Shopify CEO on the Future of Software Engineering and AI

Editor's note:

I recently listened to the ACQ2 podcast featuring Shopify CEO Tobi Lütke, and it was one of those experiences that leaves you with a dozen open tabs and your brain buzzing with ideas. What struck me most wasn't just the business insights (though Shopify's journey from a $1.5 billion IPO to nearly $200 billion today is remarkable), but Lütke's deeply philosophical approach to this technological moment we're living through.

What's truly impressive about Shopify is that they've returned to their pandemic peak market value while actually reducing their workforce. The company now operates with fewer employees than before, yet their efficiency and results have dramatically improved. This is the same CEO who famously declared that if AI can solve a problem, they won't accept hiring someone to do it. When I interviewed with Shopify, I experienced this firsthand—their entire interview process has already been completely AI automated. It's fascinating to see a company truly walking the walk on AI transformation.

He describes feeling the "privilege of a lifetime" to witness another platform shift, and I think he's absolutely right. We're experiencing something profound: there's now "software that no one wrote," as he puts it, which we must actively "interview" to understand its capabilities and limits. That framing alone changed how I think about working with AI.

The Art of Evaluating AI: The Toby Eval

What fascinated me most was Lütke's systematic approach to understanding AI. He maintains what he calls the "Toby eval"—essentially a personal test suite of prompts with expected results that he runs against every new model release. This isn't abstract benchmarking; it's artisanal, hands on exploration.

Actually, I have been doing this for a long time.

Think about that for a moment. The CEO of a $200 billion company maintains his own folder of AI tests, personally evaluating each model like a craftsman testing new tools. His goal? To figure out how to structure these models to become "idealized non judgmental teachers for people." This kind of direct engagement is how genuine insight emerges.

Code as First Draft: The New Economics

Lütke's observation about the economics of software creation really resonated with me. He points out that code was once "by far the most expensive thing to produce." Now, with AI, much of the boilerplate and gluing code can be treated as a first draft—something abundant rather than precious.

This shift is revolutionary. We can now afford to experiment, to build and discard, in ways that were previously unimaginable. It reminds me of how word processing changed writing: when mistakes became cheap to fix, the entire creative process transformed. We're seeing that same transformation in software, but amplified.

The historical parallel he draws is compelling too. In his apprenticeship days, software was viewed like civil engineering—build the bridge once, and it's essentially done. But modern software "evolves forever." AI accelerates this evolution, making continuous improvement not just possible but economically inevitable.

The Ceiling and Floor Paradox

Here's where Lütke's thinking gets really nuanced. He articulates a fundamental tension in tool design: the balance between "raising the floor" and "not lowering the ceiling."

Most successful tools raise the floor—they help everyone achieve "good enough" results, maybe a 7 out of 10. That's valuable; it democratizes capability and creates business value. But the danger is that in making things easier, we might inadvertently constrain what's possible at the high end. We might lose the space for the 10 out of 10 masterpieces.

He references early mobile development tools like PhoneGap or the HTML5 movement as cautionary tales. These "lossy abstractions" pretended the world was simpler than it was, ultimately constraining developers from building truly great software. The lesson? AI should amplify human potential without limiting the heights we can reach.

Living in the Future

Lütke's approach to innovation is refreshingly pragmatic. To ensure Shopify builds for tomorrow's consumers, he deliberately lives in "everyone else's relative future." He runs beta versions of operating systems on his primary phone, treating inevitable bugs as "interesting constraints" rather than annoyances.

This isn't just about testing software; it's about cultivating a mindset. By immersing himself in emerging technologies, he develops intuition for what's coming. It's the difference between reading about swimming and actually getting in the water.

The practical application at Shopify is what he calls helping customers "hill climb"—constantly improving their businesses through AI. One example stuck with me: a carpet seller struggling with European sales used AI to transform product images from Malibu beach houses to Parisian apartments. Sales instantly tripled. No expensive photo shoot needed. That's the kind of previously impossible business optimization AI enables.

Context Engineering: The New Fundamental Skill

Perhaps the most actionable insight from the conversation was Lütke's identification of "context engineering" as the foundational skill for the AI age. It's the ability to state a problem with enough context that it becomes "plausibly solvable" without additional information.

This resonates deeply with my own experience. The better I've gotten at providing context to AI, the more valuable the outputs become. But here's the unexpected benefit Lütke mentions: mastering context engineering makes you a better communicator overall. Much of what we call "politics" in organizations, he observes, is actually just bad context engineering or fundamental disagreements on assumptions.

Shopify teams are codifying their unique principles into "Constitutions" (borrowing the term from Anthropic). These documents define their specific tradeoffs and choices—the decisions where another company might plausibly choose differently. By running projects against these Constitutions using AI, they create both an evaluation framework and a guide for necessary discussions.

Leadership in the Age of AI

Lütke's views on leadership versus consensus particularly struck me. He asserts that "consensus is always the absence of leadership." Consensus prioritizes safety, typically resulting in 6 or 7 out of 10 outcomes. It raises the floor but lowers the ceiling.

Great works, he argues, come from a single vision executed by a team, not from an agreed upon vision. In managerially run companies, legitimacy is invested in the plan—changing anything requires enormous energy. In founder run companies (or "refounded" ones like Satya Nadella's Microsoft), legitimacy is invested in an individual, allowing the company to pivot on a dime.

His chess analogy is perfect: though machines play the best games, humans still watch humans playing chess. AI capabilities should be viewed as instruments that humans play to explore new creative spaces. The successful people five years from now will be those who best know how to use these assisted tools.

The Quiet Revolution

One observation that really made me pause: the Turing Test passed without fanfare or front page news. We've already normalized something that was once the holy grail of AI research. Lütke believes people a thousand years from now will study our era as a crucial moment when humanity gained incredible individual agency through technology.

Listening to this conversation, I'm convinced we're not just witnessing a technological shift but a fundamental change in how humans create and solve problems. Lütke's insights aren't just about building a company; they're about thoughtfully building the future. His ideas challenge us to think beyond the hype cycles and consider how we can be more creative, more effective, and ultimately more human in an age of intelligent machines.

The message is clear: this isn't about humans versus machines, but humans with machines, creating solutions neither could achieve alone. And that's a future worth building toward.