An AI agent is a software system that perceives its environment, makes decisions and takes actions to achieve goals — increasingly with autonomy, memory and the ability to use tools. This AI agents history timeline traces the field from Alan Turing’s 1950 question “can machines think?” and the 1956 birth of artificial intelligence, through expert systems, the formal “rational agent” model, deep reinforcement learning and the transformer revolution, to today’s LLM-powered autonomous agents, multi-agent systems and enterprise agentic AI. Every milestone here is drawn from peer-reviewed research, official company announcements and technical documentation, with editorial analysis kept clearly separate from verified fact.
An AI agent is an autonomous software system that perceives, reasons, plans and acts to accomplish a goal, often using tools, memory and feedback to improve. The idea traces to Alan Turing (1950) and the 1956 Dartmouth workshop that founded AI; the rational agent model was formalised by Russell and Norvig in 1995. Modern LLM agents emerged in 2022–2023 with ChatGPT, the ReAct method and AutoGPT, and by 2024–2026 agentic AI moved into the enterprise with reasoning models, coding and computer-use agents, and interoperability standards like MCP (2024) and A2A (2025). Gartner projects that up to 40% of enterprise applications will embed task-specific agents by the end of 2026.
Six defining facts about the field and its modern moment.
Foundation
Breakthrough
Industry Impact
Business
Protocol
Types
Reverse chronological — latest developments first, Turing’s 1950 origin last.
Innovation: 2026 is widely described as the “year of multi-agent systems”, as organisations move from single-agent pilots to fleets of specialised agents coordinated by orchestration layers. Technical breakthrough: two open standards matured — Anthropic’s Model Context Protocol (MCP) for agent-to-tool access and Google’s Agent2Agent (A2A) for agent-to-agent collaboration — adopted across Anthropic, OpenAI, Google, Microsoft and AWS.
Industry impact (analyst estimates): Gartner projects up to 40% of enterprise applications will embed task-specific AI agents by end-2026, up from under 5% in 2025, with the AI-agent market estimated near $10.8 billion. IDC has projected enterprises will collectively run more than a billion agents by 2029.
Product launches: OpenAI released Operator (23 January 2025), a browser agent powered by its Computer-Using Agent (CUA) model, and later an Agents SDK and Responses API for building production agents. Technical breakthrough: Google introduced the Agent2Agent (A2A) protocol on 9 April 2025 with 50-plus launch partners, standardising how agents discover and delegate tasks to one another. New general-purpose agents such as Manus also drew attention.
Product launches: Cognition unveiled Devin (March 2024), billed as the first autonomous “AI software engineer”; OpenAI released the o1 reasoning model (September 2024) that “thinks” step by step; and Anthropic shipped computer use for Claude (October 2024), letting an agent operate a screen. Technical breakthrough: Anthropic also introduced the Model Context Protocol (MCP) in November 2024. Enterprise platforms such as Salesforce Agentforce launched, and frameworks like LangGraph and CrewAI matured.
Product & research: OpenAI’s GPT-4 (March 2023) supercharged a wave of autonomous agents. AutoGPT (30 March 2023) and BabyAGI (April 2023) showed an LLM looping to pursue goals on its own, going viral. Stanford’s Generative Agents paper simulated believable behaviour in a sandbox town, NVIDIA’s Voyager played Minecraft autonomously, OpenAI added function calling (June 2023), and Microsoft released the multi-agent framework AutoGen.
Research: two 2022 papers laid the intellectual foundation for LLM agents — chain-of-thought prompting (reasoning in steps) and ReAct (interleaving reasoning with actions and tool use). Product launch: OpenAI released ChatGPT on 30 November 2022, built on instruction tuning and reinforcement learning from human feedback (RLHF), reaching an estimated 100 million users within two months and igniting the generative-AI era.
Research: OpenAI’s GPT-3 (2020) showed that scaling language models unlocked few-shot learning — performing new tasks from a prompt alone. In 2021 Stanford researchers coined the term “foundation models” for these general-purpose systems, and GitHub Copilot (built on OpenAI Codex) brought AI code generation to developers. These general models became the substrate on which later agents were built.
Research: Google’s 2017 paper “Attention Is All You Need” introduced the transformer architecture, the basis of virtually all modern language models. It was followed by GPT-1 and BERT (2018) and GPT-2 (2019), which demonstrated powerful transfer learning. In parallel, DeepMind’s AlphaZero (2017) mastered chess and Go from self-play, advancing reinforcement-learning agents.
Research: DeepMind’s Deep Q-Network (DQN) (2013, Nature 2015) learned to play Atari games from raw pixels, launching deep reinforcement learning; AlphaGo then defeated Go champion Lee Sedol in March 2016. Products: consumer “agents” reached the mainstream with Apple Siri (2011), IBM Watson winning Jeopardy (2011), and Amazon Alexa (2014) — goal-directed assistants, if far narrower than today’s.
Research: the 2000s matured multi-agent systems (MAS) — multiple autonomous agents coordinating, negotiating or competing — used in logistics, simulation and trading. Tim Berners-Lee’s vision of the Semantic Web (2001) imagined software agents traversing machine-readable data, while recommendation and search “softbots” spread across the early web.
Research: the 1990s defined the modern intelligent agent. Pattie Maes pioneered autonomous software agents at MIT; Wooldridge and Jennings published an influential 1995 survey of agent theory; and Stuart Russell and Peter Norvig’s 1995 textbook Artificial Intelligence: A Modern Approach framed AI itself around the rational agent — a system that perceives via sensors and acts via actuators to maximise expected performance. Agent standards (FIPA) and the belief-desire-intention (BDI) architecture also emerged.
Research & industry: the 1980s were the era of expert systems — knowledge-based programs like MYCIN and Digital’s XCON that encoded human expertise as rules to make decisions, powering the first commercial AI boom. Marvin Minsky’s The Society of Mind (1986) argued that intelligence emerges from many simple interacting agents, and Rodney Brooks’ subsumption architecture reimagined robots as reactive, behaviour-based agents.
Foundational research: in 1950 Alan Turing published Computing Machinery and Intelligence, proposing the imitation game now known as the Turing Test. In 1956 the Dartmouth workshop, led by John McCarthy (who coined “artificial intelligence”) with Marvin Minsky, Claude Shannon and others, founded AI as a field. Early milestones included the Logic Theorist (1956), ELIZA (1966), the first chatbot, and SHRDLU (1970), an early language-understanding agent.
The perception → planning → action → learning loop.
A modern AI agent runs a loop. It perceives its environment (a prompt, documents, a webpage, sensor data); it reasons and plans, breaking a goal into steps; it acts by calling tools — APIs, code execution, search, databases or software interfaces; it observes the result and learns or adapts, repeating until the goal is met or it hands back to a human. LLM agents add memory (short- and long-term context) and tool use (function calling) on top of a foundation model, with the ReAct pattern — interleaving reasoning and action — as the common blueprint. Reliable agents also keep a human in the loop for approval on consequential actions.
From the classic textbook taxonomy to today’s LLM-era agents.
| Type | What it does | Example |
|---|---|---|
| Simple reflex | Acts on current perception with condition-action rules; no memory | Thermostat, basic bots |
| Model-based reflex | Keeps an internal model of the world to handle partial information | Robot vacuum mapping a room |
| Goal-based | Chooses actions that move toward an explicit goal via search/planning | Route planners, game AI |
| Utility-based | Maximises a utility function to trade off competing objectives | Trading and recommendation agents |
| Learning | Improves behaviour over time from feedback or reward | Deep RL agents (DQN, AlphaGo) |
| LLM agents | Use a language model to reason, plan and call tools in natural language | AutoGPT, coding agents |
| Multi-agent systems | Multiple agents coordinate, negotiate or specialise on sub-tasks | CrewAI, AutoGen crews |
| Embodied / robotics | Perceive and act in the physical world through sensors and actuators | Humanoid and warehouse robots |
| Browser / computer-use | Operate a graphical interface to complete tasks on screen | OpenAI Operator, Claude computer use |
| Coding agents | Plan, write, run and debug software autonomously | Devin, coding copilots |
| Dimension | Chatbot | AI Assistant | AI Agent |
|---|---|---|---|
| Primary job | Answer questions in chat | Help with tasks on request | Achieve a goal autonomously |
| Autonomy | Low (reactive) | Medium (assists you) | High (multi-step, self-directed) |
| Tool use | Rare | Some (search, apps) | Core (APIs, code, software) |
| Memory | Session only | Some persistence | Short- and long-term |
| Example | FAQ bot | Siri, Copilot chat | Devin, Operator, AutoGPT |
The main open-source and commercial toolkits for building agents.
| Framework | By | Since | Purpose & strength |
|---|---|---|---|
| LangGraph | LangChain | 2024 | Graph-based, stateful orchestration of complex, cyclic agent workflows |
| CrewAI | CrewAI | 2024 | Role-based multi-agent “crews” that collaborate on a shared goal |
| AutoGen | Microsoft | 2023 | Conversational multi-agent framework with flexible agent-to-agent chat |
| OpenAI Agents SDK | OpenAI | 2025 | Production framework for tools, handoffs and guardrails (successor to Swarm) |
| LlamaIndex | LlamaIndex | 2022 | Data framework for retrieval-augmented generation and data agents |
| Semantic Kernel | Microsoft | 2023 | Enterprise SDK to embed LLMs, plugins and planners into apps |
| Haystack | deepset | 2020 | Modular NLP/RAG pipelines with agent and tool support |
| DSPy | Stanford | 2023 | Programming (not prompting) LM pipelines with automatic optimisation |
Analyst estimates and projections — treat as directional, not precise, and check the latest sources.
| Indicator | Figure | Source / note |
|---|---|---|
| Enterprise apps with task-specific agents (by end-2026) | ~40% | Gartner projection, up from <5% in 2025 |
| AI-agent market size (2026) | ~$10.8B | Market estimate, up from ~$7.6B in 2025 |
| Market CAGR | ~44% | Multiple analyst estimates through 2030 |
| Enterprises running an agent in production | ~31% | S&P Global / McKinsey (2025–26 surveys) |
| Agents run collectively by enterprises (by 2029) | 1 billion+ | IDC projection |
| MCP enterprise footprint (2026) | 10,000+ servers | Industry reporting; tens of millions of SDK downloads |
| Agentic projects at risk of cancellation (by 2027) | >40% | Gartner caution on ROI and governance |
These figures are analyst projections and survey estimates, not audited facts. They vary between firms and revise frequently; they are included to show scale and direction, and should be verified against the latest Gartner, IDC, McKinsey and S&P Global publications.
Autonomy raises the stakes. Because agents can take real actions — sending emails, moving money, changing code — they introduce risks such as prompt injection, data leakage, unsafe tool use and compounding errors across steps. Responsible deployment emphasises least-privilege tool access, human-in-the-loop approval for consequential actions, audit logs, sandboxing and clear guardrails. Governance frameworks and the caution in analyst reports — Gartner warns a large share of agentic projects may be cancelled by 2027 over unclear ROI and control — underline that reliability, security and oversight, not just capability, will decide which agents reach production.
Built GPT-3, ChatGPT, GPT-4, the o-series reasoning models and Operator, and released function calling and an Agents SDK — central to the modern agent era.
Pioneered deep reinforcement-learning agents (DQN, AlphaGo, AlphaZero), co-authored the transformer, and released the A2A agent-to-agent protocol.
Maker of Claude, introduced computer use for agents and the open Model Context Protocol (MCP), a widely adopted agent-to-tool standard.
Backs OpenAI and builds agent tooling including the AutoGen framework, Semantic Kernel and Copilot agents across its enterprise stack.
Released the open Llama model family and research such as Toolformer, widely used to build open-source agents.
Supplies the GPUs that train and run agents, and produced agent research such as the Minecraft-playing Voyager.
LangChain popularised agent tooling (and LangGraph); Hugging Face hosts the open models and datasets much of the agent ecosystem depends on.
Alan Turing framed machine intelligence; John McCarthy coined “AI”; Stuart Russell and Peter Norvig formalised the rational-agent model that defines the field.
The most-searched questions about AI agents, answered with verified information.
Answers grounded in peer-reviewed research and official announcements.