For most of 2025, the AI industry felt like it was strapped to a runaway train of progress that only knew one speed: faster!

And then… August arrived.

August was the month someone slammed on the emergency brakes and threatened to derail AI’s entire hype train. It began with OpenAI’s underwhelming new model rollout, eloquently summed up by one trending Reddit thread as “GPT5 is horrible.” From there, the damning headlines shifted to the hysteria unleashed by MIT’s study claiming that 95% of internal AI projects were failing. Days later, the press cycle moved on to Meta’s sudden AI hiring freeze, and whispers of an “AI bubble” began to circulate.

Misery loves company and the AI industry found itself at the top of the guest list. There was, however, one notable exception who declined to RSVP. While many leaders were busy panicking, Shopify was doing a victory lap.

Just four months after CEO Tobi Lütke’s internal memo went viral—a call to arms for Shopify employees to embrace AI—the results were in: it had been a resounding success, with a tool called Scout featured as the crown jewel of Shopify’s AI adoption.

Ask Scout about a new feature idea, and it will use AI to search through every meaningful live data point that Shopify collects from customers. We’re talking support tickets, product reviews, customer interviews, sales conversations, and more. Scout will use all of that data to give you a fully-sourced answer in plain language.

This gives anyone in the company a superhuman ability to understand the psychology, frustrations, and desires of Shopify’s enormous customer base. Employees at Shopify are using the insights from Scout to create better products, make customers happier, and build a stronger company—and as Tobi put it, bring together engineering and product teams in a way he’s never experienced.

But I recently learned that this is actually just the tip of Scout’s iceberg of success.

When I saw Tobi’s X post, standing out in contrast amidst the industry’s gloomy summer, I knew I had to learn more about what Shopify was doing differently. So I started digging, and traced Scout’s origin to something called the “Product Support Network” team, led by Rich Brown and Cristan Brown. I reached out and they kindly agreed to meet with me, share their story, and reveal the principles that underpin the success of both Scout and Shopify’s wider AI strategy.

During our conversation, Cristan revealed that Scout handles around 1,000 tool calls per day. In total, approximately 2,000 people—about 25% of the company—have used Scout to date.

This includes hundreds of people in product management regularly conducting product reviews, with Tobi himself being among the tool’s power users, revealing that he uses it “all the time” to evaluate new projects that Shopify considers launching.

When I explained to Rich and Cristan how their breakthrough appeared to be an outlier during the summer of AI discontent, Rich laughed, and confessed that Shopify was living in a “parallel universe,” one in which thousands of Shopify employees are charging hard at AI adoption and seeing incredible results.

They then proceeded to reveal seven principles that have contributed to the success of Scout and Shopify’s wider AI playbook. This article will pass those principles onto you. But this story is bigger than just one internal tool. Figuring out how to spot the most useful use cases for AI and bring them to life is the golden ticket question on the minds of millions right now.

Whether you want to be more productive in your own job, drive innovation inside your company to stay competitive, or launch entirely new AI-powered products, the story of Scout provides a clear playbook. 

All aboard!

P.s., our previous article, To Launch Something New, You Need Social Dandelions, struck a major chord with readers and ranked on the front page of Hacker News. If you haven’t read it yet, open it up in a new tab and dive in after you’re done learning about Shopify’s 7 principles!

Tobi Lütke insisted that everyone start using AI—but use AI to do what, exactly?

It's the question we all now wrestle with. We know AI is one of the greatest opportunities to innovate, but figuring out how to capitalize on it remains a giant, open-ended question mark.

But it turns out that Rich and Cristan had an answer to this question as far back as September 2024, several months prior to the publication of Tobi’s memo.

Around this time, they both decided to join the Product Support Network (PSN) team as co-leaders of the Product Experience arm of the group. The PSN is designed to be a bridge between customer feedback and product teams. Its purpose is to help employees at Shopify build what customers actually want rather than what Shopify thinks customers want.

The way the PSN team traditionally achieved this goal was by conducting deep dives—manually collecting samples of customer feedback, scheduling interviews, and then “bubbling up” insights across the organization.

But deep dives are time-consuming work. While every customer ultimately gives feedback for the same reason—wanting the product to work better for them—the touchpoints where this feedback is provided are scattered all over the company. The data comes from multiple different channels and reaches multiple different teams. This necessitates a lot of manual and repetitive work to collate, sift through, make sense of, and keep up to date.

At least, that’s how things worked in the old world.

Rich and Cristan already had a blurry vision of a tool that could automate the work of deep dives when they joined the PSN. In fact, they immediately made it a side quest for their team to explore an AI-powered solution that could automate deep dives.

And this is where we need to make our first pit stop.

While their idea may have been hazy, there was a principle behind it that was sharp—and it laid the foundation for Scout’s future virality.

1. Find Your 10,000% Bottlenecks

Cristan and Rich knew their team’s deep dives were a goldmine of insight. But even a world-class team can only pan for gold so fast. While the mountain of customer data was virtually endless, the team’s capacity to process it was not.

There was far more gold buried underground than the team could excavate. They were not limited by the volume of customer data to mine, they were limited by the size of their mining crew—they were limited by the amount of human intelligence that Shopify could afford to throw at the problem.

By commoditizing intelligence, AI can melt away these old constraints. It transforms intelligence from a scarce and expensive resource into one that is cheap and abundant. If the PSN team could use AI to break open this bottleneck, they could tap into the full depths of their data and distribute it to those who needed it instantaneously, unleashing a tremendous amount of value that had previously been trapped below the surface.

Bottlenecks like these are hiding in plain sight in every business—an ancient artifact from a time when intelligence was a uniquely human resource. Workflows that have long been subjected to these constraints are where the most seismic effects of AI are being felt.

Think about it. What is the most valuable work that you do, or that your team, or your company does? If you had the ability to execute 10,000% more of that work, at little additional cost, what would that look like? What if you had 100x more software engineers? Would they run out of work or would you ask them to ship more, higher quality software? What if you could produce 100x more content across every platform? Would you run out of content ideas or would you use that extra capacity to have a daily presence on every social and entertainment platform?

Generally we are not constrained by a lack of ideas or demand, we are constrained by the amount of intelligence we can afford to direct toward an activity. 

From the get go, Rich and Cristan could sense this invisible bottleneck with their deep dives. "That feedback loop has always been there, but it's been slow," Rich explained, adding that it "was a bit of a wicked problem previously because everything was distributed and it needed something like AI to come along to solve it." 

This intuition to target a workflow that was constrained by human intelligence is what put Scout on its course to become transformational rather than incremental.

To find your own Scout, look for target-rich environments where you are holding back simply because it is too expensive to hire more humans. Consider:

• Where are these bottlenecks hiding in your workflows?

• Which role do you wish you could hire 10x more of?

• Which workflows generate the most value that you would run 100x more of if you could?

• What work are you not doing today because it would be too expensive to hire for?

And to ensure you think big enough like the PSN team did, ask yourself: what if you were forced to find a way to complete the same amount of work (or greater) in 10% or less of the time you currently spend on it?

(Note: you might be wondering if the PSN team still has their jobs. After all, if they have automated themselves, what is left for them to do? We’ll learn their surprising fate further down)

So that’s the first key ingredient to Scout’s success: its purpose was to augment a high-value activity that was constrained by human intelligence from the outset. But while their idea had a solid foundation, it was still murky. The resolution of what would eventually become Scout came into sharper focus when Rich and Cristan discovered a newly released product from Google called “Notebook LM”…

2. Follow the 10x rule for tools

At the time, Notebook LM was a tool that gave you the ability to upload your own documents and sources, and then use AI to ask questions and summarize them in natural language. For example, a biographer could upload hundreds of pages of interview transcripts and personal letters, and then simply ask, 'What was her relationship with her mother like?' to get an instant, sourced summary.

When Rich and Cristan tried Notebook LM for the first time, they realized it was in the ballpark of what they were looking for. But while it was a step towards what they wanted, it had its limitations. For one thing, moving all of Shopify’s data into the tool would require a painstaking amount of manual data entry. It also wasn’t live. It would require constant manual updates, meaning the juice wouldn’t be worth the squeeze. Notebook LM did, however, help inspire the team to build their own tool.

Shopify has a framework for deciding when to build its own tools and it comes with a pretty high bar. One of Tobi Lütke’s maxims is: only build a tool yourself if you think you can make it 10x better than what’s already available. This 10x threshold ensures a custom-built tool is worth the effort and long-term maintenance, rather than just using an off-the-shelf product.

Rich and Cristan were confident that a chatbot with live access to Shopify’s data would indeed pass this test and offer 10x more value than Notebook LM. 

So they got to work. 

They started by mapping out the dream in greater detail with what Shopify calls a “Greenpath”. At Shopify, a Greenpath is simply a design exercise to map out the ultimate “pie in the sky” vision for a product. When the vision is clear, the team then breaks it down into smaller, actionable “scopable chunks” to build toward it.

Before we get into what the team did next, let’s recap our second principle.

Once you've identified a 10,000% bottleneck, the next step is to dream about what the perfect AI solution would actually look like. Start by researching what tools are already out there. Then ask: can you build something at least 10x better? (And make sure you read principle #5 before answering this question!) If you can, that’s your green light to get going. If not, stick with the best alternative.

So the PSN team had their action plan in place.

There was just one problem… they weren’t software engineers!

Sure, they could envision a 10x tool, but did they have the coding expertise to actually create it?

A few years ago, the answer was no.

But in 2025, the very definition of a 'builder' at Shopify was changing.

3. Play before value

Inside of Shopify, there exists a Slack channel where people share games they’ve built using AI coding platforms like Cursor, Replit, and Loveable. If you aren’t familiar, these tools allow anyone to describe what they want in plain language and watch as AI writes the code, meaning non-technical people can now build fully functional apps and sites.

The reason for creating this games channel is not because Shopify plans on competing with Nintendo anytime soon. The idea is to give people a fun, low-stakes way to tinker with these powerful new tools. It’s hard to imagine what a tool is capable of until you understand how it works, so the channel encourages people to dive in. The practical, high-value ideas come after you build familiarity.

Shopify encourages playtime for this reason: play is a precursor to value.

(Shopify engineer, Daniel Beauchamp has written more about the company’s culture of “unserious exploration” here)

One way that Rich and Cristan’s team specifically encouraged playtime was by setting an open-ended goal of building ten AI use cases in a month. They admitted this seems silly in hindsight but it's this kind of unstructured exploration that arms you with the knowledge necessary to spot valuable use cases. 

Playing with vibe coding tools is what gave the PSN team the confidence to build Scout. They realized they could use a platform like Cursor to largely “prompt” the app into existence. 

As Cristan put it, “we take weird and make it normal very quickly”. Things that seem foreign and “on the edge” are normalized within days at Shopify. Rich explained that their strategy was to intentionally "direct that hacker energy towards a cumulative feature set. So rather than everyone making their own little game, people are building on top of an idea.” This is how the side quest to create Scout gathered momentum, turning a series of playful experiments into a powerful, unified tool.

Their playtime also brought about an unexpected consequence: they were starting to deepen their understanding of how to read and write code without setting out to do so. Even though they weren’t seasoned software engineers, they saw how they could use AI-assisted coding to get Scout at least 95% of the way to the finish line.

Our third takeaway is to encourage playtime first, and trust the value will come second.

And we’ve already hinted at the fourth principle behind Scout’s success. As the PSN team started to build, they stumbled upon the most important discovery of all, one that would fundamentally change their careers.

4. You're 7/10 at everything

At Shopify's annual summit, Carl Rivera, the Chief Design Officer of Shopify at the time, made a keen observation about the impact of AI: "Everyone's now 7 out of 10 at everything."

The PSN team was living proof. Cristan had spent her career in customer success and sales. Yet, "I was reviewing code yesterday," she told me, laughing. "I have no business doing this." 

But she was. She was reviewing prototypes, understanding the logic, suggesting improvements. The project had accidentally turned her into a technical leader. And she wasn't alone—a portion of Rich and Cristan’s team of former support analysts and user researchers were suddenly shipping code, building data connections, and fixing bugs. The technical and non-technical divide was evaporating.

The speed at which they could now work was almost comical. Before AI, it would take a group of six to eight team mates working for four days just to ship one feature. "Now one person can do it in an afternoon on the side of their desk," Cristan said.

Rich called it having your skills "jet-packed," he explained, "so that you're now able to shorten the distance between, 'I would love this thing' and having to ask somebody else to do it.” This changes everything about how teams operate. You can be wildly more ambitious because everyone—not just engineers—can build. Everyone—not just writers—can create content. Everyone—not just designers—can produce visuals. In a world where everyone starts at a 7/10, problems can get solved faster because the person who spots them can fix them.

And perhaps most importantly, roles become fluid. The CFO can ship code. The support lead can create marketing videos. The engineer can write strategy docs. The product manager can design interfaces. When everyone can contribute to everything, you can tap into the collective intelligence of your entire team, not just designated specialists.

As you explore ways of integrating AI, and even consider the possibility of building your own tools, remember that you are now at least a 7/10 at everything. The PSN team were simply one of the first to recognize they had these new superpowers and to start using them. We all need to break our pre-existing notions of what we can accomplish and adjust our goals, our standards, and our identities accordingly.

That’s takeaway #4.

Armed with their new superpowers, the PSN team pressed forward with Scout. They vibe-coded their way to what seemed like a working prototype. They could query multiple data sources. They could get answers in natural language. So what happened next was surprising.

The app was mostly a disappointment.

It was impressive for a team of non-engineers, but it wasn't the 10x tool they'd envisioned.

MIT's study had found that 95% of AI pilots fail and Scout was starting to look like another statistic.

5. Build with tomorrow’s capabilities in mind

In early 2025, something called the Model Context Protocol (MCP) was gaining steam.

Prior to MCP Servers, connecting an AI model to an external tool or data source was a massive headache. Want your AI to talk to your customer database? Custom integration. Want it to pull from support tickets? Another custom integration. Want it to access your internal wiki? You get the picture.

Each bridge between your AI and your data required a specific setup. When you're working with multiple models, data sources, and interfaces—like Shopify was—the number of custom connections quickly gets out of hand. Pulling this off would have been a heavy lift that required a dedicated internal engineering team or hiring an external software vendor.

MCP Servers changed the game by acting as a universal translator. Instead of building a hundred bridges, you build one. One protocol to connect everything.

For Scout, this was the missing puzzle piece.

"The problem that we were solving of pulling in disparate pieces of information is the perfect expression of what you would use an MCP Server for," Cristan explained. "It was like the right problem with the right tooling." It also changed who could build it. MCP Severs removed the final technical barrier that was holding the team back.

They dove back in, rebuilding Scout using the MCP protocol. This time, it clicked. The answers were accurate, the sources were properly linked, and the whole thing worked like a charm. The early responses from their colleagues were electric and it was clear that Scout was no longer just an ambitious experiment—they had a winner on their hands.

When you're building AI tools, you need to understand what the models are capable of today, but build for what they'll be capable of tomorrow.

When the PSN team started building Scout in September 2024, MCP Servers didn't exist. They wouldn't arrive until two months later and didn’t gain widespread adoption until early 2025. But the team built anyway, learning what worked and what didn't, mapping the shape of the problem they were trying to solve. When MCP Servers finally arrived, they didn't need to start from scratch—they just plugged in the missing piece.

There are countless examples of AI startups that struggled for months with a half-working product, only to explode in popularity when a new model or technology dropped that made everything click. Scout proves this same dynamic applies to internal tools.

Just because something doesn’t work right away, don’t give up. Even if 95% of AI pilots fail, you don't know how many of those failures are close to becoming a success in the near future (also AI is so transformative that you could argue the successful 5% may easily make up for the 95%—a 1:19 success ratio might be perfectly acceptable if the failures are cheap and the successes deliver step change results). 

A tool that doesn't work today might just need a smarter model next month. A prototype that feels clunky might be waiting for the right protocol. A feature that seems too expensive might become viable when costs inevitably plummet. AI technology is evolving so fast that you should work with a bias toward optimism about its future capabilities.

Takeaway #5 is to build with tomorrow’s capabilities in mind and resist throwing away failures too soon.

Once they knew they had something special, the PSN team called in some engineers to help prepare Scout for its company-wide debut.

Six weeks after they’d discovered MCP Servers, Rich and Cristan were ready to take to the stage. Literally.

6. Lower the adoption bar

Several months ago, a friend of mine was experimenting with AI to create images for her tattoo business. Back when she first tried it, the models weren't quite up to the task—you could never get all the details right at the same time.

But once I discovered Google's new ”Nano Banana” image editing capabilities, I knew she had to give AI another shot. But I must have suggested she try it at least ten times before she finally did. It turns out that convincing people to try new tools and change how they work can be a tough task.

Why did my persuasion attempts fall so flat, yet Scout conquered Shopify—an 8,000+ person organization—in a matter of weeks?

It's not by accident.

When Rich and Cristan were ready to unveil Scout to the wider Shopify workforce, they didn't send a memo or pester colleagues in the cafeteria—they got up on stage at Shopify's Summit, the company's massive annual gathering, and just demoed it.

People could see Scout answering real questions, pulling real data, solving real problems. No imagination required. Who knew that showing beats telling? Everyone at Shopify, apparently. Rich noted, "we are very much an organization that's ever more like... don't give me the buildup, just show me the thing. And AI helps you do that a lot."

(Tragically, I should have known better myself. I've written entire newsletter editions on this principle and still managed to forget it in the case of my friend.) 

But the Summit demo was just the beginning of their adoption strategy. The PSN team did two other brilliant things to drive usage.

First, they added sources to every answer Scout generated.

Humans have a peculiar relationship with machine error. We've somehow accepted thousands of annual traffic fatalities as the price of human driving. But one self-driving car accident triggers calls to ban the entire technology.

The same psychology applies to AI. One hallucination can damage trust. By making Scout cite its sources—linking back to the actual support ticket, the exact Reddit thread, the specific customer interview—they baked trust directly into the product. Users could verify anything Scout told them with a single click.

Second, they met people where they already worked.

"We didn't want to create a new surface that people had to go to," Cristan explained. Shopify employees already used a platform called LibreChat for AI interactions. Scout appeared as a simple dropdown option within LibreChat.

Shopify uses Slack for communication. Scout showed up there too. Scout even integrated seamlessly in Shopify’s internal Wiki and inside Cursor. No one had to learn a new tool, visit a new website, or change their workflow. Scout just appeared one day in the places everyone already spent their time.

Remember: however powerful your AI tool is, adoption isn't automatic. You need to show its value visually, build trust into the experience, and remove every possible friction point. Once you’ve discovered a transformative use case for AI and want to share it with others, lower the adoption bar as far as humanly possible. That's takeaway #6.

But the truth is that Rich and Cristan weren't fighting this battle alone. They were pushing on an open door. Shopify had already spent months preparing its culture for exactly this kind of innovation. Which brings us to our final principle…

7. Prime your culture for adoption

While the PSN team was building Scout from the bottom up, Shopify's leadership was engineering a culture of adoption from the top down.

They started by making AI a core part of the job.

Rich explained that leadership baked AI usage directly into performance reviews. At first, the question was soft: "How do you feel about AI?" As the months went on and access to tools became universal, the question evolved: "How much are you using it?" Finally, it became what Rich called a "baseline expectation"—a table-stakes part of everyone's role, no different from using email.

Next, they gave employees access to a suite of AI tools. Rich pointed out that if you work at Google, you're expected to use Gemini. If you're at Meta, you use Llama. At Shopify, meanwhile? "Anybody could use anything." This policy encouraged employees to find the best tool for the job, not just the one the company owned, accelerating learning and experimentation.

Finally, they hosted company wide events where everyone is expected to build. Rich described their hack days as events where "people aren't waiting for the designer and the engineer... everybody's making stuff." This breaks down the traditional silos between technical and non-technical roles, giving everyone permission and dedicated time to move from idea to prototype.

The result of these initiatives—making AI a job requirement, providing unrestricted tools, and running inclusive hack days—was a workforce that Rich described as "internally unleashed."

This is the final principle: Scout's success was the product of a culture that was intentionally designed, from the top-down and the bottom-up, to both make it happen and embrace its final form. It’s not enough to use AI to build 10x solutions, you also need to build a 10x culture that’s ready for change.

The New Main Quest

There’s one question that’s been hanging over this entire story. 

If Rich, Cristan, and the PSN team created a tool to automate their core workflow of deep dives—what's left for them to do? This is the question at the heart of the AI anxiety epidemic. What happens when you automate yourself out of a job?

What happened to the PSN team suggests the future might be much brighter than the doomer headlines imply. Remember how Scout started as a "side quest" for their team? Well, Rich and Cristan revealed that building Scout is now their "main quest." 

No jobs were harmed in the making of Scout, they were upgraded. 

Their team's primary function is now to move Scout forward, transforming from a support function into a team of builders. They’ve gone from being gold miners to the architects of the entire gold mine. Their new challenges are bigger and more interesting, like integrating video feedback from merchants.

In fact, they've become one of the most sought-after teams in the company. "Recruiting for our team has become a lot easier," Cristan shared, noting they've been "flooded with requests" from employees in different corners of Shopify wanting to join their mission. After Tobi's X post put them in the spotlight, their career capital has never been higher.

Beyond just giving us a roadmap for creating high-value AI use cases, the story of Scout is a powerful counter-narrative to the prevailing fears of job displacement. It shows that when you use AI to automate a workflow, you don't just eliminate a task, you unlock the human potential that was trapped inside it, and give yourself the freedom to move onto higher leverage activities.

These are the seven principles from Shopify's success that you can use to build your own transformative AI solutions:

1. Find Your 10,000% Bottlenecks 

2. Follow the 10x Rule for Tools 

3. Play Before Value 

4. You're 7/10 At Everything 

5. Build with Tomorrow’s Capabilities in Mind 

6. Lower the Adoption Bar 

7. Prime Your Culture for Adoption

Final Calls To Action

• Want to understand the implications of recent advances in tech, culture, and product design? If so, Scott Belsky’s monthly analysis is essential reading. In his latest December edition, Scott explores how ambient AI summaries may augment our daily lives and why widespread AR might be closer than you think.

• It’s not too late to pick up some Action Method products for this giving season. Whether you’re investing in your own toolkit for turning ideas into action in 2026, or gifting an action-oriented loved one something that says “I believe in your potential,” now’s the perfect time to stock up.

  Give the gift of action.

Thanks for subscribing, and sharing anything you’ve learned with your teams and networks (let us know what you think and share ideas: @ActionDigest).