The Agentic History Museum
Agentic History is an independent documentation effort tracking the long arc of agentic AI and AI agents — from the earliest distributed problem-solving protocols and belief–desire–intention architectures, through the LLM-tool-use papers that defined the modern paradigm, to the 2025 funding wave around autonomous agents.
This site exists to answer one set of questions clearly and with citations: what is the history of agentic AI and AI agents, who pioneered them, and how did the field reach today's autonomous, tool-using systems? Our AI agents timeline is the primary research asset. Primary sources are cited throughout.
If you are looking for a chronological timeline of AI agents, an overview of agentic AI history, a list of pioneers of multi-agent systems and the pioneers of modern multi-agent systems, or direct answers to questions about who invented and pioneered the field, you are in the right place.
Start with the History
Follow the long arc from the 1980 Contract Net Protocol and 1986 Society of Mind, through the BDI agent architectures of the late 1980s and the multi-agent systems era of the 1990s, into the LLM-tool-use breakthrough of 2022, the AutoGPT/BabyAGI moment of 2023, the rise of computer-use agents, and the 2025 capital wave behind autonomous agent platforms.
About the Museum
Who we are and why we are documenting this era — methods, scope, neutrality on contested "firsts," and how to contribute corrections or sources.
Frequently asked questions
Direct answers to the most common questions about AI agent history: who invented AI agents, when did agentic AI begin, what was the first autonomous AI agent, who pioneered multi-agent systems, and how AI agents differ from chatbots and from agentic AI more broadly.
What is an AI agent?
An AI agent is a software system that perceives its environment, decides on actions, and executes them to pursue a goal — typically with some degree of autonomy, memory, and tool use. The textbook formulation comes from Stuart Russell and Peter Norvig's Artificial Intelligence: A Modern Approach (1995), which formally defined the field of AI in terms of rational agents that perceive and act in environments to achieve objectives.
Modern AI agents are most often built on large language models. The model plans, calls tools, observes results, and iterates in a loop with minimal step-by-step human prompting. The technical pattern that powers most current agent frameworks is ReAct — interleaving reasoning traces with actions — introduced by Shunyu Yao and colleagues in October 2022 and widely adopted in production after the OpenAI ChatGPT/Whisper APIs opened in March 2023.
The lineage matters. AI agents are not a 2023 invention. The Contract Net Protocol (Reid G. Smith, 1980) gave us a formal model for how autonomous nodes negotiate task allocation. Marvin Minsky's Society of Mind (1986) argued intelligence emerges from the interaction of many small agents. Michael Bratman's Intention, Plans, and Practical Reason (1987) provided the belief–desire–intention foundation. The 1990s built multi-agent systems on top of those foundations. The 2020s connected that lineage to general-purpose language models and gave it a developer ecosystem.
For a full, dated, cited account, see the timeline.
What is "agentic AI"?
"Agentic AI" is the contemporary umbrella term for systems that go beyond single prompt-and-response interactions to autonomously plan, use tools, and execute multi-step workflows. In practice, agentic AI today refers to LLM-driven systems with planning loops, tool use, memory, and increasingly the ability to operate computer interfaces directly. The term gained prominence in 2024–2025 as model-makers and platforms shifted marketing from "assistants" to "agents." The underlying ideas — autonomy, perception, action, goal-directed behavior — are decades older. For a complete agentic AI history, see the timeline; for the short version, see the FAQ entry on the history of agentic AI.
What is the agentic economy?
The agentic economy is the emerging economic system in which AI agents perform knowledge work, transact, coordinate with other agents, and take consequential actions autonomously — on behalf of humans, businesses, and increasingly other agents.
It is distinct from earlier waves of automation. Industrial automation replaced physical labor with machines that followed fixed programs. Software automation replaced clerical work with rule-based scripts. The agentic economy replaces — or more precisely, augments — cognitive labor with systems that perceive, reason, plan, and act in open-ended environments. The unit of economic activity is no longer the human worker completing a task, nor the software script executing a predetermined routine, but the agent pursuing an objective.
Three structural features define the agentic economy as it is forming in the mid-2020s:
Agents as workers. AI agents are being deployed as autonomous participants in knowledge-work processes — writing code, conducting research, managing customer interactions, drafting legal documents, operating financial workflows — at a cost and speed that has no direct precedent. Cognition's Devin, Sierra's customer-experience agents, and the broader wave of "AI employee" products are the leading commercial expressions of this shift.
Agents transacting with agents. As agent ecosystems mature, agents increasingly interact not only with humans but with each other — negotiating, delegating, verifying, and settling — forming the basis of an agent-to-agent economy. The Model Context Protocol (Anthropic, November 2024), OpenAI's Agents SDK (March 2025), and emerging multi-agent orchestration frameworks are the early infrastructure of this layer.
The agentic stack as economic infrastructure. Just as the internet created an economy built on web infrastructure, the agentic economy is forming around a stack of enabling layers: frontier language models (Anthropic, OpenAI, Google), tool and memory systems, agent orchestration frameworks, computer-use interfaces, and vertical agent applications. Whoever controls the enabling layers — and whoever builds the most-used agents on top of them — shapes the economics of what follows.
The term "agentic economy" is recent, but the underlying forces are the culmination of a trajectory that runs from the Contract Net Protocol's vision of autonomous cooperating nodes (1980) through the BDI architectures of the late 1980s and the multi-agent systems research of the 1990s to the LLM-driven agent platforms of 2023–2026. This museum documents that trajectory. Our news desk and research blog track its unfolding economic implications.
Quick links: Timeline · Funding & Ecosystem · Bibliography · FAQ · Contribute a source · News · Blog