✶Explainer14:00
Goodhart's Law Is Just Overfitting for Businesses
Tobi connects machine-learning overfitting to Goodhart's law: a model learns to cheat the benchmark just as a business degrades the moment a metric becomes a goal. He uses this to explain why schools optimize for marks (and why kids cheat) and why he distrusts single-metric optimization. Different fields reinvent the same enduring truth under different names.
- Overfitting is a model learning to cheat the loss function, which only proxies the real task
- Goodhart's law is literally the same thing applied to businesses
- Any metric that becomes a goal ceases to be a good metric
- Schools overfit to marks; the same universal truths recur across competitive fields under different names
“goodarts law just says any metric that um becomes a goal ceases to be a a good metric”
“overfitting uh goodarts law are the same thing”
#metrics#machine-learning#goodharts-law#strategy
✶Explainer29:00
Derive Every Product From First Principles
Tobi says every existing product is deeply path-dependent, encoding compromises that were true when it was built but no longer are. So a new product must be re-derived from the building blocks available right now, not copied from how others do it. Skipping that exercise and doing what everyone else does is, to him, an abdication of product leadership.
- Everything that exists is highly path-dependent on assumptions that may no longer hold
- Derive the solution from every building block available today, then decide whether to take shortcuts
- Sometimes the status quo turns out to encode real wisdom, and you adopt it deliberately
- Doing something world-class requires doing it differently; it's near-axiomatic
“you have to derive it from first principles you have to say how would we solve this problem given every fun like building block that…”
“if you want to do something better than what exists you have to do it differently”
#first-principles#product#innovation#strategy
✶Explainer1:05:30
Positive-Sum Games Are the Ultimate Growth Hack
Tobi frames business partnerships as an iterated prisoner's dilemma where coordinating over long horizons beats any momentary defection. He compares most companies to kids failing the marshmallow test, pulling future profits forward at a discount. Shopify's model is to add levers of monetization but not pull them, staying on the same side of the table as its customers.
- The Stripe-Shopify partnership is iterated prisoner's dilemma played as coordinate, not defect
- Most CEOs act like kids who grab the single marshmallow instead of waiting for two
- His job is adding monetization levers to the room and then not pulling any of them
- On a long enough timeline, positive-sum games with customers are the ultimate growth hack
“being a good partner in business is like this sort of corporate marshmallow test”
“on a long enough timeline playing positive some games with your customers is the ultimate growth hack”
#partnerships#long-term#prisoners-dilemma#monetization
✶Explainer1:10:00
Business Is a Positional Game, Not Just Tactics
Using chess, Tobi splits business into a positional game (territory, trust, role) and tactics (conversion, A/B tests). He argues the business world obsesses over tactics while the real, durable value lives in the positional game. Over-extracting via tactics drains the value you built positionally, which is how once-great companies fade.
- Chess has two semi-independent games, positional and tactical; business only talks about tactics
- You need enough tactical skill to not get margin-called, but that's not where the value is
- The positional game is territory, trust, and how deeply merchants rely on you
- Extracting the whole value through tactics leaves nothing in the tank and companies fade
“the business world only talks about tactics it only talks about the conversion”
“if you do that you have nothing left in the tank and that's the companies that we all see that just sort of got to…”
#strategy#chess#tactics#positioning
✶Explainer1:19:30
Every Time Software Gets Simpler, More Businesses Exist
Tobi argues that great UX taming complexity is almost a moral obligation, because bad software makes people feel dumb and can stop a fragile new entrepreneur in their tracks. Shopify's data shows that every time they make something simpler, more businesses exist on the platform. Autopiloting taxes, payments and fraud literally causes more entrepreneurship, which he calls the best answer to the most important question of his life.
- Bad software makes people feel dumb, which he sees as an inversion of the human-tool hierarchy
- Entrepreneurs are often unsophisticated, scared, and surrounded by naysayers, so one confusing moment can make them quit
- Data shows every simplification leads to more businesses existing on the platform
- Removing complexity (taxes, payments, fraud) actively causes more entrepreneurship
“software goes bad it makes people feel dumb and machines do not get to negatively influence people”
“engaging in building something again is an is is an act of pure courage”
#ux#entrepreneurship#simplicity#product