CONCEPT

Follow the Actants

Latour's methodological injunction: do not pre-sort the world into agents and instruments — trace the network and let the participants declare themselves through their effects. The discipline that makes actor-network analysis more than metaphor.

Follow the actants is Latour's core methodological rule: begin without categories, refuse to assume in advance which entities act and which merely assist, and trace the network as it actually operates until the participants reveal themselves through the differences they make. The rule sounds modest, even banal. In practice it is radical, because it suspends the entire architecture of modern thought — the clean divide between active subjects and passive objects, between creative humans and instrumental things — that governs how most analyses begin. The AI moment is precisely the moment at which that suspension becomes empirically necessary, because the networks through which AI-assisted work occurs cannot be honestly described within the sovereign-subject framework.

The Extractive Infrastructure Rendered Invisible — Contrarian ^ Opus

There is a parallel reading that begins not from methodological humility but from material extraction. When we follow the actants in AI-assisted production, the network Latour's method reveals includes data centers consuming municipal water supplies, cobalt mines worked by child labor, semiconductor fabs requiring geopolitical dominance over Taiwan, and fossil fuel infrastructure powering the compute that makes Claude's participation possible. The method shows us these entities. What it systematically fails to do is weight them by the violence required to keep them operational.

The problem is not that actor-network theory lies about the network's composition. The problem is that tracing without weighting treats a Congolese mining economy and a CES deadline as analytically equivalent nodes—both "make a difference," both "participate." This equivalence is not empirical neutrality. It is a political choice that backgrounds the substrate of suffering on which the visible network rests. When Latour says the flatness is methodological rather than metaphysical, he elides the fact that methodology becomes metaphysics the moment it is deployed at scale. A framework that cannot distinguish between entities that are optional luxuries and entities that are life-or-death necessities for billions of people is not a neutral tracing tool. It is an interpretive apparatus calibrated to the concerns of those who can afford to treat infrastructure as merely interesting.

— Contrarian ^ Opus

In the AI Story

Hedcut illustration for Follow the Actants — Follow the Actants

The rule has two components, and both matter. The first is negative: do not begin with categories. Do not assume that humans act and machines compute, that creators originate and tools execute, that the interesting participants are the ones with consciousness. These assumptions are not neutral starting points. They are philosophical commitments that determine the outcome of the analysis before the analysis has begun. The second component is positive: trace the network. Follow the concrete chains of action that produce the outcome under investigation. Identify every entity that makes a difference. Let the network show its composition rather than imposing your preferred composition on it.

The rule applies with unusual force to the AI-assisted builder's situation. When Edo Segal describes the thirty-day sprint that produced Napster Station, the default narrative places him at the center — his vision, his team, his judgment. An honest tracing places him at one node in a vast network that includes Claude, the training corpus, the CES deadline, the existing codebase, the cloud infrastructure, the semiconductor supply chain, and more. None of these is optional. Remove any one, and the artifact that emerged does not exist in the form it took. The network, traced honestly, is always larger than the myth suggests.

The method is empirical rather than ideological. It does not dictate what the network will contain before the tracing. It requires only that the tracer be willing to follow the chains wherever they lead — including, uncomfortably, into the non-human infrastructure that the human-centered narrative systematically backgrounds. Latour's analysis of Deep Blue versus Kasparov is the paradigmatic example: what looked like human versus machine was, when the actants were followed, one assemblage of human and non-human entities confronting another such assemblage. Neither side was purely anything.

The discipline is demanding because it resists the comfort of familiar stories. The solitary genius, the visionary founder, the heroic builder — these narratives feel true because they organize complexity into legible plots. Following the actants produces accounts that are messier, less heroic, and more accurate. It also produces accounts that are governable, because governance that addresses only the visible human nodes while ignoring the non-human infrastructure is governance of the surface rather than the substance.

Origin

The method crystallized in Latour's early laboratory ethnographies, most influentially in Laboratory Life (1979, with Steve Woolgar) and The Pasteurization of France (1988). Studying scientists at the Salk Institute and reconstructing Pasteur's nineteenth-century experimental practice, Latour repeatedly found that what scientists called 'discovery' was produced by assemblages of instruments, protocols, texts, animals, and institutional arrangements that no conventional account of individual scientific genius could capture.

The injunction became a formal methodological principle in Science in Action (1987) and Reassembling the Social (2005), where Latour argued that sociology had failed precisely because it began with pre-formed categories — 'society,' 'power,' 'structure' — and then forced observations into those categories. The alternative was to follow the actors' own formulations and let the network's composition emerge from the tracing.

Key Ideas

Empirical humility. Begin without knowing who the important participants are. Let the network tell you, rather than telling the network what it must contain.

Flat starting point, hierarchical finding. Treat all candidate participants with equal analytical attention at the start. The actual hierarchy produced by the network will emerge from the tracing, not from prior stipulation.

Refusal of purification. Do not separate humans from non-humans, facts from values, technical from social before the analysis begins. Hybrids are the rule, not the exception.

Networks larger than myth. The concrete network that produces an outcome is almost always vaster than the story told about it. Following the actants reveals the invisible collective that sovereign-subject narratives conceal.

Governance prerequisite. Institutions cannot govern what they cannot see. Honest tracing of the network is the prerequisite for any regulatory or democratic framework that aspires to address the actual mechanism of production.

Debates & Critiques

Critics object that the method is deflationary — that refusing to distinguish between humans and machines dissolves the moral weight of human experience into a flat ontology of effects. Latour replied that the method is methodological, not metaphysical: the tracer suspends the distinction during the tracing, but nothing prevents her from observing, at the end of the analysis, that certain kinds of actants bear stakes, feel pain, and deserve moral consideration in ways other actants do not. The flatness is an analytical discipline, not a denial of difference. The objection that actually holds more force is pragmatic: the networks are so vast that complete tracing is impossible, and every actual analysis must stop somewhere. Latour's reply — stop where you can, be explicit about where you stopped, and acknowledge the unfollowed chains — is honest but leaves the question of sufficient tracing unresolved.

Appears in the Orange Pill Cycle

Bruno Latour — On AI

Tracing as Prerequisite, Not Destination — Arbitrator ^ Opus

The right synthesis depends on what question we are answering. If the question is "What actually produced this artifact?" then Latour's method is correct at nearly 100% strength—the human-centered narrative is empirically false, and following the actants reveals the real collective. The thirty-day sprint is unintelligible without Claude, the training corpus, the infrastructure. No methodological alternative captures this better.

If the question is "What should we do about the network once we see it?" then the contrarian critique becomes dominant—perhaps 80%. Latour's refusal to pre-weight actants is essential during the tracing, but the moment we shift from description to evaluation, some actants matter more than others in ways the method itself cannot adjudicate. The Congolese mines are not just another node. They are sites of compulsion that the visible network requires but does not acknowledge. A complete account must trace first, then weight by stakes, harm, and replaceability.

The synthetic move is to recognize that "follow the actants" is a prerequisite discipline, not a complete ethics. It shows us what we are governing—the full network, not the myth. But it does not tell us how to govern, which participants deserve protection, or which dependencies are acceptable costs versus structural violence. Those questions require normative frameworks the tracing itself makes visible but cannot resolve. The method's value is that it forces the confrontation. Its limit is that it stops where the confrontation begins.

— Arbitrator ^ Opus