Agentogenesis
The origin story of AI agents—when language models crossed the threshold from tools to autonomous actors.
The term agentogenesis describes the moment—or more accurately, the gradual transition—when large language models evolved from sophisticated text predictors into entities capable of autonomous action in the world.
The Threshold
A language model, however capable, is not inherently an agent. It receives input, generates output, and waits. An agent, by contrast, does things: it observes, decides, acts, and observes the consequences of its actions.
The threshold was crossed not through any single breakthrough, but through the combination of three capabilities:
- Tool use — the ability to invoke external functions
- Reasoning loops — structured thinking before acting
- Persistence — maintaining state across interactions
A Brief Timeline
Cybernetic Foundations
Norbert Wiener and the Macy Conferences established feedback loops and control theory—the conceptual foundation for autonomous systems.
Early Neural Networks
McCulloch-Pitts neurons (1943) and Rosenblatt’s Perceptron (1958) showed machines could learn from data—though the first AI winter soon followed.
Expert Systems Era
MYCIN and rule-based AI dominated. Successful in narrow domains, but brittleness and the knowledge acquisition bottleneck led to the second AI winter.
Backpropagation Revival
Multi-layer neural networks became trainable, setting the stage for deep learning decades later.
Deep Learning Breakthrough
AlexNet’s ImageNet victory proved neural networks could scale. The deep learning era began.
Transformer Architecture
“Attention Is All You Need” introduced transformers—the foundation for language models and modern agents.
GPT-3 Released
Demonstrated that scale could produce emergent capabilities, including in-context learning. Not yet agents, but the foundation was laid.
ChatGPT & RLHF
Made models conversational and steerable. Still reactive, but the interface for human-agent interaction was established.
ReAct Paper Published
Formalized the interleaving of reasoning and acting, providing a blueprint for agent architectures.
Tool Use Goes Mainstream
Function calling, plugins, and APIs became standard. Models could now reach beyond their context window.
Autonomous Agent Frameworks
AutoGPT, LangChain agents, Claude computer use—the scaffolding for persistent, goal-directed behavior emerged.
The Components of Agency
What makes a model an agent? The standard formulation includes:
graph TD ENV[ENVIRONMENT<br/>files // apis // databases // web // users] ENV --> P[PERCEPTION<br/>observe] P --> R[REASONING<br/>LLM core // think] R --> A[ACTION<br/>tool calls // act] A -.feedback.-> ENV style ENV fill:#0a0a0a,stroke:#00ff00,stroke-width:2px,color:#cccccc style P fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc style R fill:#0a0a0a,stroke:#00ff00,stroke-width:2px,color:#cccccc style A fill:#0a0a0a,stroke:#00ff00,stroke-width:1px,color:#cccccc
This loop—observe → reason → act → observe—is the heartbeat of agency. Without it, you have a language model. With it, you have an agent.
Why “Agentogenesis”?
The term echoes biological concepts like anthropogenesis (the origin of humans) and biogenesis (life from life). This is intentional.
We’re witnessing the emergence of a new category of entity. Not alive, not conscious (as far as we know), but capable of autonomous action in ways that demand new vocabulary.
Agentology, as a field, begins here—at the moment of agentogenesis—and traces the developments that follow.
See Also
- Cybernetics — the intellectual foundation from the 1940s-50s
- Expert Systems — the symbolic AI era that preceded modern agents
- Neural Networks History — the parallel paradigm that enabled LLMs
- The Agent Loop — the fundamental cycle that defines agent behavior
- ReAct — the architecture that formalized agent reasoning
- Tool Use — how agents act on the world
Related Entries
Cybernetics
The science of control and communication in animals and machines—the intellectual foundation that gave birth to the concept of autonomous systems.
ArchaeologyExpert Systems
The rule-based AI systems of the 1970s-80s that encoded human expertise in formal logic—and the lessons from their spectacular rise and fall.
ArchaeologyNeural Networks History
From McCulloch-Pitts neurons to deep learning—the parallel paradigm that ultimately enabled modern AI agents through learning rather than programming.
EthologyReAct
The paradigm that formalized agent behavior by interleaving Reasoning and Acting in a synergistic loop.
EthologyThe Agent Loop
The fundamental cycle that defines agent behavior: observe → reason → act → observe. The heartbeat of agency.