4P Marketing: Four AI Agents Turn a Slack Brief Into a Shopify Preview

TL;DR

• Client: 4P Marketing, a performance and website agency. We build Shopify websites for their clients.
• Input: a Slack message describing how the client wants the website to look: the look, the assets and, later, the corrections.
• Four agents: Intake, Build, Verify and Feedback hand the work to each other through a file, so a job survives after the chat ends.
• A person is the gate. A preview reaches Aidana only after the audit has passed. She opens it, and she sends the link. Nothing in the system posts to Slack on its own.

The problem

Agency feedback is messy. A single Slack thread mixes colour codes, marked-up screenshots, "like on that other site", German and English, and comments that contradict an earlier PDF. A developer has to read it all, work out what's being asked, implement dozens of small changes, check them on mobile and desktop and write up what was done. That's slow, and points get lost. Two jobs on the same store can also overwrite each other.

The four agents

Aidana is the gate in the middle. A preview reaches her only after the audit has passed. She opens it and she sends the link. The agents draft the client message; she sends it. A preview without a passing audit cannot be shared, and every approval is recorded by name.

What it does

• Turns a Slack brief into a Shopify draft without two jobs on the same store overwriting each other.
• Keeps every change on an unpublished theme. The live theme is never where the work lands.
• Runs three full audit passes before a person is asked to look.
• Follows the client clock agreed with 4P: first draft by day 3, two feedback rounds, final version on day 15.

A real round of feedback

For one of the agency's beauty clients, a single round looked like this:

• Three sources merged: the agency's Slack thread, the end client's meeting notes and a 14-page "open points" PDF.
• A full technical review opened every page of the shop: 524 product pages, 130 collection pages and 21 content pages. It found 94 individual points across 10 areas.
• Root causes, not just symptoms: 523 of 524 products had been imported without real variants, so the variant selectors in the mockups couldn't appear. That's a data fix, not a design fix.
• Things the automation couldn't do (Shopify blocks code access to trial themes) were reported clearly instead of being skipped silently.

Under the hood

• Slack as the interface. The agency doesn't learn a new tool.
• A hand-over file between agents. Each request is a file with its status, so work survives restarts and the next agent always knows where to pick up.
• Draft-first, always. Every change lands on a copied draft theme with a preview link.
• Visual proof. Screenshots at mobile and desktop widths, assembled into a before/after PDF per round.

Results

Connected projects

• Same "understand → build → check" loop as → F1 Pit Stop Shopify builds
• Same multi-agent thinking as → Antoinette
• The agency is managed as a client in → Artikuno OS