Lessons of a Physicist Building a Company

Reality is messy, noisy and not sterile

I went from physics into startups, along the way I had to learn many hard lessons and fell on my nose more than once. Here are lessons I learned along the way.

Physics is a beautiful abstraction. It's a science that combines a reductionist view and first principles thinking. It is about understanding the world bottom up. Experiments are designed to be independent of the observer, and as clean of noise as possible. Physics is about understanding. StartUps however are a discipline about Optimization along several dimensions. The playing field is inherently messy, large parts of the environement are outside of your control, the rules are constantly changing and everything is in constant movement. You are not an observer but a player - one of many - and as such you change the outcome.

- Physics is like Chess, StartUps like Poker -

While in physics you can plan concrete steps to reach a conduct an experiment and get a definite outcome, you cannot do this in startups. Instead of planning elaborate steps to guarantee a specific outcome, you want to optimize for maximizing your chances of the desired outcome - success -.

Each Startup is a different game, but the currencies for success remain the same - treat each as a game that needs to be optimized for:

• Game 1 (Product): Users want value.
• Game 2 (Distribution): The Product needs to be distributed to the users.
• Game 3 (Talent): Employees want equity + meaning.
• Game 4 (Capital): VCs want unicorns, not profits.
• Game 5 (Regulation): Regulators want stability.
• Game 6 (You): You want existential purpose + wealth.

The system's output equates to the Nash equilibrium of all these games pulling simultaneously. Your startup doesn't "solve" a problem; it renegotiates the equilibrium.

What this means: Your pitch isn't a physics talk. It's a credible signal that you've hacked the equilibrium: "If you fund us, we'll poach 3 key engineers from Google (Game 3), triggering their M&A team to overbid (Game 1/4), forcing the EU to regulate fast (Game 5) - and you'll get your 1000x."

Lesson 2: From mean to Power Laws

This one is pretty much straight out of Thiels book "Zero to One". You're used to Gaussian errors and ensemble averages. The business world is Pareto-distributed and non-ergodic.

• Your first 10 customers are not a sample of the next 10,000. They're a different regime - early adopters who obey different psychological laws - see "Crossing the Chasm" by Geoffrey A. Moore. Interpolating is fatal.
• VC returns follow power laws: 1 bet returns the fund, 9 die. You can't "diversify" your startup's strategy to reduce variance, but you want to maximize convexity (optionality). That means doing things that have small, fixed downside and unbounded upside. Like shipping a scrappy MVP to a niche, not building a perfect product for "the market". However with the recent emergence of vibe coding voices have become louder to ship fast - faster than ever.
• Talent is power-law distributed: One engineer = 10x another. Your hiring "system" is not a pipeline - it's a search for outliers.

Takeaway: Replace every dashboard average (avg. revenue/user, avg. churn) with a Lorenz curve (who generates what % of revenue). You'll see your "system" is 3 whales and 10,000 free riders. Design for them, not the average.

Lesson 3: Stocks and R-loops vs B-loops

Each of these Games are governed by a stock described by Meadows' stock-flow dynamics.

• Founder morale (depletes with every "no")
• Team psychological safety (erodes with each pivot)
• Investor confidence (depreciates quarterly)

Key insight: Morale is a stock with no equilibrium. It decays unless actively replenished. Your "runway" is not just cash - it's morale converted to dollars.

Feedback loops to master:

• R-loop (Reinforcing): Ship feature → user love → team morale ↑ → ship faster. This is the real engine.
• B-loop (Balancing): Hire fast → culture dilutes → decision speed ↓ → firefight → hire more. This is death.

Your job: Architect one R-loop and cut every B-loop at the root. Everything else is noise.

Lesson 4: The Boundary is Your Moat (and Your Prison)

In startups, the boundary is your competitive moat.

But here's the trap: You get to choose the boundary, but the market gets to move it.

• You define "our market" as mid-market SaaS.
• A competitor offers a free tier → they redefine the boundary to include individuals.
• Your 10-person sales team is now irrelevant.

Lesson: Define your system boundary not by customer segment, but by feedback speed. The startup that can close the loop fastest (learn-build-ship) defines the category. Everyone else is playing catch-up. If improvement becomes exponential, the one with the fastest feedback loop wins.

How: Instrument your startup's OODA loop speed (Observe-Orient-Decide-Act). If it takes you 2 weeks to ship a user request and a competitor takes 2 days, your boundary is dissolving.

Lesson 5: Reflexivity - The System is a Mirror

This one is contrary to physics experiments: The observer is part of the system, and the system optimizes around you. It becomes a mirror of your actions and decisions. - Utilize it!

• You raise a big Series A → you're now expected to hire 30 people → you dilute culture → you ship slower → you miss milestones → you raise a down round. The capital created the failure mode.
• You announce AI features → customers expect magic → you overpromise → you ship buggy code → trust collapses. Your marketing was the product loop.

Founder-level systems thinking: Every action is not a "move" but a signal that rewrites the rulebook. You are not in a game; you are negotiating the game's existence.

The skill: Second order thinking. Not "What happens if I do X?" but "What happens to the incentive structure after I do X, and how will actors re-optimize?"

Progression of the Toolkit to master for clarity

• Causal Loop Diagrams → Incentive Diagrams: Map how each stakeholder's reward function creates loops. Identify who wins when you lose.
• Monte Carlo → Scenario Planning: But scenarios aren't probability-weighted. They're possibility-weighted. Assign capital to the scenario that doesn't kill you, not the one that's most likely.
• Lagrangian → Business Model Canvas: The canvas isn't a plan. It's a coupled differential equation between value prop (potential) and channel (kinetic). Derive the fixed points.
• Blue-Green Deployments → Pre-Mortems: For every decision, simulate the systemic failure mode: "If this works, which R-loop does it break?"

Things I had to unlearn

• Precision → Speed: A 90% correct decision shipped today beats a 99% decision next month. The system evolves too fast for p-values.
• Truth → Legitimacy: You're not seeking a discovery; you're seeking permission to exist from users, investors, and talent. Legitimacy is the stock you actually trade.
• Objectivity → Positionality: Your VCs, employees, and competitors are not noise. They're co-authors of your reality. You must choose your position (e.g., "we're the privacy-preserving AI") and defend it as a boundary condition.

Final Thoughts

Pick a startup idea. Before writing code, draw a single causal loop diagram with:

• 3 stocks (e.g., user trust, team morale, investor confidence)
• 2 R-loops (one you want, one you fear)
• 1 B-loop that will kill you in 18 months

If you can't map it, you don't understand the system enough to survive in it.

Physics teaches the universe's indifference. Startups teach you the universe is adversarially indifferent, it wants to kill you, but only if you let it define the rules.

Your move.