Notes: Tobi Lütke on First Principles Thinking
Four questions [Adler frame]
Q1. What is it about? Two intertwined theses: (1) first principles thinking as a practice of re-deriving solutions from foundational assumptions rather than accepting path-dependent status quo; (2) every person has far more unrealised potential than they believe, and a leader’s job is to hold people to that potential rather than their current level. Shopify’s product philosophy — simplify as moral obligation and growth strategy — falls out from both theses combined.
Q2. How is it argued? Primarily through analogy and worked examples rather than formal argument. The analogies are cross-domain (machine learning/overfitting as Goodhart’s Law; chess positional vs. tactical play; thermostat as product-management metaphor; Michael Jordan’s energy source as a template for identifying one’s own). Worked examples: Shopify’s remote-work decision, Shopify’s original product vision vs. incumbent e-commerce complexity, monorepo architecture decision. Tobi is self-aware that his reasoning “runs in a meta-language that isn’t directly something I can translate into words” — the interview is partly an attempt to reverse-engineer his own algorithm.
Q3. Is it true? The path-dependence observation is solidly argued: most existing solutions do embed assumptions that were valid at time of design and have since become false; the discipline of asking “would I make the same decision today with current inputs?” is a legitimate corrective. The Boolean-flip model (not just nudging a constraint, but flipping it entirely when the underlying assumption has inverted) is a useful heuristic, though it risks over-application — sometimes constraints are partially weakened rather than fully inverted, and a full flip overreaches. The Goodhart’s Law = overfitting equivalence is accurate and elegant. The maximising-human-potential thesis is aspirational rather than argumentative; Tobi doesn’t offer a mechanism, just a conviction.
Q4. What of it? For product leaders: treat complexity as an active harm (not just inelegance), because it kills businesses before they start. For founders: identify the load-bearing assumptions in incumbent solutions, check whether they are still valid, and re-derive only from assumptions that hold. For leaders: hold people to their potential, not their current level — and be harder on yourself than on others. For long-term strategy: positional game (territory, trust, relationship depth) compounds; tactical game (conversion optimisation, pricing moves) extracts; the former funds the latter.
Glossary
Path dependence — the property of existing solutions whereby their current form reflects the constraints operative at the time they were designed, even when those constraints no longer hold. Tobi identifies path dependence as the primary adversary of first principles thinking.
First principles thinking (Tobi’s formulation) — re-evaluating the function a solution is meant to serve, then re-deriving what the solution should look like given current building blocks. Distinct from incremental improvement (starts from current implementation) and from analogical reasoning (starts from what competitors do).
Boolean flip — when a foundational assumption in a solution stack changes from true to false (or vice versa), the correct response is not to nudge the dependent decision — it is to re-run the entire derivation tree, which may land in a very different place. Remote-work decision is the canonical example: when “are people allowed to leave a house?” flipped from yes to no, the entire office-strategy tree needed re-running, not just adjusting.
Overfitting / Goodhart’s Law — Tobi treats these as the same phenomenon in different domains. Overfitting: a model learns to optimise the training proxy rather than the actual task. Goodhart’s Law: a metric that becomes a target ceases to be a good metric. Both describe the failure mode of optimising a proxy that diverges from the underlying goal.
Local maxima — a position reached by a series of individually good decisions that nonetheless leads to a suboptimal overall outcome. Tobi: “Most of the time you end up in a bad part of a tree, in a local maxima of a path-dependent environment, by only making good choices.”
Positional game — in chess: building structural advantages whose value accrues over many moves. In business: accumulating trust, supply, demand, and product territory whose value compounds over years. Contrasted with tactical play.
Tactical game — in chess: immediate move-to-move execution. In business: conversion optimisation, pricing changes, A/B tests. Necessary to survive; insufficient to compound. Danger: extracting via tactics the full value built through positional play, leaving nothing in the tank.
Tobi Tornado — rapid, compressed change management: Tobi observes a problem, has a direct conversation, and either updates his priors or cancels/redirects the project immediately. Framed as respect for finite career time and fairness to people working on projects that won’t succeed.
Energy source — Tobi’s frame for understanding what drives a builder. His own: dissatisfaction with status quo. Michael Jordan’s (observed in The Last Dance): rivalry and manufactured grievance. The concept is that building companies requires someone to “inject heat” into the system — room-temperature organisations cannot outperform anyone.
Exothermic — Tobi’s characterisation of the best leaders: they are wellsprings of energy that drive others, rather than drawing energy from the organisation.
Simplification as moral obligation — software that is unnecessarily complex makes users feel inadequate, inverting the proper relationship between tools and people. Beyond the ethical argument: every unit of complexity removed expands the addressable market (more potential entrepreneurs can reach launch), making simplification a growth strategy.
100-year vision — planning horizon used to anchor long-term strategic decisions. At 100 years, the specific product cannot be specified, but the mission can. Used to enforce a “coordinate, don’t defect” posture in partnerships (iterated prisoner’s dilemma framing).
First principles as function re-evaluation [§ First Principles Thinking]
Tobi’s operational description of first principles thinking:
“A perfect product lead is almost like a thermostat for high quality product. You re-derive literally every decision that is valuable, every foundational assumption, every foundational direction. You want to see the observation you’ve made in the meantime since you last derived the next step. Re-running the entire function over the state that is now updated, the higher fidelity information, would you come to the very same thing?”
The key phrase: re-running the function over updated state. This is not a one-time derivation but a periodic re-check. The trigger for re-running is new information that updates the input state — not time, not a scheduled review, but a change in foundational conditions.
The failure mode this corrects: skipping the re-derivation exercise and defaulting to what incumbents do. Tobi treats this as “an abdication of product leadership,” not just strategic laziness.
Remote work as worked example [§ Boolean Flip]
The remote-work decision is the fullest worked example Tobi gives of Boolean-flip reasoning:
- The standard logic for in-person work: collaboration requires synchrony → synchrony works best face-to-face → shared physical space is optimal.
- Each assumption was reasonable in 2004 (Shopify’s founding year). By 2020, each had weakened: async tooling, video infrastructure, distributed work culture had all advanced.
- Additional inputs fraying the decision independently: Shopify was already spread across 4–5 cities; Tobi was spending full days in back-to-back video calls from the Ottawa office, with no co-located attendees; Ottawa’s talent density (1M people) was insufficient for a company scaling to 10,000+.
- COVID flipped the Boolean: “are people allowed to leave a house?” changed from yes to no.
- Re-running the derivation tree from that new input landed in a completely different place: not “allow more remote work” but “become a remote-native company, porting the internet into the company rather than porting the office online.”
Tobi’s observation on the danger of local maxima: “Most of the time you end up in a bad part of a tree, in a local maxima of a path-dependent environment, by only making good choices. People think that making a good choice inoculates you from making mistakes — both of those things are incorrect.”
Goodhart’s Law as anti-OKR argument [§ Metrics and Taste]
Shopify operates without traditional OKRs. Tobi’s argument:
- No metric is a complete heuristic for a complex business: businesses have millions of tensions, and optimising one metric drives trade-offs that damage other dimensions.
- OKR systems cause promotion decisions to be tied to driving specific metrics, which causes people to optimise the metric rather than the underlying goal (overfitting).
- Shopify compensates with extremely sophisticated data infrastructure (“I can dig into every constituent atomic bit part that makes up a cohort formed 15 minutes ago”) used for information, not for target-setting.
- The unquantifiable (craft satisfaction, delight, love, hate) overlaps with the most valuable product decisions by approximately 80%. Excluding that 80% is not neutral — it systematically deprioritises the highest-value space.
- “The most powerful, not unquantifiable things in the world of business are fun and delight. If all metrics are pointing down but everyone says, ‘My God, I’m having so much more fun,’ the very next thing that will happen is all metrics will start going up.”
Simplification as moral obligation and growth strategy [§ Complexity and Entrepreneurship]
Tobi’s argument that simplification is both an ethical obligation and a market-expansion strategy:
Ethical argument: software that is unnecessarily complex makes users “feel dumb.” Machines are tools that should amplify human capability. Complexity that does not serve the user inverts this relationship.
Growth argument (empirically grounded): “Every single time we make a complex thing simpler, more businesses will exist on the platform.” The mechanism:
- Potential entrepreneurs are in a fragile psychological state — facing nay-saying, financial pressure, uncertainty.
- Any point where software creates confusion or stops forward progress is a potential exit point: “They close the browser and say, ‘I’m not cut out for this.’”
- These exit points do not filter out bad businesses — they filter out good businesses that were close to the threshold many times.
- Removing complexity removes exit points → more businesses cross the threshold → platform grows.
The example given: tax configuration. Instead of a tax configuration screen, Shopify simply handles taxes correctly. The user never encounters the complexity.
Positional vs. tactical game [§ Chess Analogy]
The chess analogy has two parts:
Positional game: What territory are you accumulating? What trust do merchants have in you? Are you the platform they want to add capability to, or the one they want to optimise out? This is long-run, non-quantifiable, the source of durable competitive advantage.
Tactical game: Conversion rates, A/B tests, pricing changes. Immediately legible, immediately rewarding. Necessary to survive (don’t get “margin called”) but insufficient to compound.
The failure mode for companies that over-index on tactics: “they extract through tactics the entirety of the value that they have created and that is theirs to take through a positional game. If you do that, you have nothing left in the tank.”
The 100-year vision frame enforces the positional posture: “Positive-sum games have incredible returns, especially in the worlds of software, where you can experience exponentials very easily. On a long enough timeline, playing positive-sum games with your customers is the ultimate growth hack.”