AI engineering · forward deployed

AI engineering for Shopify Plus stores

Forward deployed engineering for £1m+ stores. We build the Claude agents, MCP servers and Claude Code skills that run our own Shopify Plus business — and we deploy the same stack to clients in 8 to 12 week embedded engagements.

Book a discovery call See what we’ve built ↓

Multi-£m

Shopify Plus stores under our own management

7,000+

variants in our catalogue, all touched — created, edited, repriced, retagged, metafield-tuned

200+

collections with content we wrote, rewrote, and optimised

248

SKU lifecycle decisions in a single day, variant-aware (one example of the work we automate)

Daily

production AI agents across our stores and clients

We don’t sell AI. We run on it.

Most agencies pitching “AI for ecommerce” are bolting GPT wrappers onto retainers they were already selling. We took the other path. We’re operators. We run a multi-£m Shopify Plus store as our own business, and we built a stack of Claude agents, custom MCP servers and Claude Code skills out of necessity — to run that store with fewer hands and better decisions than agencies could deliver.Then clients started asking us to build the same thing for them. So now we do both: we run our store, and we ship the same engineering to other operators in fixed-scope engagements.When you hire us, you’re not hiring a team that talks about AI. You’re hiring the people who built the systems they use every day, and who’ll build yours the same way.

What we’ve built

The list below is six specific systems we’ve shipped and use in production. They sit on top of broader catalogue work — every one of our 7,000+ variants has been touched in some way. Products created from supplier specs and listed properly. Existing products edited, retagged, repriced, metafield-tuned. Meta titles and descriptions rewritten across hundreds of pages. Collection content authored and optimised. Custom Claude Code skills authored in-house for the workflows we run weekly.Six examples below, but the breadth is what matters: this is how we run a multi-£m store with a small team.

Merchandising triage agent

A variant-level Shopify clearance system using Claude Sonnet plus a custom MCP server we wrote against the Shopify Admin GraphQL API. It pulls 90-day sales velocity per variant, applies a rule we landed on after some painful learning — only clearance-tag a product if every variant is slow, and cut only the slow variants if it's a mixed bag — and writes the changes back through GraphQL. In one day it triaged 248 SKU lifecycle decisions. 47 hardware. 201 e-liquid. The same job used to eat about 40 hours a week of merchandiser time.

Catalogue operations engine

Bulk operations across our entire catalogue. We built a Python agent against the Shopify Admin GraphQL API that handles batching, rate limits, error retry, and verification per variant. The first job was a cost-update run across 7,000+ variants — 83% coverage on active products in one shot. Since then it has handled bulk metafield writes for SEO content blocks, bulk meta title and description rewrites across hundreds of pages, redirect imports, price-rule application, attribute normalisation across legacy product data, and product creation from supplier spec sheets.

SEO content sub-agents

Custom Claude Code agents for collection rewrites and product descriptions. Each one is brand-voice tuned, fed structured input from Ahrefs and GSC, and writes output that meets our internal SEO checklist — primary keyword in H1 and first paragraph, semantic variations, FAQ blocks where they fit, internal links to sibling collections. We use them daily across 200+ collections on our own store. The output isn't perfect — nothing AI writes is — but the editing time is roughly 80% less than writing from scratch.

Custom MCP servers

In-house MCP servers connecting Claude to Shopify Admin GraphQL, Ahrefs, GSC, Sanity, and our internal merchandising data store. They live inside our Claude Code workflows and let Claude do real work on real systems, not just talk about it. We're open-sourcing one of these soon.

Claude Code skills library

A library of in-house Claude Code skills covering the workflows we run weekly. Specific skills include collection creation and content upload, Shopify product listing imports (turn pasted supplier data into properly listed products with metafields, variants and SEO populated), weekly merchandiser reporting, clearance triage, content writing for collection and product pages, SEO content optimisation, and competitor analysis. Skills are reusable instruction sets that turn an ambiguous request into a deterministic workflow.

Stone and Sky operations platform

Outside Shopify entirely. We built an end-to-end Next.js plus Sanity plus Vercel platform managing 28 HMO units across 5 properties for a property finance company. Liaison packs, fire and safety logs, evacuation plans, incident reporting. Live in production, used daily by the operations team. Real example of what we ship when the brief is "build the operating system this business runs on".

How forward deployed engineering works for ecommerce

Most agencies sell hours. The forward deployed engineering model is different — and it’s the model we run.A forward deployed engineer (FDE) embeds with a customer, owns a problem end-to-end, and ships production software inside the customer’s environment. The role started at Palantir, scaled at companies like Anthropic, OpenAI and Sierra, and is now becoming the standard way technical AI work gets delivered. It works because most enterprise AI deployment isn’t a model problem — it’s a context problem. You can’t deploy useful AI without deeply understanding the workflows, constraints and edge cases of the business you’re deploying it into.The same logic applies to Shopify Plus. Plugging a generic AI tool into a multi-£m store and expecting it to work is the same mistake enterprises made in 2023. What actually moves the needle is someone who sits inside your business, understands your catalogue, your customers and your operations, and ships the systems that fit your specific reality.That’s what we do. We’re a small consultancy that operates like an internal FDE team — embedded, accountable, shipping working code, not deliverables in PowerPoint.Read the full explainer: what a forward deployed engineer actually does →

The engagement model

We work in 8 to 12 week embedded builds. Fixed scope, fixed price, agreed up front. No retainers.

Weeks 1 to 2

Scope and embed

We get full read access to your Shopify Admin, GA, GSC, Ahrefs, and any internal tooling. We sit in your Slack or Teams. We talk to whoever owns merchandising, customer service, ops. By end of week 2 we've identified the two or three workflows that, if AI took them over, would compound the most over the next year. We agree the scope in writing.

Weeks 3 to 8

Build and ship

Most weeks we ship something. The first thing usually goes live by week 4 — a small agent doing real work on real data. From there we layer in additional workflows, custom MCP servers if needed, and integration into your existing tooling. You see weekly demos, not monthly status reports.

Weeks 9 to 12

Hand off and document

The systems we built are owned by you. We document everything — the architecture, the prompts, the schemas, the runbooks. We train your team to operate them. We agree what ongoing optimisation looks like, if any. We're not interested in retainer dependency. After week 12 you have production AI systems running in your business that you understand and own.

What we deploy

The technical stack we use, every engagement.

Models

Claude Sonnet, Claude Opus, occasionally Haiku for cost-sensitive workflows

Orchestration

Claude Code, custom Python and TypeScript agents, MCP servers

Shopify integration

Admin GraphQL API (deep — variants, metafields, metaobjects, bulk operations, redirects)

Data sources

GSC, Ahrefs, internal databases, your existing analytics

Infra

Vercel for any web surface, GitHub for code, your existing tools where they exist

Open standards where possible. Custom code where it actually matters. Nothing locked behind our toolchain.

Open source

We’re shipping a couple of our internal MCP servers and Claude Code skills as public repositories so anyone — clients or not — can see how we work. Links go here when the repos are live.Coming soon: hollowpoint/shopify-mcp-server and hollowpoint/claude-skills-shopify.

Frequently asked questions

Is this an SEO service or a software project?

It's a software project. We build production AI systems — agents, MCP servers, automations — that solve specific operational problems in your business. We do offer SEO services separately, but they're a different engagement. AI engineering is engineering work, with engineering deliverables.

How is this different from hiring an AI agency?

Most AI agencies sell ongoing retainers and don't build software you own. We sell fixed-scope engagements and hand off working code that runs in your environment. We can do this because the same stack runs our own store every day — we're not learning on your project.

Do we own what you build?

Yes. Code, prompts, schemas, MCP server configurations — all of it. We document everything and train your team. We can offer an optional ongoing optimisation arrangement, but it's never a lock-in.

What kind of stores do you work with?

Shopify Plus stores doing £1m+ a year, primarily UK and EU based. We've worked across consumer retail, wellness, and B2B-flavoured ecommerce. The technical stack matters less than the operational complexity — we work best with stores that have real merchandising, real customer service load, and real catalogue depth.

Can you work alongside our existing tech team?

Often the better arrangement, actually. Your team knows your business; we know how to ship Claude-based systems quickly. Embedded engagements work well as a sprint that augments an internal team rather than replacing one.

What does it cost?

Engagements typically start at £8k to £15k per month for the embedded period, sometimes higher depending on scope. We'd rather have an honest conversation about budget and scope than publish a generic price. Book a call.

How fast can we start?

We typically have one engagement slot opening every 6 to 8 weeks. If we're a fit and the scope is clear, we can usually start within 4 weeks of the first conversation.

Book a discovery call

Tell us about your store and where you think AI engineering would help. We’ll do a 30-minute call, no pitch, and either tell you yes we can help and here’s what an engagement looks like, or no this isn’t a fit and here’s what we’d do instead.Talk to us