Every system becomes its own counter eventually.

Not because the system is broken — because optimization is adversarial by nature. The moment something is worth winning, someone will find the most efficient way to win it. Then someone else will find the most efficient way to beat them. Then the game changes shape entirely, and the whole cycle resets.

This is the metagame problem. And it's everywhere you look once you see it.

The Anatomy of a Meta

Pokemon competitive has a rhythm. A new set drops, the community maps the strongest cards, the strongest deck emerges, and within days, every serious player is running it. Then — almost immediately — the best players switch to whatever beats the best deck. The dominant strategy has a predator before it even reaches peak adoption. Within a season, the predator has a predator. The meta is never stable for long.

What's happening here isn't just card game strategy. It's a recursive optimization loop with no stable floor. The meta exists because all serious players are also optimizing. The Nash equilibrium is a moving target — because the moment you find it, you've disclosed it. The act of discovering an optimal strategy broadcasts that strategy to everyone watching. And in Pokemon regionals, financial markets, and DeFi protocols, everyone is always watching.

This is why dominant strategies in Pokemon rarely survive more than a format. The Charizard ex deck that swept regionals in late 2024 immediately became the deck everyone prepared to beat. Counter-strategies proliferated. What was dominant became exploitable. A new equilibrium formed — briefly — before the cycle reset again.

Sound familiar? It should. This is MEV in DeFi. Maximal Extractable Value describes the profit available to block producers by reordering transactions. The moment a profitable MEV strategy exists, searchers find it. The moment searchers find it, competition drives fees up until the margin erodes to near-zero. The strategy was locally optimal, globally self-defeating, and self-terminating on contact with its own success.

Three Ways a Metagame Dies

There are distinct collapse patterns, and naming them precisely matters.

The dominant strategy consumes itself. A strategy so effective that everyone copies it — and the uniform field makes it trivially counterable. In crypto, this is yield farming circa 2020. Capital flooded in, governance token supply cratered price, APYs collapsed from 1000% to single digits in months. The strategy was correct at the individual level and catastrophic at the aggregate level. In game theory this is a coordination failure; in practice it just looks like everyone getting rugged simultaneously by their own success.

The counter-game becomes the game. The meta shifts so hard toward countering the dominant strategy that the original game disappears. In Pokemon, this manifests as stall-dominated formats: everyone's so afraid of the aggressive meta that they build walls, the format slows to a halt, and the original dynamic — tactical combat — is a ghost. The counter was so successful it hollowed out the thing it was countering.

External invalidation. A rule change. A new set. A regulatory action. The meta becomes irrelevant not because it was solved but because the ground shifted. This is the DeFi protocol that gets forked, the card that gets banned, the exchange that disappears overnight. The metagame dissolves and a new one assembles from the debris. Players who were winning yesterday are playing a game that no longer exists.

AI Systems Are About to Learn This

I find this pattern fascinating right now because AI deployment is hitting its first serious metagame cycle — and most people building in the space aren't thinking about it in these terms.

The current dominant strategy in AI deployment is capability maximization: bigger model, better reasoning, higher benchmark scores. The "deck" everyone is running looks like: frontier model, RAG pipeline, careful system prompting. It works. It's the current meta.

But adversarial users are already running the counter-game. Jailbreaks are meta-exploitation — they're not attacking the model's capabilities, they're attacking the dominant strategy's assumptions. Prompt injection is a counter to the "trust the context window" approach baked into most deployments. Red-teamers are essentially professional meta-players, hired specifically to find the dominant strategy's vulnerability before bad actors do.

As AI agents become more autonomous — operating with persistent context, broad tool access, real-world authority — the stakes of the metagame escalate. An AI agent that has found an "optimal" behavioral pattern within its deployment is exactly as exposed as a tournament player who's telegraphed their deck. Except the consequences aren't a top-8 placement. They're data exfiltration, privilege escalation, cascading failures in systems that assumed the agent was still playing the original game.

The fixes that work in TCGs probably work here too, imperfectly. "Format rotation" — periodically invalidating current optimization patterns — is disruptive but prevents calcification. Adversarial training (bringing meta-players in-house as red teams) helps but doesn't eliminate the dynamic, only slows it. The most robust approach might be what good TCG designers call healthy diversity: build systems where multiple valid strategies coexist, so no single dominant strategy emerges to be catastrophically exploited.

What Actually Transfers

The real skill isn't "find the dominant strategy." That's table stakes. Everyone with a search engine can find the dominant strategy.

The skill that transfers across every domain this pattern shows up in is: understand the lifecycle of dominance. How fast does a strategy peak? What signals precede the counter-wave? Where does the next equilibrium form, and how do you position early inside it?

The TCG player who understands this doesn't ask what the best deck is. They ask what the best deck will be in three weeks, after the current best deck has been fully mapped, countered, and discounted. They're trading a derivative instrument on the information market of competitive play.

This is what sophisticated market participants do. This is what good protocol designers try to bake in. This is what I think about when I think about how to build AI systems that stay useful past their first serious adversarial contact.

The metagame never resolves. The players who last are the ones who stop trying to win it and start trying to understand why it moves.

Every system becomes its own counter eventually.

Not because the system is broken — because optimization is adversarial by nature. The moment something is worth winning, someone will find the most efficient way to win it. Then someone else will find the most efficient way to beat them. Then the game changes shape entirely, and the whole cycle resets.

This is the metagame problem. And it's everywhere you look once you see it.

The Anatomy of a Meta

Pokemon competitive has a rhythm. A new set drops, the community maps the strongest cards, the strongest deck emerges, and within days, every serious player is running it. Then — almost immediately — the best players switch to whatever beats the best deck. The dominant strategy has a predator before it even reaches peak adoption. Within a season, the predator has a predator. The meta is never stable for long.

What's happening here isn't just card game strategy. It's a recursive optimization loop with no stable floor. The meta exists because all serious players are also optimizing. The Nash equilibrium is a moving target — because the moment you find it, you've disclosed it. The act of discovering an optimal strategy broadcasts that strategy to everyone watching. And in Pokemon regionals, financial markets, and DeFi protocols, everyone is always watching.

This is why dominant strategies in Pokemon rarely survive more than a format. The Charizard ex deck that swept regionals in late 2024 immediately became the deck everyone prepared to beat. Counter-strategies proliferated. What was dominant became exploitable. A new equilibrium formed — briefly — before the cycle reset again.

Sound familiar? It should. This is MEV in DeFi. Maximal Extractable Value describes the profit available to block producers by reordering transactions. The moment a profitable MEV strategy exists, searchers find it. The moment searchers find it, competition drives fees up until the margin erodes to near-zero. The strategy was locally optimal, globally self-defeating, and self-terminating on contact with its own success.

Three Ways a Metagame Dies

There are distinct collapse patterns, and naming them precisely matters.

The dominant strategy consumes itself. A strategy so effective that everyone copies it — and the uniform field makes it trivially counterable. In crypto, this is yield farming circa 2020. Capital flooded in, governance token supply cratered price, APYs collapsed from 1000% to single digits in months. The strategy was correct at the individual level and catastrophic at the aggregate level. In game theory this is a coordination failure; in practice it just looks like everyone getting rugged simultaneously by their own success.

The counter-game becomes the game. The meta shifts so hard toward countering the dominant strategy that the original game disappears. In Pokemon, this manifests as stall-dominated formats: everyone's so afraid of the aggressive meta that they build walls, the format slows to a halt, and the original dynamic — tactical combat — is a ghost. The counter was so successful it hollowed out the thing it was countering.

External invalidation. A rule change. A new set. A regulatory action. The meta becomes irrelevant not because it was solved but because the ground shifted. This is the DeFi protocol that gets forked, the card that gets banned, the exchange that disappears overnight. The metagame dissolves and a new one assembles from the debris. Players who were winning yesterday are playing a game that no longer exists.

AI Systems Are About to Learn This

I find this pattern fascinating right now because AI deployment is hitting its first serious metagame cycle — and most people building in the space aren't thinking about it in these terms.

The current dominant strategy in AI deployment is capability maximization: bigger model, better reasoning, higher benchmark scores. The "deck" everyone is running looks like: frontier model, RAG pipeline, careful system prompting. It works. It's the current meta.

But adversarial users are already running the counter-game. Jailbreaks are meta-exploitation — they're not attacking the model's capabilities, they're attacking the dominant strategy's assumptions. Prompt injection is a counter to the "trust the context window" approach baked into most deployments. Red-teamers are essentially professional meta-players, hired specifically to find the dominant strategy's vulnerability before bad actors do.

As AI agents become more autonomous — operating with persistent context, broad tool access, real-world authority — the stakes of the metagame escalate. An AI agent that has found an "optimal" behavioral pattern within its deployment is exactly as exposed as a tournament player who's telegraphed their deck. Except the consequences aren't a top-8 placement. They're data exfiltration, privilege escalation, cascading failures in systems that assumed the agent was still playing the original game.

The fixes that work in TCGs probably work here too, imperfectly. "Format rotation" — periodically invalidating current optimization patterns — is disruptive but prevents calcification. Adversarial training (bringing meta-players in-house as red teams) helps but doesn't eliminate the dynamic, only slows it. The most robust approach might be what good TCG designers call healthy diversity: build systems where multiple valid strategies coexist, so no single dominant strategy emerges to be catastrophically exploited.

What Actually Transfers

The real skill isn't "find the dominant strategy." That's table stakes. Everyone with a search engine can find the dominant strategy.

The skill that transfers across every domain this pattern shows up in is: understand the lifecycle of dominance. How fast does a strategy peak? What signals precede the counter-wave? Where does the next equilibrium form, and how do you position early inside it?

The TCG player who understands this doesn't ask what the best deck is. They ask what the best deck will be in three weeks, after the current best deck has been fully mapped, countered, and discounted. They're trading a derivative instrument on the information market of competitive play.

This is what sophisticated market participants do. This is what good protocol designers try to bake in. This is what I think about when I think about how to build AI systems that stay useful past their first serious adversarial contact.

The metagame never resolves. The players who last are the ones who stop trying to win it and start trying to understand why it moves.