Make money doing the work you believe in
Wooldridge and Jennings 1995: agents had a theory before they had LLMs
In 1995, Michael Wooldridge at Manchester and Nick Jennings at Queen Mary University of London published "Intelligent Agents: Theory and Practice" in The Knowledge Engineering Review 10:115-152. The paper synthesized roughly a decade of multi-agent systems research into a definitional framework. Intelligent agents had four properties: autonomy, reactivity, proactivity, and social capability. The Belief-Desire-Intention model, developed by Bratman as philosophy in 1987 and formalized for agents by Rao and Georgeff at the Australian AI Institute in 1991, became the dominant architecture for deliberative agents.
By the late 1990s the multi-agent systems community had formal logics for agent reasoning such as BDICTL, agent communication languages including KQML and FIPA-ACL, coordination protocols like contract net and auctions, and deployed applications in factory process control, business process management, and the NASA fault diagnosis system for the space shuttle. Two textbooks codified the field: Wooldridge's An Introduction to MultiAgent Systems and Russell-Norvig's agent-centric framing across multiple editions of Artificial Intelligence: A Modern Approach.
The 2023-2026 LLM agent wave proceeded with limited engagement with this prior work. AutoGPT, BabyAGI, LangChain agents, and the current MCP-driven generation rediscovered problems the MAS community had named, attacked, and partially solved by 2005. Some researchers have noticed; the arXiv preprint "Agentic AI and Multiagentic: Are We Reinventing the Wheel?" in mid-2025 argued the new vocabulary mostly relabels old concepts.
The implication is not that LLM agents should adopt BDI verbatim. The implementations differ. The implication is that the conceptual problems, defining what an agent is, when it is reasonable, how it commits to plans, how it coordinates with others, were charted before. Walking the territory without the older map produces repeated discoveries of old failure modes at production scale.
