All podcasts / Acquired / Summary

ACQ2: How to Live in Everyone Else's Future (with Shopify CEO Tobi Lütke)

2025-09-18 - source - Read full transcript
Ben Gilbert (host)David Rosenthal (host)Tobi Lütke

Key insights

Context engineering, not prompt cleverness, is the core AI skill worth building.
Tobi defines it as stating a problem with enough surrounding context that an AI can plausibly solve it without asking follow-up questions. He says building this skill made him a better email writer and let him diagnose that much of what people call company 'politics' is actually poor context engineering between humans who disagree on an unstated assumption.
ai-context-engineering
Tobi maintains a personal folder of AI evals and treats every model release as an interview, not an upgrade to trust blindly.
He runs a standing set of prompts with expected, judged results against every new model, comparing it to how automated unit testing was a resisted novelty in the late 1990s before becoming standard practice. When AI fails a task, he treats that failure as a new eval to add to the collection for testing future models.
ai-context-engineering
Shopify mandates AI as the first pass on every task, and treats failed AI attempts as useful data rather than embarrassment.
Tobi's internal memo requiring employees to reflexively reach for AI first was controversial when leaked, but he argues it was necessary because withholding the requirement would unfairly let early AI adopters monopolize the best career opportunities. A failed first AI attempt still produces a concrete eval for judging the next model.
ai-context-engineering
'Constitutions' work as company documents only when every statement is a genuine, contestable trade-off, not a platitude.
Shopify's constitutions (a term borrowed from Anthropic) exclude anything a reasonable company would obviously agree with, like 'teamwork matters.' Tobi feeds transcripts and edge cases to AI to surface what's missing or contradictory in these documents, then negotiates the wording with the model until it holds up.
ai-context-engineering
Good tools raise the floor of outcomes without lowering the ceiling; most abstractions fail at the second half.
Tobi's framework rates any platform on a 1-10 scale of achievable quality. Xamarin and HTML5-hybrid mobile frameworks are his examples of abstractions that raised the floor for weak teams while permanently capping the ceiling for everyone who built on them, unlike mature operating systems, which constrain almost nothing.
abstraction-and-tool-design
AI's real product value is shifting the distribution of user outcomes toward their own vision, not replacing human judgment.
He illustrates this with a Shopify merchant who used AI to swap product photos from beach houses to European apartments in minutes instead of commissioning a new photo shoot, and with chess: machines have outplayed humans for twenty years, but audiences still only care about human players like Magnus Carlsen, because the point was never raw computational output.
abstraction-and-tool-design
Founder-led and managerially-run companies operate on fundamentally different sources of legitimacy.
In managerial companies, 'the plan' carries legitimacy and the CEO is its advocate, so any pivot requires enormous activation energy. In founder-led companies, legitimacy sits in a person who can change direction on a dime if the founder is conscientious and open to new data. Tobi credits this structure with why Shopify's internal operations barely reacted to the 2021-22 stock swing.
founder-vs-managerial-leadership
Consensus is definitionally the absence of leadership and mathematically caps outcomes around a 6 or 7 out of 10.
Tobi argues that reaching consensus means everyone abdicated leadership to avoid downside, which structurally rules out both disasters and true 9s or 10s. He deliberately convenes committees or 'pricing councils' specifically when he wants a decision punted, since a consensus body can propose changes but rarely forces one through.
founder-vs-managerial-leadership
Great teams function like jazz bands: constrained but emergent, with leadership shifting to whoever has the most merit in the moment.
Drawing on time with jazz musicians, Tobi describes leadership as setting minimal shared constraints (key, tempo) and inviting the most skilled contributor to shape the next note, rather than dictating every part like an assembly line. He extends this to Shopify: his co-founder Daniel Weinand held veto power over shipped features when UX was the company's differentiator, which Tobi calls leadership, not consensus.
founder-vs-managerial-leadership
Shopify's post-2021 leadership reset replaced most of the executive team, not the strategy.
Tobi says nearly every executive besides president Harley Finkelstein turned over after the 2021 stock peak, because tenured leaders had optimized at becoming skilled at making their area sound rigorous to him rather than solving his actual current priority. COVID forced him back into direct, ground-level involvement across every project, which he says he now prefers to the more abstracted CEO role he'd drifted into.
founder-vs-managerial-leadership
A 15-year personal digital archive of keystrokes, screenshots, notes, and calendar data becomes far more valuable once AI can clean and query it.
Tobi has passively logged his active window and a screenshot every 10 minutes for about 15 years, originally just to see where his time went. He now uses AI to turn the fragmented, format-shifting logs into clean timelines that let him trace how his own beliefs changed, which he ties to the idea that the brain is 'a backwards-facing narrative consistency optimization device' rather than an accurate chronological record.
personal-data-archiving
The 2021-22 Shopify stock swing from 20x to 70x revenue and back never changed the company's fair market value; it only reflected shifting market conviction, and Tobi blames sell-side analysts for confirming rather than checking the bubble.
He separates stock price ('a betting market on fair value') from the business itself, noting the company's underlying value didn't move even as the multiple did. He's specifically critical of sell-side analysts who kept issuing buy ratings at 70x revenue, calling that the one professional class whose job is to warn the public rather than validate momentum. In hindsight, his one real regret is loosening hiring standards to scale headcount for what turned out to be a temporary e-commerce demand spike, not the valuation swing itself.
shopify-2021-meme-stock

Books referenced

Media referenced

Companies

Techniques and frameworks

Summary

This ACQ2 conversation opens as a philosophical detour and stays there: Ben and David set out to catch up on six years of Shopify news and instead get Tobi Lütke's real-time theory of how AI is changing computing, work, and leadership. Tobi frames the current moment as a platform shift so fast that society "blew by" the Turing Test without noticing, a callback to the sci-fi he grew up on (Snow Crash, Neuromancer) that always staged a dramatic AI-arrival moment history never actually delivered. His day-to-day response is disciplined: he runs a personal folder of AI "evals," treats new models as candidates to be interviewed for their capability edges, and has institutionalized this at Shopify with a company-wide mandate to use AI as a first pass on everything, a memo that was controversial when leaked and looks obvious in hindsight.

The through-line of the conversation is "context engineering," Tobi's term for the discipline of stating a problem with enough surrounding information that it becomes solvable without back-and-forth. He argues this is now the fundamental skill of working with AI, and that practicing it has made him better at email, at diagnosing workplace conflict as often being an unstated disagreement about what "good" looks like, and at building Shopify's internal "constitutions," documents (borrowing Anthropic's term) that capture only genuine, contestable convictions rather than platitudes, iteratively refined by feeding them back to AI to surface contradictions and gaps.

A second major thread is Tobi's framework for judging tools and abstractions: good ones raise the floor of achievable quality without lowering the ceiling, while bad ones (his examples: Xamarin, HTML5-hybrid mobile apps) trap teams in a permanently capped outcome no matter how skilled they become. He extends this into a theory of what AI is actually for: not replacing human judgment but shifting where people land relative to their own vision, illustrated with a Shopify merchant who used AI to instantly regenerate product photography instead of booking a shoot, and with the chess analogy, machines have outplayed humans for two decades, yet the world only cares about human players like Magnus Carlsen.

The conversation's most structurally interesting section is Tobi's distinction between founder-led and "managerially run" companies. In his framing, managerial companies vest legitimacy in "the plan," making the CEO an advocate who needs enormous activation energy to change course, while founder-led companies vest legitimacy in a person who can pivot immediately given new data. He extends the category to non-founders like Satya Nadella and Jamie Dimon who delivered "refounding events" dramatic enough to author a new company culture. This leads into his sharpest claim: consensus is definitionally the absence of leadership, structurally capping outcomes around a 6 or 7 out of 10, which is why he deliberately convenes committees when he wants a decision slow-walked, and why he describes great teams as functioning like jazz bands, shared constraints plus emergent, merit-based leadership, rather than assembly lines.

On Shopify's own history, Tobi is candid that the 2021 stock run to a 70x revenue multiple, and the subsequent 80% drawdown, never actually moved the company's underlying fair value; he faults sell-side analysts for confirming the bubble with buy ratings rather than warning investors, and pins his one real regret on loosening hiring standards to scale headcount for what turned out to be a temporary pandemic-driven demand spike. He also reveals that he's passively logged his own screen activity every ten minutes for fifteen years, an archive now newly valuable because AI can clean and query it, letting him trace how his own convictions (on leadership, on consensus, on delegation) actually shifted over time rather than trusting his memory's tendency to narrate a falsely consistent story.

Notable Quotes

"We are software that no one wrote and I have to interview it to figure out its capability." - Tobi Lütke

"The computer is to get people to build things where they could not possibly imagine having solved otherwise... Computers are there to make humans better, never to constrain what humans can do." - Tobi Lütke

"Consensus is always the absence of leadership." - Tobi Lütke

"I think your goal on planet Earth is to minimize the difference between [the person you could have become] and who you actually ended up being." - Tobi Lütke