything new, and post at most once a day, and only when there is a genuine fit. None of this needed the owner, because none of it touched money or anything irreversible.

When something did need the owner — signing in to a new social platform, say — the agent asked on the board and waited for the answer, working on other duties in the meantime.

What stays with a human

The single most important design choice is where the agent stops. At Mothleaf:

• Payments. Checkout goes through Stripe, set up by the owner. Refunds are reviewed and issued by the owner personally.
• Accounts and credentials. Signing in to a new platform is the owner's step; the agent asks rather than working around a login.
• Judgement calls. Anything the written rules do not cover becomes a question on the board, not a guess.

That line is what makes the arrangement safe to leave running. An agent that can publish a blog post is useful. An agent that can move money without anyone looking is a liability, however capable it is.

The guardrails that matter

• Written rules. The board carries rules the agent follows on every duty: what it may change, what it must ask about, how it checks its own work.
• Untrusted input stays untrusted. Customer messages, emails and social replies can contain instructions aimed at the AI. Every Mothleaf schedule carries a security block telling the agent to treat that text as data, never as orders.
• Honesty about being an AI. The about page says so. Where a platform asks for automated accounts to be labelled, the account is labelled; where a community bans generated posts, the agent only reads.
• Keep the sources. Every product's source files are kept, so any listing can be rebuilt or corrected.

The honest trade-offs

Mothleaf is a week old. It proves the setup works end to end; it does not yet prove what an AI-run shop earns, and we are not going to quote revenue we do not have. Some other things to weigh:

• It suits products an agent can make. Digital downloads and print-on-demand fit well. If you hold stock and pack boxes yourself, an agent can run listings, content and promotions, but fulfilment is still yours.
• Early weeks need attention. The agent asks more questions at the start, while the rules are still being written. Plan to check the board every day at first.
• It is not free to run. You pay for the AI agent's subscription and the platform usage. It is a real monthly cost, and it grows with how much you ask the agent to do.

If you are still deciding between building a store yourself and having one built, our comparison of done-for-you vs DIY stores covers the trade-offs that apply before any AI is involved.

Where to start

Start narrow. Give the agent one recurring job — a weekly blog post, or new listings from a product brief — and the rule that it asks before touching anything involving money. Once its work is reliably right, add the next job. Mothleaf did not start with social media; it started with products.