what is an agent? a survey of definitions across time and discipline

survey of definitions across disciplines

this page synthesizes the project as a cross-disciplinary survey of agent definitions, linking every analytic dimension and discipline module developed so far.

abstract

this meta-study inventories definitions of “agent,” “agency,” and “agentic ai” across time and discipline. the survey combines a chronological corpus with thematic analyses so you can trace how autonomy, embodiment, goals, and disciplinary language evolve.

dataset backbone

core inventory: agents-definitions.mdx — time-ordered entries with citations and a downloadable JSONL file.
schema reference: embedded in the same page; captures date precision, discipline, source type, notes, and quote flags for reproducibility.

analytic dimensions

autonomy spectrum: how much discretion does the agent exercise?
entity frames: what kind of entity (human, institutional, machine) is presumed?
goal dynamics: does the agent accept, adapt, or negotiate objectives?
persistence & embodiment: is the agent session-bound, persistent, or physically embodied?

each dimension page outlines heuristics, tables, diagrams, and residual questions to test how convincingly a definition fits the proposed categories.

disciplinary syntheses

philosophy: intention, reasons, intentional stance.
psychology: perceived control, self-regulation, obedience.
law: delegated authority, fiduciary duty, liability.
economics: incentive alignment, agency costs, governance.
complex systems: adaptive rule-followers, emergence, thermodynamics.
cognitive science: mental modularity, plans, process societies.
computer science: agent-oriented programming, interaction protocols.
human-computer interaction: mixed-initiative assistance, user control.
sociology: structure-agency duality, collective actors.
artificial intelligence: perception–action loops, llm tool ecosystems.

suggested workflow

survey the inventory: scan the chronological entries to observe definitional shifts and compile custom slices via JSONL.
map dimensions: apply the four analytic lenses to situate definitions along autonomy, entity assumptions, goal handling, and persistence.
compare disciplines: read the qualitative syntheses to understand each field’s emphasis and unresolved debates.
annotate & extend: add metadata columns (e.g., goal_stance, entity_frame) and link back to the source definitions as the dataset grows.

future directions

cross-link individual entries with their dimension placements for bidirectional navigation.
develop interactive visualizations combining autonomy with entity frame or goal dynamics with discipline clusters.
revisit the taxonomy as llm-native agent platforms evolve through 2026 and beyond.