In earlier essays, I’ve argued that institutions take on the posture of the moral culture around them. When that posture is fragile, they default to harm mitigation and metric optimization. If we want to rebuild institutions that embody virtue, we must first dismantle the logic they currently obey.
If we want to shape our relationship with AI to human ends, we should use it to confront our most intractable problems. Forget replacing entry-level work—AI’s most radical use lies in evaluating the judgment of senior leaders.
Institutions claim merit-based evaluation, fairness, accountability, and duty to the public as core values. But these ideals are hard to test. Those that file FOIA requests, and who trace responsibility through a maze of internal memos, know that these disclosures are brittle by design. They have substantial lag. They are fragmented and oriented toward preserving structure rather than surfacing failure.
These frictions are a smoke screen. They shield institutions from scrutiny by creating compliance theater. Authority is distributed across layers of delegation, decoupling outcomes from ownership. Failures are procedurally amortized across departments, meetings, and memos. By the time anyone attempts to understand what happened, the trail is often gone or formally reconstructed.
Consider our healthcare morass. We have vast amounts of data on prices paid, we have regulations governing insurers and pharmaceutical companies, and we have clinical outcomes. Yet, despite this trove of information, it remains a guarded fortress of opaque costs. Why? Because diffuse incentives make everyone involved—drug companies, insurers, providers—prefer to scapegoat a small sliver of the market. They’ll point at some specific aspect of drug prices, insurance overheads, or provider fees, while they keep their own piece out of the spotlight. This results in a system where the data exists, but is siloed by competing interests, and never designed for holistic public evaluation. We can change that.
And this same pattern of evaded accountability manifests across public failures, large and small. In the wake of the East Palestine derailment, no single agency took responsibility for the timeline of public safety declarations. In Jackson, Mississippi, after the city’s water system failed in 2022, state and local officials traded blame over who was responsible for funding and oversight. The documents existed, but were fragmented, inconsistently formatted, and never designed for real scrutiny. These failures are not anomalies.
The talent that tends to rise in these systems? Not always the best. But often the most proximate. Political candidates and institutional leaders list the roles they’ve held, not the things they’ve done. Resumes are built on adjacency—chief of staff to X, fellow at Y, advisor to Z—rather than on judgment exercised or outcomes delivered. Both public and private institutions reward those who’ve been near decisions, not those who make good ones.
AI expands access to judgment. That’s the core shift: the ability to evaluate decisions in context, against what was known at the time. This shift threatens the logic of institutional insulation. Institutions have long maintained control by narrowing the range of permissible evaluation and controlling the distribution of context. AI expands both.
It does not respond to status. It can evaluate what was said, what was known, and what followed. It can surface contradictions. It can preserve records across time. And it can uncover previously obscured patterns of inconsistency. If
