SIGNAL//DESK
AI/agentssrc: curated AI glossary

Agentic AI

Agentic AI is like a digital assistant that doesn't just answer questions, but can actually complete a multi-step project for you. Instead of you giving it instructions for every single click, you give it a goal, and it figures out the steps, uses tools like a web browser or calculator, and fixes its own mistakes until the job is done.

Agentic AI refers to systems designed to operate autonomously over extended workflows by chaining together reasoning, tool execution, and iterative feedback loops. Rather than a single-turn request-response model, these agents maintain state and memory to plan, execute, and refine their actions to achieve a high-level objective.

The paradigm of building autonomous, multi-step, tool-using AI systems with planning, memory, and self-correction; the agent concept applied as a general operating model.

evolution

  1. 1995 · history
    Introduction of Software Agents

    Pattie Maes and others popularized the concept of autonomous software agents capable of performing tasks on behalf of users.

  2. 2016 · history
    Deep Reinforcement Learning Breakthroughs

    AlphaGo demonstrated the ability of AI to plan complex, multi-step strategies autonomously to achieve long-term goals.

  3. 2023-03 · history
    AutoGPT and BabyAGI Launch

    The release of open-source projects demonstrated the first practical applications of LLMs recursively prompting themselves to complete multi-step objectives.

  4. 2024-09 · history
    OpenAI o1 Series Release

    The introduction of models explicitly trained to use 'chain-of-thought' reasoning to plan and self-correct before providing an answer.


← all terms