Timeline
The History of AI Agents
2000 Years in the Making
The concept of an AI agent didn't emerge in 2023 with ChatGPT. It's the culmination of over two thousand years of thinking about what it means for one entity to act on behalf of another—from Roman law to philosophy of mind, from economics to the LLM revolution.
Pre-1957
Agency Law Origins
Legal foundations establishing principal-agent relationships, authority, and liability.
Mandatum & Procurator
Roman law establishes 'mandatum' (gratuitous agency contract) and 'procurator' (legal representative), formalizing one person acting on behalf of another.
Mercantile Agency
Italian and English merchants formalize agency in commercial transactions during the Middle Ages.
Industrial Expansion
Industrial Revolution necessitates clearer rules for delegation and liability as commerce expands.
1957-1970
Philosophical Foundations
Philosophers examine what makes actions intentional and goal-directed.
Anscombe: Intention
G.E.M. Anscombe establishes that actions are 'intentional under a description'—we understand actions by understanding reasons.
Restatement (Second) of Agency
American Law Institute defines agency relationship: principal manifests assent that agent acts on their behalf with control.
Davidson: Causal Theory
Donald Davidson argues intentional actions are explained by belief-desire pairs that causally produce behavior.
1970-1989
Economic & Social Theory
Economics, psychology, and cognitive science each adapt agency to their disciplinary concerns.
Milgram: Agentic State
Stanley Milgram introduces the 'agentic state'—individuals as instruments carrying out another's wishes. Anchors one end of the autonomy spectrum.
Jensen & Meckling: Principal-Agent
Formalizes principal-agent relationships as contracts with delegated authority. Establishes vocabulary of agency costs and information asymmetry.
Minsky: Society of Mind
Mind as a 'society' of simple agents whose interactions generate intelligence—the multi-agent system metaphor.
Bratman: BDI Architecture
Belief-Desire-Intention framework becomes foundational for intelligent agent architectures. Plans structure practical reasoning.
Dennett: Intentional Stance
Pragmatic criterion: treat entities as agents when attributing beliefs and desires yields reliable predictions.
Bandura: Human Agency
Defines human agency through intentionality, forethought, self-regulation, and self-reflectiveness—properties increasingly sought in AI.
1990-1999
Computer Science Revolution
Agency transforms from philosophical concept to computational implementation.
Shoham: Agent-Oriented Programming
AGENT0 implements agents as computational entities with mentalistic terms—beliefs, commitments, obligations.
Maes: Learning Interface Agents
Adaptive software assistants that learn user preferences. Pattie Maes introduces adaptation as core agentic capability.
Wooldridge & Jennings: Intelligent Agents
Defines agents through four properties: autonomy, reactivity, proactivity, and social ability.
Russell & Norvig: Perception-Action
AI: A Modern Approach defines agents as perceiving through sensors and acting through actuators. Becomes standard AI education framework.
Epstein & Axtell: Agent-Based Modeling
Growing Artificial Societies establishes ABM—autonomous individuals whose local interactions produce emergent macro patterns.
Sutton & Barto: Reinforcement Learning
Agents maximize cumulative reward through environmental interaction. Framework proves foundational for deep RL decades later.
Multi-Agent Systems Textbooks
Ferber and Weiss consolidate MAS—autonomous entities that cooperate, compete, and negotiate to achieve goals.
2000-2019
Consolidation & Validation
Frameworks mature through education, reference works, and demonstrations at unprecedented scale.
Kauffman: Physical Agency
Proposes thermodynamic criteria for autonomous agents, highlighting tension between embodied and virtual agency.
Restatement (Third) of Agency
Updates legal doctrine while retaining core concepts of fiduciary duty and delegated authority.
Wooldridge: MAS Education
Introduction to Multi-Agent Systems trains new generations of researchers and practitioners.
Deep Reinforcement Learning
DeepMind's DQN masters Atari from raw pixels. AlphaGo defeats world Go champions. Validates RL framework at scale.
2020-2025
The LLM Era
Large language models enable agent architectures qualitatively different from hand-coded systems.
GPT-3 Release
Demonstrates few-shot learning and flexible reasoning at scale. The foundation for LLM-as-agent patterns.
ReAct: The Turning Point
Yao et al. show LLMs can interleave reasoning with tool use—iteratively planning, acting, and observing. Catalyzes the LLM-as-agent pattern.
ChatGPT Launch
Conversational AI goes mainstream. Mass audience exposure to LLM capabilities.
Generative Agents
Park et al. demonstrate LLM agents with memory, reflection, and planning sustaining believable behavior over extended time horizons.
Commercial Frameworks
LangChain, OpenAI Agents SDK, and Claude tools embed iterative tool orchestration as default infrastructure.
Five Patterns Across Seven Decades
Looking across this history reveals recurring themes that explain why contemporary definitions vary.
Broadening Entity Frames
From humans and human-AI relationships → purely computational agents (1990s) → hybrid framings where humans provide goals and AI executes (2020s).
Increasing Autonomy
Delegated proxies (1958) → perception-action systems (1995) → tool orchestrators (2025). Decision-making shifts progressively toward the agent.
Mental States → Behavior
Philosophy emphasized intention and reasoning. CS implemented these as constructs. Contemporary definitions focus on observable capabilities: tool use, iteration, task completion.
"Tools-in-a-Loop" Convergence
Rapid standardization on iterative tool orchestration. From research demonstration (ReAct 2022) to industry standard in under three years.
Hand-Coded → Learned
Explicit rules → Deep RL learning → LLMs. Decision-making knowledge shifts from programmer-specified logic to learned parameters.
Why It Matters
These patterns explain why agent definitions vary: they reflect different points in this trajectory and different disciplinary emphases.
Key Definitions Through History
Essential vocabulary that emerged from each era and remains relevant today.
| Term | Era | Definition |
|---|---|---|
| Agency Relationship | Legal (Pre-1957) | Fiduciary relationship where one party authorizes another to act on their behalf |
| Intentional Description | Philosophy (1957) | Description of action that captures the purpose or goal behind behavior |
| Principal-Agent | Economics (1976) | Contract where principals engage agents with delegated decision-making authority |
| BDI Architecture | Cognitive (1987) | Belief-Desire-Intention framework structuring practical reasoning through partial plans |
| Perception-Action Agent | CS (1995) | Entity that perceives environment through sensors and acts through actuators |
| Multi-Agent System | CS (1999) | Computational system of multiple interacting intelligent agents |
| LLM-as-Agent | Modern (2022) | LLM combined with tool access, memory, and goal structures creating autonomous agency |
Related Glossary Terms
Agent
A system exhibiting the three foundational properties of Goal, Perception, and Action (GPA). An agent pursues objectives, observes its environment, and takes actions to achieve its goals. This represents Level 1 in the three-level hierarchy.
BDI Architecture
Belief-Desire-Intention framework for structuring agent reasoning. Agents maintain beliefs (knowledge about the world), desires (goals they want to achieve), and intentions (committed plans of action). BDI provides a foundation for understanding how agents reason and decide.
Multi-Agent System (MAS)
A system where multiple autonomous agents interact, cooperate, or compete to achieve individual or collective goals. Examples include trading systems with multiple algorithms, distributed due diligence teams, or coordinated compliance monitoring.
Reinforcement Learning (RL)
A machine learning approach where agents learn optimal behavior through trial and error, receiving rewards or penalties for their actions. RL agents discover effective strategies without explicit programming, raising governance questions about learned behaviors.
LLM-as-Agent Pattern
The contemporary architectural approach where a large language model iteratively orchestrates tool calls, observes results, and adapts its strategy to achieve goals. This pattern underlies most modern AI agents in professional applications.
Principal-Agent Relationship
An economic framework analyzing relationships where principals engage agents with delegated decision-making authority. Focuses on incentive alignment, information asymmetry, and agency costs. Foundational for understanding AI alignment challenges.
Dive Deeper
This timeline is adapted from Chapter 1 of Agentic AI in Law and Finance, which includes detailed citations, disciplinary analysis, and practical applications of this historical understanding.