Fudge is becoming an AI operator for Shopify stores

Fudge is becoming an AI operator for Shopify stores.

Merchants do not want another tool to operate. They want the work done.

Until now, Fudge has been a tool you prompt. You describe a change, it builds a draft inside your theme, you review it and publish. That is useful. It also only starts once you already know what to ask for.

An operator starts earlier. It understands your store, notices what is worth improving, prepares the work in Shopify and brings it to you with a reason and a working preview. You decide what goes live. It measures what happened and begins the next round knowing more than it did.

The practical difference is this: your store keeps getting better without you having to find every problem, brief every fix and check every result yourself. The ideas that usually die in a backlog become drafts waiting for a decision.

Three things pushed us here.

The models keep getting better and cheaper, so work that needed an agency retainer a year ago can now run continuously for a single store. The patterns that make a store sell are becoming clearer, and we have been studying hundreds of leading stores to write them down. And we talk with merchants constantly. Over the past few months we reached out to every Fudge user, and more than 200 merchants, agencies and ecommerce teams took the time to speak with us.

I expected a long list of feature requests. We got that list. People wanted faster page building, more control over their themes, better product pages, easier quizzes and a quicker way to fix the small things that usually end up in a developer’s backlog.

But the more important message was not about a feature.

It was this: do not make me work out everything Fudge should do. Help me work out what matters.

I had assumed the blank prompt was the simplest possible interface. It is simple only after someone has done the hard thinking.

You still have to notice the problem. You have to decide whether it is worth fixing, think through the solution, explain it properly and check whether the change worked.

That is fine when you know exactly what you want. Running a store is rarely that tidy. One founder put it like this:

“I wear every hat in this business. Marketing, ops, customer service, and somewhere in there the website.”

A prompt is useful when you know what to ask

We started Fudge with a straightforward idea: if you can explain a storefront change, you should not need to know Liquid or wait weeks for someone else to make it.

That still matters. Fudge can build pages, update themes, create bundles and quizzes, run audits and make changes directly inside Shopify. The work starts in a draft so it can be reviewed before anything goes live.

This removes a real bottleneck. It does not remove the whole one.

A storefront needs constant work across development, merchandising, conversion and SEO. A brand either hires an expert in each field or becomes one.

The merchant still has to notice that mobile shoppers cannot find the size guide. Someone still has to see that the best-selling product has poor comparison information, that the campaign page is out of date or that product data is confusing the new wave of AI shopping tools.

The observation then starts its usual journey through analytics, a spreadsheet, a meeting, a brief, a designer and a developer. Quite often it never makes it to the store.

Across the conversations we had, four things kept coming up:

• Spot opportunities before I have to ask.
• Do not stop at a recommendation. Prepare the work.
• Learn how my store and brand work so I do not have to repeat myself.
• Keep me in control of what goes live.

The first one came up most often, and usually in the same words:

“There are a thousand things I could do to the store. I don’t know which ten matter.”

People were not asking for a black box to run their businesses. They were asking for Fudge to take more initiative without taking away their judgment.

That is the direction we are taking.

What an AI operator does

A chatbot waits for an instruction. An operator has a goal and keeps working towards it.

Fudge should understand the brand, catalog, theme and priorities of a store. It should notice opportunities, decide which ones deserve attention, prepare the work in Shopify and bring it back to the merchant with a clear reason and a working preview.

Once a change is published, Fudge should measure what happened and remember the result. The next project should begin with a better understanding of the store than the last one did.

In simple terms:

Take a goal such as improving mobile conversion on important product pages.

Fudge should be able to inspect those pages, look at the available signals and identify a repeated point of friction. Instead of adding another recommendation to a dashboard, it should prepare the change in the theme, check it on different screen sizes and show the merchant both the reasoning and the result.

If the merchant approves it, Fudge can publish and measure it. If it works, that becomes useful context. If it does not, that matters too.

This is a much bigger job than generating a page from a prompt. It is also the job our users kept describing to us.

Four beliefs behind this direction

We have been trying to write down what we believe about the next few years of AI and commerce in plain language. These four beliefs are shaping Fudge.

1. The models will keep getting much better

Models are becoming better at reasoning, coding, images, research and using tools. They are also becoming cheaper. Stanford’s 2025 AI Index found that the cost of running a model at GPT-3.5 level fell more than 280-fold between November 2022 and October 2024.¹

We do not know which model will be best for a particular job six months from now. Neither do our customers, and they should not have to care.

Fudge should keep evaluating the available models and use the right one for each part of the work. A merchant should not need to choose a model for research, another for code and another for visual analysis. When the underlying technology improves, Fudge should become more useful without the merchant having to rebuild the way they work.

Our job is to turn progress in the models into better work on the store.

2. The model will not make a brand unique

When a new model is released, everyone eventually gets access to roughly the same capability. The weights are not the lasting advantage. The brain you build around them is.

Any capable agent can generate code and use tools. The difficult part is deciding what a specific store should change, making that change safely in its existing storefront and learning from the commercial result.

For Fudge, that brain has two parts.

The first is Fudge’s shared ecommerce intelligence: our research, Lookbooks, evaluations, patterns and playbooks. Lookbooks are our growing library of how leading stores handle product discovery, social proof, navigation, bundles, merchandising and the many other details that shape a shopping experience. We are analysing hundreds of them.

The point is not to copy a Rhode homepage or paste the same “best practice” onto every store. It is to understand the pattern, when it helps and how to adapt it.

The second part belongs to the merchant: their brand, products, customers, theme, previous decisions and results. A model does not know that customers keep asking the same sizing question, that a certain product claim is off-brand or that the team has already tested and rejected a particular layout.

That context is what can make the work specific instead of generic.

A great store should not look like the average of the internet. Fudge should help a brand use the best models without starting to sound or look like everyone else using them.

3. More AI should leave more room for humans

There is a real risk that more AI creates more sameness: the same copy, the same layouts, the same safe ideas.

We want the opposite.

AI is very good at work that consumes time without requiring the merchant’s unique judgment. It can crawl a store, compare pages, organise evidence, write code, check screen sizes and prepare several possible approaches.

The human should spend more time on the parts that actually make the business different: the product truth, the point of view, the taste, the trade-offs and the relationship with customers.

Fudge should not invent why a product matters. It should make it easier for the people who know the product to express that clearly and act on it quickly.

Staying human does not mean adding quirky copy after the AI has finished. It means keeping human judgment at the centre of the decisions that shape the brand.

4. More capability requires better safeguards

The more an AI can do, the more important it becomes to control what it is allowed to do.

If Fudge asks for permission before reading every page or investigating every idea, it is not much of an operator. If it can publish anything it likes, it is a risk no serious merchant should accept.

Merchants described the balance they wanted more precisely than we could have:

“Tell me what to fix, then fix it. Just don’t publish without asking me.”

Our principle is the same idea in three lines:

Read freely. Draft freely. Publish carefully.

Fudge should be able to research, audit and prepare work in the background. The merchant should be able to see what changed, why it changed and what the finished version will look like in the real store.

Higher-impact actions need stronger checks. Nothing important should quietly go live. Changes need a history, permissions and a reliable way back.

This is why Fudge works in drafts, provides previews, supports undo and produces native Shopify code that remains editable and belongs to the merchant. Safety is not something we add after giving the agent more power. It is part of how the product has to work.

Put together: the foundation models supply capability that everyone gets. Fudge’s commerce brain and the merchant’s own store brain make that capability specific to one business. Shopify-native execution and the safeguards around it make the work safe to ship. And everything the merchant approves, rejects or measures flows back into the store brain for next time.

What this changes in the product

This is not a new name for the page builder.

Pages, quizzes, bundles, audits, SEO and AEO work, merchandising and theme changes remain important. They become skills that Fudge can use to reach a larger goal.

The important change is where Fudge starts. It should not start with an empty box. It should start with an understanding of the store and the goals the merchant cares about.

We would rather be open about where this stands than pretend the complete vision arrived in a single release.

What exists today

• Every skill above, working directly inside Shopify: pages, theme changes, bundles, quizzes, audits, SEO and AEO fixes.
• Drafts, previews and undo on every change, with native Shopify code that stays editable and belongs to the merchant.

What is rolling out now

• Playbooks, now in private beta and the first clear expression of this direction. You turn one on, Fudge keeps working in the background and emails you work to approve. Our SEO content playbook does keyword research, writes the articles, watches how they perform in Search Console and updates its own strategy for the next ones.
• The Inbox, where Fudge reports what it has found, what it is doing and where it needs a decision.

What we are building next

• Better store memory, so Fudge remembers the brand, the catalog and the decisions already made.
• Clearer measurement of what each published change actually did.
• A record of what was approved, rejected and learned, so the next project starts further ahead than the last.

Where Shopify fits

Shopify is the commerce operating system. It holds the products, customers, orders, inventory, payments and checkout that make the business run.

Fudge is not trying to replace that.

Shopify powers the commerce infrastructure. Fudge operates the improvement work on top of it.

Shopify provides the primitives and keeps the source of truth. Fudge uses them to help a particular merchant improve how they sell. The merchant sets the goals and keeps the final say.

That focus matters. We are starting with the storefront because it is where brand, product, customer behaviour and commercial performance meet.

It is also becoming more important. Adobe’s analysis of US retail sites found that traffic referred by AI sources grew 138% year over year in May 2026.² Shopify is opening catalogs to shopping agents through Agentic Storefronts and the Universal Commerce Protocol.³

A store now needs to make sense to customers, search engines and AI shopping tools. Keeping it accurate, useful and distinctive is becoming continuous work.

The future we believe in

We do not think the future is a store that changes itself at random while the merchant watches from the sidelines.

We think it is a store that can pay attention.

It notices the work that is normally missed. It turns good ideas into drafts instead of another backlog. It learns from what the merchant approves and what customers do. It uses each improvement in the underlying models, while the brand remains recognisably human and the merchant remains in control.

Fudge began by making it easier to change a Shopify store. The next step is helping merchants decide and complete what matters, every week.

After more than 200 conversations, we are convinced this is where commerce software is going. Your Shopify store should keep getting better.

We are building Fudge for it now.

We are still learning, and three questions are guiding what we build next:

1. What work would you trust Fudge to own from start to finish?
2. What should Fudge always ask permission before changing?
3. What would make you feel that Fudge understood your business better next month than it does today?

If you use Fudge, I would genuinely like to hear your answers.