Ritavan on System Gambits and Finding Leverage

Ritavan argues that the investor’s favourite question — ‘does this company have a moat?’ — asks the wrong thing. A moat is a static tick-box abstracted from a fixed picture of the world; what actually protects a business is whether it runs a system that compounds faster than rivals can match. His framework, the System Gambit, names the move that builds such a system: sacrifice your standing in the game you are winning to cross into a new game whose structure compounds in your favour. Co-hosted by investor Kai Wu of Sparkline Capital, whose work on intangible assets runs alongside the argument.

Key ideas

1. A moat is an artifact; a compounding system is the thing. Ritavan rejects the moat checklist because a checklist is a list of binary items abstracted out from a fixed causal model of how the world works. Change the rules of the game and the checklist is useless — it describes the old game. The defensibility question is not ‘how thick is the wall?’ but ‘does this business compound in a way and at a rate a rival simply cannot?’

2. The three conditions of a system gambit. A move qualifies only when all three hold at once: a self-improving loop (every iteration structurally gets better, not just bigger); path dependence (it must be built loop by loop, so no rival can buy it on the market with more capital); and management logic antagonism (to copy you, a rival would have to break what currently makes them money). Miss one and you have a nice asset or a wild bet — not a gambit.

3. The sacrifice is watching your own metrics collapse. Crossing systems hurts because your performance in the old system — the one you can still measure — visibly falls. Skanderbeg abandoned his fortress to a small garrison and harried the besieging Ottoman army from the hills; it looked like losing right up until he won the war. Amazon ‘sacrificed’ a decade of EBIT to build its flywheels.

4. Adoption is meaningless; closing the loop is everything. The microscope and the telescope are the same technology — two lenses in a tube — yet the microscope ‘hallucinated’ for ~200 years while the telescope revolutionised astronomy overnight. The difference was a testable causal model to iterate against. The lesson for AI: you can drive adoption to 100%, but value comes only from pointing the tool at a bottleneck where you can close the loop with high-quality signal.

5. Within-paradigm AI optimises you into irrelevance. Because your data is generated by your existing business, bolting AI onto it trains you deeper into the game you already play — and anyone with a credit card and an API key can copy that. Edge comes from paradigm change, not incremental optimisation. Ritavan calls the purely defensive, follow-the-crowd move the system anti-gambit — rational, but not where advantage is won.

Content

The moat as tick-box, and why it fails

Ritavan’s quarrel is not with moats but with treating a moat as a checklist entry. The image — measuring the thickness of a fortress wall — implies a fixed world in which the item can be verified once and ticked. He traces the investing checklist to Mohnish Pabrai’s adoption of Atul Gawande’s The Checklist Manifesto, and draws the crucial boundary: checklists work where the underlying causal model is fixed. A surgeon’s checklist holds because human physiology does not change with a social-media post; a pilot’s holds because aerodynamics is stable. A checklist is an artifact abstracted from that stable model. When the model itself can change — as it can in business and markets — the artifact stops being reliable, and the discipline should be to drop a level and reason about the system, not the tick-box.

Kai Wu maps this onto his own line that ‘code is not the moat’ for software companies: whether code is a moat depends entirely on the game. When only a few could write it, code was scarce and defensible; once AI writes it cheaply, the game changes and the moat evaporates. The investor’s job becomes second-order — identify which system is now the dominant paradigm, ask what is scarce in that system, then find companies whose assets are priced for the old paradigm but valuable in the new one (his example: Getty’s re-rating once proprietary image data became AI-training input).

Skanderbeg: inverting the paradigm

The historical spine of the argument is Gjergj Kastrioti — Skanderbeg — the Albanian commander trained by the Ottomans who broke away to fight them. Against a vastly larger army, defending his fortress by the checklist (tall walls, deep moat, outlast the siege) led to one predictable outcome: a successful Ottoman siege. Skanderbeg instead left a token garrison, slipped away before the army arrived, and harassed the besiegers night after night from the surrounding hills. The same fortress, the same armies — but by changing the causal model of the fight, he turned the enemy’s marshalled mass from a strength into a fixed, vulnerable target. To the Ottomans he looked like a coward who ran; he did not mind looking like a loser short-term if it won the war. David with the sling looks foolish, too — right up until the asymmetry pays.

Microscope versus telescope: the AI lesson

Ritavan’s sharpest metaphor for AI is optical. The microscope and the telescope are essentially the same invention. Robert Hooke and Antonie van Leeuwenhoek pointed theirs at biological samples and produced, for roughly 150–200 years, ‘visibility without understanding’ — the same sample viewed twice looked different, because lighting and preparation were unstandardised and, more deeply, there was no causal model of physiology to interpret what was seen. Galileo pointed the same technology at the sky and revolutionised astronomy within months, because astronomy had a testable causal model (and a rich data set from Tycho Brahe for Kepler to fit) that could be iterated against hard, shared feedback: Galileo in Italy and an astronomer in Paris saw the same moon.

The transfer to AI is direct. Generative models are stochastic — the same prompt can return different answers, microscope-like. Value does not come from the tool’s raw capability or from adoption metrics (‘AI adoption’ as a dashboard KPI is, he argues, meaningless — you can be at 100% adoption of a microscope that reveals nothing). It comes from pointing the tool at a bottleneck where the loop closes with high-quality signal. Kai Wu adds the complement: what made Claude Code a breakthrough was less the model than the harness — file access, tool calls, terminal commands — and the setting: software is a closed loop (does it compile?) with codified knowledge, an ideal place for the tool, unlike, say, therapy. Both invoke Goodhart’s law against the brief ‘token-maxing’ fashion of measuring AI success by spend.

Nokia versus ASML: agility without a causal model

Ritavan reads Nokia and ASML from the outside as a study in causal models. Nokia was superb at speed and agility — it saw the smartphone coming, logged and documented it, and tried to react fast. But reacting fast to others’ moves without a causal model of your own game is aping, not strategy: copy the Japanese player, copy the iPhone, and you are neither playing basketball nor soccer. ASML, by contrast, looks like a maker of extraordinary ~$400m machines, but its real, inimitable capability is the institutional causal model of how a specific machine performs in a specific fab, on a specific design, under specific conditions — knowledge held by engineers embedded on the ground (a forward-deployed model ASML ran decades before Palantir made it a story). That understanding of the game cannot be put on a balance sheet, which is precisely why few are incentivised to build it and why it is where the edge lies.

Amazon and Walmart: multi-paradigm compounding

Amazon is, for Ritavan, the greatest system gambit of the era — and it ran across three paradigms at once. It built an industrial warehousing-and-logistics business (predictability, physical footprint), a digital storefront (learning loops on user data, dynamic pricing, recommendations), and then crossed them into reinforcing feedback loops (understanding demand lets it shape inventory and price). Fulfilled-by-Amazon added the platform paradigm — opening the infrastructure to third-party sellers over whom Amazon holds leverage — and Prime tied all three together. Sacrificing ~5% EBIT for a decade to build this was, in his words, trivial against the value created; Jim Collins’s flywheel is simply a compounding loop.

Walmart’s counter is the instructive part. For years it aped Amazon (better CRM, some digital tooling) — the defensive anti-gambit. The turn came when it leaned on what Amazon structurally cannot copy: physical presence and generational, in-store customer relationships. That is management logic antagonism (Amazon opening physical stores would break its own model) plus path dependence (trust built over decades). Fusing online and physical customer understanding, plus a loyalty programme, Walmart built its own multi-paradigm compounding machine. The lesson: it does not matter where you start — pre-dot-com startup or half-century-old family firm — what matters is reasoning from first principles about the system and leveraging your own asymmetric strengths, not the crowd’s.

The investor’s one takeaway

Asked to distil the book to a single lesson for investors, Ritavan answers with a question: what game are you choosing to play? At any level of assets or size, the opportunity is to pick a different game — different assets, a different time axis (playing beyond the quarter everyone else optimises) — where your own asymmetry compounds. The book offers eight such gambits (the first: paradigm change and goal displacement); he likens them to a set of exercises whose reps train you to spot the pattern. For the investor watching from outside, the discriminating question is: who is merely showing a drop in performance and a value trap, and who is quietly building a structural loop that will compound?

Related

• Ritavan — guest; author of The System Gambit
• Kai Wu — co-host; quantitative investor, Sparkline Capital, on intangible assets
• Matt Zeigler — host
• System Gambit — the framework the episode introduces
• The System Gambit — the book
• 7 Powers — Helmer’s structural theory of advantage, which the System Gambit both extends and challenges
• Compounding — the loop the gambit is built to create
• What Makes a Great Investor — theme this episode feeds