First Principles Thinking

First principles thinking is reasoning from foundational truths rather than from inherited assumptions or analogies. Tobi Lütke’s formulation is more specific than the common gloss: it is the practice of re-evaluating the function a thing is meant to serve, then deriving what that thing should look like from scratch — rather than accepting the current implementation as a given and incrementally improving it.

The core model: function re-evaluation over assumption stacks

Every existing solution embeds a stack of assumptions, many of which were correct at the moment they were made but have since been superseded by changed circumstances. The longer a solution has existed, the more assumption layers have accumulated — and the more likely it is that some layer has silently become false.

First principles thinking, as Tobi defines it, is the practice of asking: what is this thing actually for? Then: given what we now know, what would we build if we started from that function?

This is distinct from analogical reasoning (“what does company X do in this situation?”) and from incremental improvement (“how do we make the current thing 10% better?”). Both of those start from the existing implementation and iterate. First principles starts from the function and re-derives.

Path dependence as the target

The primary adversary is path dependence: the tendency for current solutions to reflect the constraints of when they were designed, even when those constraints have dissolved.

Tobi’s worked example is remote work at Shopify. The conventional logic for why companies should have offices goes:

Collaboration requires synchronous communication.
Synchronous communication works best face-to-face.
Therefore, people should share a physical space.

Each step was a reasonable approximation in a world without high-bandwidth video, persistent messaging, async documentation culture, and distributed tooling. By 2020, every assumption in the chain had weakened considerably. The question was not “should we do remote work?” but “if we were designing a company today for the actual job of coordinating human work, what physical infrastructure would we specify?”

Shopify’s answer: zero mandatory co-location. The decision was not a response to COVID forcing the issue — Tobi describes it as a conclusion already reached, which COVID simply accelerated.

The Boolean flip pattern: when a constraint that has been false for years is still treated as operative, the right move is not to nudge the constraint — it is to flip it entirely and re-derive from there. Incremental remote-work policies (two days a week, hybrid mandates) fail because they treat the office as the default and remote as the exception. A first-principles view flips the Boolean: no office, then adds back physical co-location where it genuinely adds function.

Goodhart’s Law as overfitting

Tobi maps Goodhart’s Law (“when a measure becomes a target, it ceases to be a good measure”) onto the language of machine learning: optimising a metric is equivalent to overfitting on a training signal. The metric was a proxy for a function; optimising the proxy drifts away from the function.

The first-principles corrective is to keep re-asking what the underlying function is, rather than continuing to optimise the current proxy. Metrics are inevitably approximations of goals; the discipline is to notice when the approximation has become the goal.

Positional vs. tactical game

Borrowed from chess: positional play builds structural advantages whose payoff comes in moves 20–30; tactical play executes on the current position. Most businesses default to tactical play because it is immediately legible (wins are visible now) and because positional advantages require sitting with uncertainty while the position develops.

First principles thinking is positional by nature: it requires accepting a period where the re-derived solution looks worse than the incremental-improvement alternative, because the re-derived solution is building toward a structurally better position. Organisations that lack the ability to tolerate that period will systematically under-invest in first-principles work.

Simplification as first-principles application

Tobi’s argument that simplifying software is a moral obligation — not merely a design preference — is a first-principles conclusion:

The job of software serving small businesses is to let those businesses run.
Complexity imposed on small businesses consumes management bandwidth that the business cannot afford.
Therefore, complexity that does not serve the user’s job is actively destructive, not just inelegant.

The growth strategy implication: simplification is also a market-expansion strategy. Every unit of complexity removed makes the product accessible to a customer who previously found it too hard. This converts what looks like a product-quality decision into a demand-generation decision.

Dissatisfaction as first-principles energy source

Tobi identifies dissatisfaction — not satisfaction, not positive vision — as the emotional foundation for first-principles work. The ability to look at something that everyone else considers fine, feel genuine dissatisfaction with it, and act on that dissatisfaction is what drives the re-derivation process.

This is distinct from contrarianism (disagreeing for the sake of disagreeing) and from negativity (dwelling on what is wrong). It is a directed dissatisfaction: the feeling that something could be categorically better, paired with the motivation to find out what that would look like.

Where mainstream views differ

The popular formulation of first principles thinking — associated with Elon Musk via the “physics approach” framing — emphasises breaking problems into fundamental truths and building up from there. Tobi’s formulation is narrower and more operational: it centres specifically on path dependence and function re-evaluation, rather than first principles as a general epistemological stance.

The distinction matters in practice. Musk’s formulation can become a style of argument (“I’m reasoning from first principles, you’re not”). Tobi’s formulation is a diagnostic: find the assumption in this solution that has silently become false, then re-derive.