Every week, some piece of AI competence gets demoted from skill to default. A prompt trick becomes a button. A careful context routine matters less when the window expands. A five-step workflow becomes a native agent behavior. A wrapper starts to feel like canal infrastructure after the railways arrive.
The work was real. It saved time. It clarified judgment. It made projects possible that would otherwise have stayed in ideation. It also aged in public.
The personal churn is only the small version. Everyone else around the table is also getting new powers. You can do things that you could not do before. So can your competitors, your employer, your customers, the frontier labs, the platforms, and the largest states in the system.
“Arms race” names some of the pressure, but it makes the game too simple. An arms race imagines speed along one visible track. The AI pressure is stranger: people are building conversion systems before they know which outputs will convert into future value.
Heavy Eurogames describe that pressure better than most of the usual tech metaphors. Their drama is rarely a single decisive attack. It is the slow construction of a position. A cube becomes a resource. A resource buys a card. A card changes the value of a city, a route, a worker, or a province. Early moves matter because they commit you to a theory of the later game.
AI work has acquired that shape. An AI position may include local artifacts, institutional assets, and physical constraints: prompts and evals; workflows and distribution; capital, chips, energy, and law. Their value depends on the scoring rule that emerges later: what becomes cheap, what remains scarce, and who can still be trusted to decide the difference.
First moves
A beginner sees a turn as a turn: take wood, build road, draw card, place worker. A stronger player sees the chain. This resource is two turns away from a card. This card changes the value of a city. This city matters only if the later scoring condition makes cities matter. The board is full of objects that look concrete but are really claims about future conversion rates.
Serious AI use already works this way. The simple form is prompt in, answer out. The stronger form treats the prompt as one move inside a setup: context that can be retrieved, examples that calibrate the model, source bundles that constrain the answer, review habits that catch fluent nonsense, and the discrimination to know when the output sounds better than it is.
The next level up is abstracting the relation the sequence controls. A road can be a wooden piece, a route, or leverage over another player's timing. Behavior analyst and chess master Francis Mechner observed that chess masters report relations such as “lines of force” and “powers of a piece.”11. Mechner, F. (2010). Chess as a behavioral model for cognitive skill research: Review of Blindfold chess by Eliot Hearst and John Knott. Journal of the Experimental Analysis of Behavior, 94(3), 373–386. The master sees not just a rook and a sequence of events but also strategic capacity.
Apropos artificial intelligence, the visible piece may be a prompt or an app, a model or a data center. The stronger player asks what condition makes the piece score: model capability, inference cost, power access, distribution, liability, trust, or the ability to evaluate a result before it reaches a user.
To analyze AI behavior functionally is to ask which relation among context, model, task, consequence, and review makes the system behave as it does. Once you can systematically conceptualize those relations, you can see why two moves that look unrelated on the surface may be playing for the same scoring card.
A player who looks past chatbot features toward substations, cooling, chips, model routing, or regulatory trust is making the same abstraction at another scale. The consumer-facing tool is the visible piece. The bottleneck underneath is the relation every AI strategy has to pass through.
A missing rulebook
In a Eurogame, uncertainty is bounded. You may misread the strategy, but the contingencies exist somewhere: printed on a card, inferable from the rulebook, discovered through repeated play, discussed on BoardGameGeek, compressed into a strategy guide. The rules rarely change because a publisher shipped a stronger model on Tuesday.
AI removes that comfort. Today's workbench can be repriced by a release note, a price cut, a benchmark jump, a legal rule, a procurement norm, a safety incident, or a data-center bottleneck. The hidden scoring card is only part of the uncertainty. The board itself can change while the player is still building.
The United States and China are not simply racing along the same track. Kyle Chan describes the American position as heavily oriented toward frontier models and AGI, while China also pursues model quality but puts more emphasis on efficiency, open source diffusion, applications, robotics, energy, and deployment.22. Douthat, R., & Alvarez Boyd, S. (2026, May 14). China’s not the problem. We are. [Op-ed conversation with Kyle Chan]. The New York Times. The same pressure appears at the human scale in Yi-Ling Liu’s reporting: American elites talk about high agency, Chinese founders hunt for the fengkou, and workers in both countries describe being managed, harvested, or reduced to nonplayer characters in someone else’s game.1111. Liu, Y. (2026, May 12). The shared feeling of being harvested by the future. The New York Times.
If the final card is imminent and singular in impact, betting on frontier model capability looks stronger. If the game lasts longer, the stronger position may accrue from world-building: cheap deployment, energy-backed infrastructure, robotics, diffusion, and integration into daily services. Neither side simply waits for the rulebook. Their investments help write it, while everyone else learns which parts of the future have already been assigned to them.
The end of eras
Or what survives
In Brass: Birmingham, the board has a memory with an expiration date.33. Order of Gamers. (n.d.). Brass: Birmingham rules summary (Version 1.2). The game moves from canals to railways. At the end of the first era, level-1 industries are swept off the board, while higher-level industries survive. Some early infrastructure scores and disappears. Some carries forward.
Some AI work is level-1 industry: valuable because of today's awkwardness, fragile once the awkwardness disappears. Elaborate prompt chains matter while the model struggles to plan. Manual context-packing matters while the window is small. Glue code matters until the platform absorbs the obvious integration. This work can be exactly right for the current era and still become bookkeeping by May.
The older infrastructure in Brass can still score. It can fund the next move. A player can spend a long time profitably trading in the old era. Many current AI workflows will work the same way. They still solve the original problem. In a less competitive setting, players may thrive for years with the old deck.
In a more competitive game, holding onto old cards converts to a loss. The pieces remain on the table. They just stop giving a player access to the action spaces that matter.
The pattern scales above the individual. A lab can build around an evaluation that stops separating models. A firm can reorganize around an integration that becomes native to the platform. A state can pour resources into a bottleneck that matters for this generation of systems and less for the next. Current-era value is real. Carry-forward value is the harder question.
And yet, “survives the next release” is also a bet. A coding script can become a checkbox. A memory system can be over-engineered in light of a longer context window. An energy strategy can pay off only if the compute build-out stays energy-bound. The label “durable” does not settle the scoring rule. It exposes the bet.
Hidden multipliers
My introduction to heavy Eurogames was Through the Ages, a half-day lesson in watching early civilizational choices become valuable only after later events start scoring them.44. Czech Games Edition. (n.d.). Through the Ages: A new story of civilization [Rules]. In Concordia, the value of a house comes through the scoring cards you acquired.55. Rio Grande Games. (2018). Concordia Venus rules (Version 11). The same board can be worth different amounts depending on the multiplier attached to it.
Multipliers can discount as well as reward. If you built heavily toward a category that scores weakly, the pieces remain on the board. They just convert poorly.
AI is applying fractional multipliers to work that used to convert more cleanly. Before ChatGPT's public release, a polished first draft often signaled scarce fluency. A person who could summarize a messy document, turn requirements into code, or produce acceptable copy owned a visible differentiator. Those abilities still cash out. They do not score the same way when the first pass appears inside every interface.
Price changes the multiplier directly. Epoch's analysis of LLM inference prices found rapid but uneven declines: across selected benchmark thresholds, the price of reaching fixed performance levels fell between 9x and 900x per year, with a median of 50x per year.77. Cottier, B., Snodin, B., Owen, D., & Adamczewski, T. (2025, March 12). LLM inference prices have fallen rapidly but unequally across tasks. Epoch AI. Anthropic's Claude Haiku 4.5 announcement made the same rule change concrete: performance that had recently been near the frontier became available at roughly one-third the cost and more than twice the speed of a model from five months earlier.88. Anthropic. (2025, October 15). Introducing Claude Haiku 4.5. A move that once cost too much to repeat can become normal operating expense.
Other work can reprice upward for the same reason. The plumber, the electrician, the compliance officer, the procurement lead, and the domain expert gain or lose value according to the scarce part of the system: access, liability, trust, embodied repair, institutional knowledge, or the authority to decide whether an automated result is good enough to use.
The same logic turns “AI infrastructure” into more than a slogan. Compute needs power and cooling; models need chips and data rights; deployment needs evaluation and distribution. If every player can generate plausible output, advantage moves toward whatever remains difficult to obtain, route, verify, power, insure, or trust.
Productivity gains can feel strangely unsatisfying for this reason. Doubling output is not doubling score if the scoring card now says raw output is worth a fraction of what it used to be. You can be faster, better equipped, and more productive while the market value of the thing you learned to produce is falling underneath you.
The multiplier is not yet drawn.
Game changers
In a multi-player game, your action may do more than improve your own position. It can change the board state other players inherit. You take the last worker space, unlock a technology, flood a market, trigger an event. What was a choice for you becomes a constraint for someone else.
AI intensifies that mechanic because the largest players can move the board beneath everyone else. You can build an app that would have been impossible a year ago. A platform can absorb the app into a default feature. You can automate a task that used to cost a day. A frontier lab can make the same automation cheap enough for everyone. You can find a niche. A larger player can decide the niche is now a distribution feature.
A frontier lab releases a cheaper, more capable model. For the lab, the release is designed to improve position: attract usage, justify capital spending, force competitors to respond, and reduce the cost of serving each user. For a firm, the same release becomes a new action space: adopt it, ignore it, reorganize around it, or explain why not. For a worker, the same release may arrive as a rule change: AI skills required, hiring slowed, entry-level role redesigned, the same team expected to produce more.
One player's move becomes another player's game mechanic. A cheaper model is margin for one actor, budget relief for another, and a new baseline expectation for a third.
A benchmark can play that role too. Stanford's 2026 AI Index reports that capability is outpacing benchmarks designed to measure it: Humanity's Last Exam improved by 30 percentage points in one year, some evaluations meant to remain difficult for years saturated in months, and OSWorld agents reached 66.3 percent accuracy while still failing roughly one in three structured attempts.66. Stanford Institute for Human-Centered Artificial Intelligence. (2026). Technical performance. In 2026 AI index report. A benchmark that separates positions is a public score for a lab, a procurement signal for a firm, and a threat or opportunity for a worker. Its strategic value falls when the score stops separating real positions.
The cost is lost access. A lab can miss a model cycle and still raise again, or fail outright. A firm can absorb AI well, waste money on it, be acquired, or disappear. A worker can gain leverage, lose leverage, or find that the entry rung into a field has been removed. “The game changed” is not an abstraction when it changes which actions a player is allowed to take next.
A strong position keeps the next turn open.
Endgame
Endgame starts when the next available turn matters more than the current score.
Some turns cash out what the player can already do. They produce a report, a course, a client deliverable, a model release, a procurement win, a policy announcement, a launch. Cash-out work matters. It can pay rent, validate the setup, or fund the next move.
Other turns change what the player can do next. A lab may buy cheaper inference or stronger distribution. A firm may build the review path that lets AI enter accountable work. A worker may turn examples into judgment. These moves leave the player with a better next turn.
Both kinds of work are necessary. Both can be misread. A setup that never cashes out becomes private theater. A productive output that leaves no position behind can feel like progress while the table moves past it.
The clock changes behavior before the outcome is settled. Jasmine Sun reported a Silicon Valley mood in which founders, engineers, and executives talk as if the window for human economic leverage may be closing: build wealth now, get inside the winning side now, automate before competitors automate you.99. Sun, J. (2026, April 30). Silicon Valley is bracing for a permanent underclass. The New York Times. Once people believe that, model releases become labor-market events.
Stanford's Digital Economy Lab has reported evidence consistent with early-ladder damage: early-career workers, ages 22 to 25, in the most AI-exposed occupations saw relative employment declines while more experienced workers in the same occupations remained more stable.1010. Brynjolfsson, E., Chandar, B., & Chen, R. (2025, November 13). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence [Working paper]. Stanford Digital Economy Lab. The causal story remains open. The pattern still matters because the junior role is where a person learns the board.
If the entry rung disappears while you are trying to step on it, the long run is not the one you get to live. The short run can make a person feel less like a player than a nonplayer character: present on the board, useful to someone else’s game, but not choosing the turns.
At every scale, the score that matters is the next available action. A lab's headline matters if it buys cheaper models, better distribution, or more reliable systems. A firm's AI program matters if it changes the workflow and accountability structure. A state's strategy matters if it builds the energy, talent, deployment channels, and legitimacy needed for the next era. A worker's move matters if it places them where judgment can compound, instead of on the visible axis where generic AI fluency is already crowding.
ClawSuite Relay
A local-first command relay for turning AI work into explicit, reviewable terminal actions.
AI in practiceThe decision before the decision
AI shapes the choice before the visible decision point.
AI in practiceA machine that says sorry
Anthropomorphic AI gives us familiar but flawed controls.