agent definitions — computer science
on this page
overview
computer-science perspectives formalize agents as software entities with states, communication protocols, and execution semantics. the cluster includes agent-oriented programming (aop), bdi architectures, and software engineering roadmaps for multi-agent systems.
signature traits
- mentalistic abstractions: shoham’s agent languages describe beliefs, commitments, and choices as first-class programming constructs.
- interaction protocols: agents communicate via message-passing or performative acts, enabling coordination.
- organizational modeling: AOSE research structures agent societies, roles, and workflows for complex software projects.
illustrative definitions
- 1991 — yoav shoham, AGENT0: introduces a simple agent language where commitments drive autonomous action.
- 1993 — shoham, “agent-oriented programming”: codifies agents with beliefs, obligations, and choices—precursors to BDI logic.
- 2001 — jennings, sycara, wooldridge, AOSE roadmap: surveys agent-oriented software engineering, focusing on interaction and social abilities.
relation to other dimensions
- autonomy spectrum: targets moderate-to-high autonomy—agents act without line-by-line instructions but within specified protocols.
- entity frames: straddle hybrid and machine frames; agents are software but modeled with human-like mental states for design clarity.
- goal dynamics: supports adaptation and negotiation via communication primitives, especially in multi-agent settings.
- persistence & embodiment: typically persistent digital systems; embodiment arises only when agents interface with robots or sensors.
open questions
- how can traditional aop concepts inform contemporary llm-based agent frameworks that rely on prompt engineering instead of explicit mental models?
- which legacy communication standards (kqml, fipa) remain useful in modern orchestration stacks?
- can we reconcile formal agent logics with stochastic behavior from foundation models?