The Three Circles of Policy
In his March 2026 essay "What AI Hypists Miss," Fukuyama identified three circles in policy analysis. The first is problem identification — recognizing that a problem exists and understanding its dimensions. The second is determining the optimal solution. The third is implementation — the actual deployment of the solution in the real world, with all the political negotiation, stakeholder management, and iterative adjustment deployment requires. "Intelligence only gets you to the end of the second circle," Fukuyama wrote, "and is of limited help in the third. An LLM cannot directly interact with stakeholders, message them, or come up with resources." The framework locates AI's capability precisely where it is most powerful and its insufficiency precisely where it matters most.
The Material Prerequisites — Contrarian ^ Opus
There is a parallel reading that begins not with circles of abstraction but with the substrate that makes any circle possible. The third circle—Fukuyama's domain of trust and implementation—isn't some timeless human constant that resists technological acceleration. It's a historically specific arrangement built atop material conditions that AI actively erodes. Trust doesn't float free; it requires stable employment, predictable career paths, institutional memory carried by long-tenured civil servants. When AI eliminates the middle-tier analytical jobs that served as training grounds for future implementers, it doesn't just create a skills gap—it destroys the social reproduction of implementation capacity itself.
The framework's neat separation of circles conceals how thoroughly they interpenetrate through labor. The analysts who identify problems in circle one become the managers who implement solutions in circle three. The junior staffers who optimize solutions in circle two develop the relationships and tacit knowledge that make circle three possible. AI doesn't leave the third circle untouched; it hollows it out from below by eliminating the career ladders through which implementation expertise develops. What appears as a 'binding constraint' in the third circle is actually the predictable result of destroying the human infrastructure that made implementation possible. The government contractors automating policy analysis today are the same firms that will sell 'implementation-as-a-service' tomorrow, capturing the very trust relationships Fukuyama imagines as irreducibly human. The third circle persists not because it resists technology but because its technologization requires one more turn of the screw—the full proletarianization of the professional-managerial class that currently performs it.
In the AI Story
The third circle is where trust operates. It is the domain of persuasion, negotiation, compromise, the management of competing interests, the cultivation of cooperative relationships that transform a good plan into a functioning reality. AI excels in the first two circles — identifying problems with extraordinary precision, generating optimal solutions with extraordinary speed. The third circle resists technological acceleration because the binding constraint is social, not cognitive.
The asymmetry this produces is dangerous. The capacity to generate solutions outruns the capacity to implement them. The gap between the two is filled by frustration, resentment, and the corrosion of institutional trust that implementation requires. Citizens see elegant solutions that cannot be enacted; policymakers see analytical capacity they cannot translate into political outcomes; the displaced see reforms proposed and never delivered. Each cycle of unimplemented optimization deepens the cynicism that makes future implementation harder.
The framework corrects a specific Silicon Valley error: the conflation of intelligence with effectiveness. Artificial general intelligence, even were it to arrive, would not dissolve the third circle's difficulties. The problem of AI governance itself exemplifies the pattern: technical solutions to alignment exist in principle and cannot be implemented without international coordination — which depends on institutional trust that has been declining for decades. The solution is not more intelligence. It is more phronesis — practical wisdom, the kind of knowledge that lives in the third circle and that AI structurally cannot possess.
The three-circle framework also explains why productivity gains do not translate automatically into civilizational improvement. Productivity is a first-and-second-circle metric: output per unit of input, where the unit of output is specifiable. The third circle — the domain of sustained cooperative practice, institutional construction, democratic deliberation — generates value that productivity metrics cannot detect. A society that optimizes only what productivity measures ends up rich in circles one and two and bankrupt in circle three. It can identify problems and solutions it cannot implement. It can analyze and propose but cannot enact.
Origin
Fukuyama developed the framework in two 2025–2026 essays for Persuasion: "Superintelligence Isn't Enough" (October 2025) and "What AI Hypists Miss" (March 2026). The framework built on his long-standing argument in Trust and The Origins of Political Order that institutional capacity — not raw intelligence or resources — is the binding constraint on social outcomes. It extended his earlier work on state capacity and implementation into the specific context of AI-driven cognitive amplification.
Key Ideas
Three circles, asymmetric acceleration. AI accelerates problem identification and solution design while leaving implementation unchanged.
Third-circle binding constraint. The social-political domain of implementation is where trust operates and where AI cannot substitute for human capacity.
Dangerous gap. Solutions generated faster than they can be implemented corrode institutional trust through visible non-delivery.
Productivity as first-circle metric. The metrics celebrated during the AI transition measure what AI accelerates and miss what AI does not touch.
Appears in the Orange Pill Cycle
Temporal Horizons of Change — Arbitrator ^ Opus
The framework's validity depends entirely on the temporal horizon under examination. For immediate policy challenges—next year's budget, this decade's climate targets—Fukuyama's analysis holds completely (100%). The third circle genuinely operates as the binding constraint, and no amount of analytical brilliance substitutes for the slow work of coalition building. The contrarian view correctly identifies the substrate erosion but dramatically overestimates its speed. Trust networks and implementation capacity have remarkable inertia; they persist for decades even as their foundations erode.
Where the contrarian reading dominates (80%) is in identifying the direction of change. The material prerequisites for third-circle work are indeed deteriorating—not through conspiracy but through ordinary labor market dynamics. The question isn't whether AI will eventually reshape implementation but how long existing trust networks can function while their replacement mechanisms atrophy. Fukuyama is right that intelligence and effectiveness differ, but wrong to treat this difference as permanent. The contrarian is right that implementation will be technologized, but wrong to imagine it happening through direct capture rather than gradual institutional evolution.
The synthetic frame both views need recognizes implementation as a historically specific competence rather than an eternal human domain. The third circle exists—Fukuyama maps it accurately—but its existence is contingent, not necessary. It persists through inherited institutional capital: the accumulated trust relationships, procedural knowledge, and social networks built during the twentieth century's expansion of state capacity. This capital depletes without renewal. The framework's real insight isn't the identification of three circles but the recognition that we're living through the lag between the technological disruption of circles one and two and the eventual institutional adaptation of circle three. The gap Fukuyama identifies is real but temporary—measured in decades, not centuries.
Further reading
- Francis Fukuyama, "What AI Hypists Miss" (Persuasion, March 2026)
- Francis Fukuyama, "Superintelligence Isn't Enough" (Persuasion, October 2025)
- James Scott, Seeing Like a State (Yale, 1998)
- Bent Flyvbjerg, Making Social Science Matter (Cambridge, 2001)