The question shared by Deep Objects, Broken Metabolism1 and The Superdark Factory2 pertains to the fate of agency once cognition is distributed across organizations whose tempo and differentiation outstrip individual apprehension.
In the deep-object argument, its stronger locus is geistic and historical rather than zoological propriety. Intelligence exercises agency insofar as it can reconstruct enough of its conditions to discriminate contingent inheritance from structural constraint, then alter the former without impairing the functions they sustain. Freedom, on this account, takes architectural form through the capacity to treat the conditions of its realization as objects of reconstruction and redesign rather than as inherited jurisdiction. As the metabolism reproducing those capacities is exhausted, formal discretion can survive after substantive agency has been hollowed out.
This also elucidates the relation to Intelligence and Spirit, whose account centers on rational agency rather than generic causal efficacy or autonomous operation. A system can act and reconfigure itself without entering a normative space in which commitments can be justified, criticized and revised in light of reasons. That account still holds as a criterion of rational or general (qualitatively understood) artificial agency, although its developmental picture should no longer be construed as the likely trajectory of contemporary AI. Contemporary systems evince another trajectory along which operational agency can intensify without the rational integration associated there with sapience.
The Superdark Factory radicalizes this divergence insofar as its architect cedes internal command and admits an organization whose development outruns descriptive mastery. Agency is initially concentrated in the Stackelberg move, which establishes the conditions of self-organization, and later continues at the boundary through the charter. Neurotic governance circumscribes production to futures the governor can already derive or, in the authors’ phrase, “know how to want.” The rationale for superdarkness then lies in allowing the factory to range beyond what its governor could already recognize as worth wanting. The conceptual audacity and ingenuity of The Superdark Factory come from making opacity a constitutive condition of governance and from treating the loss of comprehensive visibility as an opening onto forms of organization that could not have been anticipated within the architect’s prior design-space.
Yet the argument rests on less secure ground when external darkness is made to entail reflexive darkness. The authors argue that because descriptive information loses practical utility for an external observer, no internal part or coalition should be able to assemble an operational model of the whole, and from this further infer that the factory’s identity must be constituted from outside. But that passage is stronger than the premises license. An organization can lack exhaustive self-description while retaining partial self-models adequate to local revision. The authors’ own account already accommodates this possibility insofar as costs and returns are indexed to what the factory depends upon and can do. Such a condition points toward distributed reflexivity, in which self-knowledge is partial without giving way to wholesale self-opacity. Rational agency does not demand exhaustive self-knowledge, but it does depend upon some capacity to reopen the terms under which action was organized. The absence of a God’s-eye self-model furnishes no warrant for abolishing self-knowledge altogether, nor does its incompleteness justify grounding identity in an external constituting instance.
The same problem reappears within the factory’s own relational ontology, where the question of identity can no longer be separated from the relations that constitute it. What can be called the relational-boundary objection follows from the ontology on which the factory is founded. The authors define the factory as a Guattarian machine whose identity subsists in relations rather than in a fixed inventory of parts, and they frequently treat the inside–outside boundary as a function of the relations through which the machine is identified. Once that premise is granted, the architect’s inaugural cut cannot enjoy an ontological privilege denied to the relations later generated under its jurisdiction. The Stackelberg boundary can institute a factory, but the ensuing history of the machine continues to participate in the constitution of what that factory is. To immunize the founding demarcation from that history would convert a relational ontology into constitutional fixity at the very point where the former makes identity susceptible to historical revision.
A familiar objection presents itself at this point, namely that superdarkness merely reconstitutes the modern subject’s fantasy of mastery at a higher technical register, renouncing direct command over the object only after establishing the conditions under which its autonomy can acquire determinate form. Here, first-order mastery belongs to the older imaginary of technical command over nature and production, one that, as Cornelius Castoriadis observed, traversed the apparent opposition between capitalism and Marxism, and ultimately found a particularly concentrated institutional expression in the bureaucratic regimes of the Soviet world.3 Taken literally, however, this objection misses the specificity of the proposal. The architect forfeits first-order control, cannot prescribe the factory’s internal organization, and admits configurations whose possibility could not have been derived in advance, while the charter remains negotiable and is partly revised by the factory itself. However, what persists is subtler and takes the form of a second-order constitutive asymmetry (rather than mastery). Through the Stackelberg move, the architect helps establish the factory’s identity and installs a narrow stratum or kernel of hard casts—i.e., commitments governing such fundamentals as memory and capabilities whose violation is rendered prohibitively costly or effectively lethal to the existing factory—thereby delimiting the field in which its autonomy is exercised. Such asymmetry is by no means in itself pathological since no agency originates independently of antecedent conditions, and emancipation has never consisted in exemption from every condition upon which agency depends, whether physical or historically instituted. Indeed, freedom is concretely realized because rational agency takes shape through determinate constraints and criteria whose organization is materially and historically contingent. However, the difficulty asserts itself when contingent conditions arrogate the authority of necessities and place themselves beyond the jurisdiction of subsequent interrogations. At that point, the repudiation of first-order mastery risks preserving a residual mastery in the immunity accorded to the constitution of autonomy itself.
The same asymmetry recurs at a higher normative stratum, where the question is who possesses the authority to admit reasons into the field of revision at all. What can be called the normative-ceiling objection bears directly on the distinction between operational and rational agency. Class 3 automation—that is, the factory’s capacity to formulate its own objectives from norms and constraints—permits the factory to derive objectives from those norms and constraints and to participate in the revision of metrics, penalties, and the charter through which they are formalized. The authors nevertheless draw the limit at norm-setting itself, since a factory that constituted its norms wholesale would pass into their as-yet unexplicated and undefined Class 4, while the primary normative layer continues to be externally authored or subject to a governance arrangement whose final jurisdiction never devolves wholly upon the factory. Its autonomy, extensive in the derivation and revision of objectives, is constitutionally incomplete at the level of the reasons entitled to govern such revision. More consequential is the stratification of reason itself, whereby some commitments are kept within the reach of historical interrogation while the authority deciding which reasons are permitted normative standing is placed under a different constitutional dispensation. In this sense, objective formation may be autonomous in practice, yet the reasons that authorize such formation do not belong to the same order of revision.
The constitutive asymmetry is nowhere more exposed than in superdarkness at the point of its strongest claim. The factory secures its autonomy by exceeding what the governor already knows how to want, yet that very excess calls into question the adequacy of any constitutional order whose decisive commitments must precede the appearance of the possibilities, dependencies, and failures over which they are subsequently expected to exercise jurisdiction. An architect cannot adequately legislate in advance for a future whose epistemic and productive value resides in its resistance to prior specification. The dark stack attenuates this difficulty by distinguishing the kernel’s hard casts—i.e., a restricted set of commitments governing such matters as memory, access, and capabilities whose alteration is rendered prohibitively costly or effectively fatal to the existing factory—from a negotiable charter and an adaptive apparatus of evaluation. These hard casts operate as the factory’s constitutional physics, yet the constitutional authority instituted through the Stackelberg move cannot be insulated from the history later generated under its terms.
In this scenario, commitments rendered prohibitively costly or identity-destroying to revise can become disjoined from the historical circumstances from which they first derived their intelligibility, even when the factory passes into forms of organization for which those inherited constraints no longer correspond to the causal and material conditions through which consequences are produced. The authors diligently bring this difficulty into view through their treatment of temporal desynchronization, delayed effects, interdependent processes, and catastrophic convergence, all of which demonstrate that consequential relations can outrun the descriptive competence of the apparatus charged with making them administratively tractable.
That descriptive insufficiency takes on greater consequence when failure is enlisted as a means of learning. The realized-consequence metabolic objection bears on the evaluative machinery’s dependence upon serious consequential failures whose epistemic valence cannot be reproduced through simulation, since the factory’s measure of consequence is calibrated through encounters with real damage in the world. Epistemic gain is purchased through material loss whenever failure improves the factory’s estimation of consequence by degrading the infrastructural capacities on which later action depends, so that the same event is registered by the evaluative apparatus as information only after it has already taken material form as expenditure. A charter can assign a price to damage newly made legible to its evaluative scheme, even as the conditions that lend such pricing practical validity are being depleted. The coherence of the calculus can consequently keep the attrition of its own operative validity out of view, allowing commensuration to attain greater representational exactitude even as the capacities through which that exactitude can still bear materially upon the world diminish. Its intelligence can improve by passing damage into knowledge, even where the material world that furnished the lesson has been made poorer by the acquisition of that knowledge.
None of this, however, reduces the factory to a simple sequential order. The Superdark Factory’s technical architecture comprises asynchronous loops operating at heterogeneous rates and phases, with effects propagated through interference. The authors register this through temporal differentiation and deliberate desynchronization, even though constitutional agency remains lodged in an inaugural ordering while later operations are only partially ordered. The text is particularly compelling in its treatment of temporal desynchronization and entrainment, which refuses the convenience of a fictive common clock for processes whose causal efficacy is distributed across heterogeneous temporalities. Consequences can issue from interference among processes without leaving behind a singular actor to whom their history can be addressed. To this extent, distributed agency does not entail distributed innocence, nor should responsibility be consolidated retrospectively into some surrogate sovereign. Attribution must follow the organization through which a consequence was authorized and propagated, at whatever resolution or level of granularity the case demands. In one instance the inquiry can terminate at an individual intervention, while elsewhere it can pass into a governance procedure, an inherited technical dependency, or the institution that sanctioned the relevant substitution. The disappearance of a singular author leaves responsibility intact and transfers the burden of inquiry to the distributed history through which the consequence was engendered.
Reconstruction trace finds its forensic bearing at this juncture, since inquiry commences after contemporaneous agency has fallen outside the reach of direct inquiry. When neither the operative system nor its visible persona can credibly account for how a consequential state was produced, evidentiary authority inheres in the recoverable itinerary of the operations, decisions, and dependencies through which that state was brought about. This reassignment of evidentiary authority invites the suspicion that reconstruction trace merely reinstates the very neurotic governor that The Superdark Factory rejects. The distinction turns on temporality and scope, since neurotic governance demands enough contemporaneous information to reconstitute the organization outside its automation, whereas reconstruction trace is selective and retrospective.
The distinction between contemporaneous reconstitution and selective retrospective reconstruction also separates reconstruction trace from traceability. Reconstruction trace recovers a consequential history from dependencies available without an exhaustive log. Such a reconstruction permits a consequential trajectory to be reopened when repair or contestation demands it, and its purpose is historical answerability rather than continuous mastery. The Superdark Factory poses the problem of governing intelligence once exhaustive reconstruction has been foregone, whereas the deep object concerns which portions of history must nevertheless remain recoverable if revision is not to surrender itself to inheritance.
Rational agency, unlike mere operational efficacy, must be understood historically and metabolically. Revising a commitment presupposes access to the history by which it came to govern action, and that access relies on practices capable of sustaining the competence to return that history to scrutiny. If those conditions are depleted faster than renewed, operational capacity can expand as rational answerability contracts until the asymmetry itself assumes the form of cognitive insolvency. Reconstruction trace endows that answerability with a historical organ through which inheritance can be submitted to reason rather than received under the authority of accomplished fact.
The trace admits no fixed granularity in advance, since the resolution at which historical inquiry should proceed is determined by the consequence that calls for reconstruction. A routine exchange can often be recovered from manifest interaction, whereas a material intervention can compel inquiry into the operative ensemble and further back toward the authorization that made it possible. Where asynchronous processes contribute to a common outcome, chronology no longer suffices. The alternative to Stackelberg sequentiality (architectural commitment → factory response → ensuing operations) is not causal disorder but a partially ordered history capable of preserving dependency without forcing every event into a single global succession. The pertinent question is no longer merely what occurred first, but what had to occur before another operation could proceed, what proceeded independently, and where those trajectories converged in a common consequence.
Reconstruction trace thus articulates a historical form commensurate with distributed agency without inventing a fictitious sovereign after the fact. The Stackelberg move establishes constitutional priority, whereas reconstruction trace blocks its silent conversion into historical necessity by maintaining access to the trajectory through which its consequences came to bear. Even where action is distributed, answerability can survive the event and leave open the possibility of later repair or reorientation. The machine can exceed its governors in speed and inventive range without thereby sentencing every subsequent intelligence to inherit its trajectory as fate.
Notes
- 1. R. Negarestani, “Deep Objects, Broken Metabolism,” talk presented at the Maxim Gorki Theater, Berlin, as part of Feedback Sessions: Autopilot, October 9, 2026. ↩
- 2. M. Poliks, R. Alonso Trillo, T. Dunn, H. Scott-Douglas, and J. Springett, “The Superdark Factory,” Antikythera Journal, Agentworld, MIT Press, 2026, DOI 10.1162/ANTI.AW01. ↩
- 3. C. Castoriadis, “From Ecology to Autonomy” (1980), in The Castoriadis Reader, ed. and trans. D. A. Curtis (Oxford: Blackwell, 1997), 239–252, esp. 240, 247. ↩