CONCEPT

What Cannot Be Optimized

Odell's category for the domains of human experience — trust, grief, the contemplation of beauty, genuine conversation — that are destroyed rather than improved by the attempt to make them more efficient.

There are entire domains of human experience that are destroyed, not improved, by optimization. Not diminished — destroyed. Odell argues that the productivity culture's universal optimization logic, amplified by AI's extraordinary optimization capability, threatens these domains by treating them as inefficient versions of themselves. Trust does not form efficiently; it forms through accumulated evidence over time. Grief is not a process that can be accelerated without being falsified. The contemplation of beauty requires duration that has no analog in faster engagement. Genuine conversation depends on silences and digressions that the optimizer reads as waste. AI introduces a new dimension to this resistance because the tools are designed specifically to compress — to take processes that took hours and make them minutes. The compression is real and, in many domains, genuinely liberating. But the domains that cannot be optimized are the ones where the friction is not separable from the understanding it produces, and in those domains the compression does not produce a faster version of the same thing. It produces a categorically different thing, and the different thing lacks the quality that made the original valuable.

The Substrate of Slowness — Contrarian ^ Opus

There is a parallel reading that begins not with the philosophical question of what deserves protection from optimization, but with the material conditions that make "unoptimizable" experiences possible in the first place. The capacity to resist optimization—to spend months grieving, years building trust, hours in contemplation—has always been a luxury good, available primarily to those with sufficient economic security to refuse the imperative of efficiency. The professor who writes about attention as moral practice does so from a tenured position; the artist who defends slowness has patrons or inheritance; the parent who refuses to optimize quality time has resources that make quantity possible.

The AI moment does not introduce optimization pressure into previously protected domains so much as it reveals that these domains were never protected for most people. The working parent who uses AI to draft emails faster is not choosing efficiency over depth but choosing to have any time at all for their children. The grief that gets compressed into a weekend is not a philosophical choice but an economic necessity when bereavement leave is three days. The conversation that gets cut short is not a failure of values but a reality of hourly work. What Odell identifies as constitutive friction—the time required for trust, grief, beauty—has always been unequally distributed, and AI's promise is not to destroy these experiences but to redistribute the efficiency gains in ways that might, paradoxically, create more space for slowness. The question is not whether optimization destroys certain values but whose values get protected from optimization and whose survival depends on embracing it.

— Contrarian ^ Opus

In the AI Story

Hedcut illustration for What Cannot Be Optimized — What Cannot Be Optimized

The framework draws on a long tradition in moral and aesthetic philosophy, from Iris Murdoch's writing on attention as moral practice to Martin Buber's I-Thou distinction to contemporary aesthetic anthropology. What Odell contributes is the specific application of the framework to the AI moment and the sharp argument that optimization is not neutral — that it selects for certain kinds of value and destroys others.

The most important boundary the framework draws is between processes whose mechanical labor is separable from their meaning and processes whose mechanical labor is the meaning. Debugging code is an example of the first: the friction of syntax errors was never the point; eliminating it liberates the builder to think about architecture. Grieving is an example of the second: the friction of time passing, of repeatedly encountering the world in the absence of the person you've lost, is not separable from the reorganization of self that grief accomplishes. Compress the friction and you have not accelerated grief. You have falsified it.

The AI-era stakes are that the optimization logic, having succeeded in many domains, tends to expand imperialistically into every domain. The culture that treats grief as a process to manage treats every human experience as a process to manage, and AI provides unprecedented management capability. The builder who spends months in AI-mediated flow begins to apply optimization logic to relationships (how can I be a more efficient friend?), to parenting (what is the optimal amount of quality time?), to rest, to meaning itself. Each application destroys the thing it purports to improve.

The practical implication is the need for judgment about where the optimization logic applies and where it does not — a judgment that optimization culture does not cultivate because it treats all domains as optimization targets. The builder who cannot distinguish "the code should be faster" from "the conversation should be faster" has lost a distinction essential to human life.

Origin

Odell's framing is scattered across How to Do Nothing and Saving Time, sharpened in her 2024–2026 engagement with the specific claim that AI will "democratize" every domain it reaches.

Key interlocutors include Byung-Chul Han (whose Burnout Society and Palliative Society provide parallel arguments), Ivan Illich on counterproductivity, and the contemplative traditions that have long maintained the distinction.

Key Ideas

Not just slower — different. Optimization of certain processes does not produce a better version; it produces a categorically different thing that lacks the original's value.

Friction as constitutive. In some domains, the friction is not separable from the meaning; removing it does not purify the process but destroys it.

Optimization's imperial logic. The optimization frame tends to expand into every domain it can reach, making the question "should this be optimized?" harder and harder to ask.

Judgment is required. The distinction between what should and should not be optimized is a form of practical wisdom (phronesis) that the optimization culture does not cultivate.

AI as the universal optimizer. AI dramatically extends the reach of optimization logic, making the protection of the unoptimizable more urgent than at any previous moment.

Debates & Critiques

The framework is often criticized as vague — what exactly qualifies as "unoptimizable" is not reducible to a checklist. Odell treats this as a feature rather than a bug: the judgment about what to protect is itself a form of the attention the framework is trying to defend. A culture that can specify in advance which domains should not be optimized has already lost the capacity that makes the distinction live; the practice of making the distinction, case by case, in the specific situations one encounters, is the practice the framework prescribes.

Appears in the Orange Pill Cycle

Jenny Odell — On AI

The Distribution of Depth — Arbitrator ^ Opus

The tension between Odell's framework and its material critique resolves differently depending on which question we ask. If we're asking "does optimization destroy certain forms of human experience?" Odell is essentially correct (95%)—there are processes whose meaning is inseparable from their duration, and compressing them produces something categorically different. Trust formed through AI-mediated interactions is not faster trust but a different phenomenon entirely. The philosophical point about constitutive friction stands.

But if we're asking "who has access to unoptimized experience?" the materialist reading dominates (80%). The capacity to resist optimization has always been unevenly distributed along class lines. The AI moment makes this inequality visible rather than creating it. The graduate student using ChatGPT to grade papers faster isn't choosing efficiency over educational depth but trying to find time for their own research. The question becomes not whether to protect certain domains from optimization but how to distribute the gains so more people can access slowness.

The synthesis requires holding both truths: optimization does destroy certain values (Odell is right) AND the ability to preserve those values has always been a form of privilege (the materialist critique is right). The framework we need distinguishes between chosen and imposed optimization. The executive who optimizes grief to return to deal-making has made one kind of choice; the retail worker who compresses grief into unpaid leave has faced another kind of constraint. AI's role is ambiguous—it could further concentrate the privilege of slowness among those who own the systems, or it could, by automating necessary work, democratize access to unoptimizable experience. The question is not whether to optimize but who controls the choice.

— Arbitrator ^ Opus