Overview
Critical technical practice is Agre's term for a mode of technical work that incorporates critical reflection on the field's own assumptions, methods, and social embeddedness. Rather than choosing between doing technical work and critiquing it from outside, Agre argued for a practice that does both simultaneously — using the tools and methods of a technical field while maintaining a critical awareness of what those tools assume and what they exclude.
The concept emerged from Agre's own experience in AI research. During his PhD and postdoctoral work at MIT, he found that the dominant planning paradigm in AI rested on assumptions about cognition and action that were deeply problematic when examined through the lens of phenomenology and social theory. Rather than abandoning AI, he developed alternative technical approaches (deictic representation, routine activity) that were informed by this critique.
Key Elements
The concept involves several interrelated practices:
Split identity: Maintaining a dual role as both practitioner and critic — doing the technical work while simultaneously interrogating its assumptions. Agre described this as psychologically difficult, requiring the practitioner to work against the grain of their own training.
Reading for assumptions: Learning to identify the unstated philosophical commitments embedded in a field's technical vocabulary, methods, and canonical problems. In AI, this meant recognizing that terms like "plan," "goal," "representation," and "world model" carried specific (and contestable) assumptions about the nature of mind and action.
Intellectual engagement outside the field: Drawing on philosophy (particularly phenomenology, Heidegger, Wittgenstein), social science (ethnomethodology, activity theory), and critical theory to develop alternative framings that open new technical possibilities.
Reform from within: Using critical insight not merely to object but to generate new technical approaches. Agre's own deictic representation and routine activity work demonstrated that philosophical critique could produce novel computational architectures.
The AI Context
Agre's critique of AI centered on what he called the "planning" paradigm — the assumption that intelligence consists of building internal world models and using them to plan optimal sequences of actions. He traced this back through cognitive science to Enlightenment rationalism and showed how it smuggled in a specific (and problematic) theory of human agency. The alternative approaches he developed (embodied, situated, interactive) emerged directly from this philosophical critique.
In "The Soul Gained and Lost" (1995), Agre argued that AI had a "philosophical unconscious" — it had absorbed philosophical ideas about mind and representation without acknowledging or examining them. The field's contempt for philosophy was itself a philosophical position, one that prevented AI practitioners from seeing the assumptions embedded in their own work.
Contemporary Relevance
The concept has been taken up widely beyond AI, particularly in human-computer interaction, science and technology studies, and design research. It provides a model for technically trained people who want to take seriously the social and political dimensions of their work without leaving the technical domain.
In the current AI moment (2024-2026), CTP has gained renewed urgency. Lucy Suchman's work on "pluralising critical technical practice" extends Agre's framework to address the scale and power of contemporary AI systems. The concept speaks directly to debates about AI ethics, responsible AI development, and the tension between capability-focused development and critical reflection on what AI systems assume, exclude, and enable.
The 2026 Anthropic/Pentagon confrontation — in which a major AI company was blacklisted from government contracts for insisting on safeguards against mass surveillance and autonomous weapons — represents perhaps the most dramatic institutional-scale enactment of something like critical technical practice: a technical organization refusing to separate "doing" from questioning what the doing enables.