Company guide · Shopify

Shopify made AI fluency
a hiring requirement.

While Meta engineered a controlled environment, Shopify went the other direction — bring your own IDE, your own AI tool, your own setup. The most permissive AI interview in big tech, and a direct extension of CEO Tobi Lütke's "AI is now a fundamental expectation" memo.

Last updated April 29, 2026. By the Nyrion team.

Why Shopify made it BYO

Shopify's CEO drew a line in the sand. In a March 2025 memo to all employees, Tobi Lütke wrote that "Using AI effectively is now a fundamental expectation of everyone at Shopify" and required teams to demonstrate AI couldn't do a job before asking for new headcount. He posted the memo publicly because, in his words, it was already being leaked.

"Reflexive AI usage is now a baseline expectation at Shopify."— Tobi Lütke, March 2025

Once that's the operating principle, hiring follows. You cannot test for "reflexive AI usage" with a sandboxed interview. The candidate's actual workflow — which assistant they reach for first, how they configure their editor, when they switch tools, what their muscle memory looks like — is the signal. Sandboxes wash all of that out.

Shopify's engineering leadership has been blunter still:

"If they don't use a copilot, they usually get creamed by someone who does."— Shopify engineering, on AI in interviews

Our take: this is the most internally-consistent AI-interview policy at any major company. Meta moved fast and built a controlled environment to test the specific skill. Shopify moved fast and removed every constraint that wasn't measuring the actual skill. Both are defensible — but Shopify's matches its CEO's stated beliefs more directly than any other company we've seen.

What the interview actually looks like

Forget phases. Forget the bug-fix warmup. Shopify gives you one problem at the start of the call, you spend the first couple of minutes walking the interviewer through your setup, and then you build.

Opening — the workflow tour

The first signal you give Shopify is your environment. Walk the interviewer through it: which IDE, which AI tool, which terminal, how you've laid out your workspace. They want to see thoughtful workflow planning. Showing up with an IDE you've never opened before is a signal in the wrong direction.

The build — open-ended evolution

One problem. Starts simple. Evolves. The interviewer adds requirements as you go — new constraints, edge cases, scaling concerns — and your job is to keep the codebase healthy as it grows. You're starting from an empty repo or close to it; there's usually no scaffolding.

Empty-repo starts are doing real work in this format. You decide the structure, the test setup, the file layout, the naming. That's a stack of architectural decisions traditional interviews never see. It's also the part where most candidates underestimate prep — they're used to problems with a function signature handed to them.

Tool switching is encouraged

If your IDE's assistant is stuck, copy the problem to Claude or ChatGPT in another window. Interviewers explicitly allow this and read it as a positive signal — you know when to switch tools, which is itself a senior engineering skill.

Shopify's three-tier evaluation framework

Across both AI rounds, Shopify is testing three distinct bands of competency. Some problems hit one band; others hit all three.

  • No-AI band. Can you write code unaided when the tool fails or it's overkill to invoke? Some sub-problems are deliberately small enough that reaching for the assistant slows you down.
  • AI-optional band. The problem is solvable without AI but materially faster with it. Are you reading the context to know which it is?
  • AI-mandatory band. The scope is calibrated past what's possible to type by hand in the time given. If you don't wield the assistant well here, you don't finish.

The interviewer is reading which band each piece of work falls into and watching whether you adjust your approach accordingly. Reaching for AI on a six-line utility is a negative signal. Refusing to use AI on a forty-minute integration is also a negative signal.

The contrarian read

Most candidates panic at "BYOE" because there's nothing to study. They're wrong. The thing to study is yourworkflow, not the format. Specifically:

  • Pick your toolchain in advance and use it daily for weeks. Don't try to learn Cursor the night before the interview. The interviewer will see exactly when your fluency breaks.
  • Practice empty-repo starts. Get fast at the decisions most interviews skip — directory structure, test setup, type config, formatter, dev script. These should take you less than two minutes.
  • Build the verification reflex. The single biggest failure mode reported by candidates is pasting AI code and moving on. Read it, run it, write a test for the specific claim it's making.

Shopify's CEO publicly tells managers to prove AI can't do a job before they ask for headcount. The interview inverts that: prove you can do the job with AI beside you. The bar is high; it's just not what people expect.

Where this goes next

Shopify's BYOE format is the leading indicator for what the rest of the industry will look like in two years. Here's why:

Sandboxed environments lose information. Meta's CoderPad-with-AI gives a controlled signal but strips out the candidate's real toolchain. As more companies realize workflow fluency is the signal they're hiring for, BYOE-style formats will spread.

Open-ended problem shapes are the next wave. The phase-by-phase structure Meta uses is a transitional form. It still feels like algorithm interviews dressed up. Shopify's evolving-single-problem format is closer to what production engineering looks like, and it's a better predictor of on-the-job performance.

The CEO-driven version of this format is hard to copy without the matching policy. Shopify can BYOE because its CEO has mandated AI fluency company-wide. Companies that haven't made that organizational commitment can't credibly run this interview — their engineers haven't built the muscles, and the interview would catch them out. Expect the BYOE format to roll out wherever leadership has gone publicly AI-first first.

How to prepare for it

  • Pick one assistant and live in it. Cursor with Sonnet, Claude Code, Codex CLI — pick one that fits your stack and use it daily for at least a month before your interview. Tool fluency is the signal.
  • Practice from-empty starts. Most prep platforms hand you a starter file. Shopify won't. Get fast at the first 60 seconds: mkdir,npm init, test runner, type config, lint, format-on-save. Make the muscle memory invisible.
  • Practice the workflow tour. The first thing the interviewer sees is you walking through your setup. Nail the 90-second version: which IDE, which assistant, which keybindings matter, why this tool over that one.
  • Build problems that evolve. Don't drill isolated LeetCode. Practice problems where requirements shift mid-build — the catalog at Nyrion is structured this way deliberately.

Frequently asked, briefly answered

Can I use any AI tool?

Yes. Cursor, Claude Code, Copilot, Codex CLI, Cline, Cody — whatever you actually use. Switching between tools mid-interview is also fine and reads as a positive signal if it's intentional ("the IDE assistant got stuck on this, I'll bounce it to Claude in another window").

What if my IDE crashes mid-interview?

Recover gracefully and keep going. Interviewers know tools fail. The signal is how you handle it — quick recovery, fall back to a second tool, don't lose 10 minutes restarting the world.

Do I need to use AI?

Not for every line. The three-tier framework explicitly includes a "no-AI band" where reaching for the assistant slows you down. Use AI when it accelerates you; don't when it doesn't.

How big is the problem?

Bigger than you'd expect for one open-ended hour. Calibrated assuming you use the assistant. Candidates who refuse to use AI report running out of time well before the interviewer's last question lands.

Is the bar lower than at Meta?

No. The bar is differently shaped. Meta tests structured execution inside a controlled three-phase format. Shopify tests workflow fluency, environment-setup speed, and the judgment to pick the right tool at every moment. Different skills, comparable difficulty.

Practice problems for the Shopify format

Open-ended, evolving, real codebases — that's the prep loop that maps to Shopify's interview. Browse the catalog and look for multi-stage problems where requirements shift between stages.

Sources