Augmenting Engineers with AI at Shopify

About Daniel Doubrovkine

Daniel Doubrovkine (aka dB.) currently runs the Augmented Engineering team at Shopify. Prior to this, he served as Principal Engineer at AWS in New York, working on OpenSearch. He is a seasoned entrepreneur, technologist and former CTO at Artsy.net. Daniel graduated from University of Geneva in late 90s with a degree in Computer Science. Daniel is the creator and maintainer of many popular open-source projects and a lifetime artist.

Improving Developer Experience at Shopify's Monorepo with AI 

In this episode of the HangarDX podcast, Ankit Jain, co-founder and CEO of Aviator, talks to Daniel about AI’s impact on software engineering and developer productivity.

Daniel shares insights on Shopify’s AI-integrated workflows and how they use AI to improve developer experience in what he calls a giant monorepo, the world's largest Ruby on Rails app. He also reflects on his career path from individual contributor to management and back, and the importance of staying hands-on with code.

What is Augmented Engineering

Augmented Engineering means augmenting engineers with AI tools or augmenting tools with AI. Paolo Arruda from my team coined the term.

We’re an engineering tooling team within Shopify's Developer Productivity organization. The distinctive aspect of our work is that we’re specifically focused on augmenting engineers with AI wherever it makes sense so they can be more productive. You can think of Augmented Engineering as a slice of DevEx that’s AI-first

Shopify is deeply committed to this. I thought my last job involved a lot of AI, but when I joined Shopify, I was blown away. I haven’t talked to a human in IT in months—bots just handle everything seamlessly.

Reflexive Use of AI

A lot of companies talk about adopting tools like Claude, Copilot, and Cursor. At Shopify, that’s just the default—it’s what we call “reflexive use of AI.” You just try AI first for every problem.

For example, internally, IT support is almost fully AI-driven. But it’s not just about adopting existing tools. It’s about seamless integration into workflows so that humans and AI collaborate constantly. That’s true in engineering, in customer support, everywhere.

That’s not easy for me because I'm a very old school. I really have to try to use AI to do something before I apply the method I’m used to. Recently, I needed to extract a list of things from a website. And so my initial way I do it is kind of crazy -  view source, copy paste, and bulk replace with multiple selects. I asked myself - Why am I doing this, this is insane?!  So I downloaded the file,  pulled up Cursor, and I told it to write me a script that extracts the list of X from this from the website, and it got me me two pages of JavaScript that I can just run in console and get back a CSV file of that output. Done, that’s it!  It's super productive power.

Developer Experience at a Giant Monorepo

We have thousands of human engineers working in what’s essentially a giant monorepo, the world’s largest Ruby on Rails app. It's a really impressive piece of technology.

A few years ago, we had hundreds of independently deployable services. That sounds good, but it meant customers could see outages because each service could break independently. The answer was to go to Monorepo and have a single CI pipeline that takes the monorepo multiple times a day, runs all the tests, and makes sure that it is deployable in waves in production. That whole workflow is highly tooled. 

We have internal tooling called Dev that manages sparse checkouts of the monorepo for you, transparently. You can bring up a local environment that mirrors all or part of Shopify. The north star of that tooling is that you can vibe code a feature in any part of Shopify using AI tools. I've seen demos where people do that, and sometimes they're not engineers, and they're able to use engineering tooling without having to understand what's going on under the hood.

Better Developer Experience with AI

I'm very active in the open-source world. One of the projects we contribute to is LibreChat, an open-source chatbot. We use it at Shopify, and my team also owns it.

In open source, setting up your dev environment often means spending half a day reading the README, running npm install a couple of times, figuring out the test setup, and hoping for the best. Only then can I go and edit the code and see my changes. 

At Shopify, our internal fork is not even a fork anymore. It's just like an internal mirror of the upstream repository because we work primarily upstream. I can go into it and do dev CD LibreChat, dev app, and voila! I have a completely working system with its multiple databases, caching servers, RAG pipeline system…I don't need to know anything about npm, package managers, or none of it. It opens a Cursor window for me, and I can go and ask the AI to add a feature. 

Once you're in the CI pipeline, AI can become a real collaborator to your pull requests, comment on things, improve things, write tests for you, tell if the tests are of low quality, etc.  

‍We think of AI as a companion or an agent that’s with you at every step. That’s what we’re building: tooling that lets AI help everywhere, from writing code to reviewing it to maintaining it.

Planning and Prioritizing DevEx Initiatives and Tooling

Shopify's mission is to make commerce better for everyone. Developer Productivity Organization's mission is to make making commerce better for everyone.

My team’s specific mission is to augment that with AI. We prioritize projects that will deliver meaningful, long-term productivity gains.

We’re not looking for problems that fit AI; we’re solving real, large-scale problems that are hard for humans alone to solve. For example, we have hundreds of thousands of tests. But coverage density isn’t great—we want to use AI to generate high-quality missing tests. Why? ultimately, we want AI agents to refactor our code safely, and for that, we need robust test coverage.

Flaky tests: A Never-Ending Problem AI Can Tackle

At our scale, flaky tests are a constant problem, and flaky tests plague your path to production. So far, we've been tackling flaky tests in all the standard ways: isolating them and running them in a different order, identifying the flaky ones, assigning them to developers and addressing them really seriously. But that's a never-ending problem. Can we do something and maybe end it? Can we have AI do this work? That would be an incredible productivity boost for the entire organization.

AI is Getting Better Fast

Shopify open-sourced Roast, an AI workflow tool. Our use of it is interesting - we wrote a workflow that grades tests, and we wrote another workflow that generates tests. One can use the other. If you get an F grade test, generate a better test, improve it, and keep iterating until you get an A grade.

​​Every Friday, somebody comes to the Shopify Slack channel and goes, "My God, the new thing just solved this big problem."

The AI is definitely getting better fast. I don't know where it ends, but I certainly want to be there.

How to Adopt AI in Your Org

The attitude shift is critical. But you also have to do the hard work of giving access to these tools to your organization. 

• If you don't have an internal chat deployed today, start there and then connect MCPs to everything in your company. These are enabling tools. Deploy them to change the way the company works.
  
  
• Give developers access to AI-powered tools like Cursor or Copilot immediately.
  
  
• Change your hiring process. Stop doing whiteboard puzzles and start allowing the use of AI in your interviews. Ask candidates how they’ve used AI in their workflow, and hire those who have tried the tools, persevered, and created amazing results. Those are the people you want! 

From IC to Manager and Back

I love coding and have never really stopped. I keep going back and forth between wanting to just write code and be an experienced IC because I enjoy it, and trying to find groups of people who can deliver something extraordinary together, multiplying each other's capabilities.

I didn’t deliberately become a manager. I just built something, and someone said, “Here’s five headcount. Do you want to lead this?” And I was suddenly a manager.

Over time, I became more deliberate about it. I wanted to set direction for a team, be responsible for the technology, work with people, and enable them to do their best work.

I like working with people and enabling them to do their best work. But I also love just writing code. Both jobs are awesome in their own ways.

Managers Should Write Code

I still write code, especially on weekends or for my pet projects. I also contribute a lot to open source. I use AI all the time when I write code, but I still like getting that dopamine fix from adding a feature or fixing a bug. I don't write a lot of code, but I write code every few days. It depends on where my team and organization are at.

A good metric of team health for a manager at any level is if they have time to write some code. It doesn’t have to be production code—but you should be able to stay connected to the craft.

AI Changing Managers' Roles

Good management requires a manager whom the team can trust. Engineers trust other engineers, craftsmen trust other craftsmen, no matter what your title is or how senior you are. They look at people by what they do and what they make. It's really “easy” to earn trust from an organization of developers. Writing code is like speaking a language. You just need to speak that language.

Managers’ roles are also changing with AI. For many things, you no longer have to delegate three levels down to a pool engineer to collect data and write code; you can do it yourself and have answers in an instant. Everyone is a developer. These tools are force-multiplying and anybody can drive them.