An AI model maps input to output. An AI agent wraps one or more models in software that can choose steps, call tools, inspect results, and continue toward a goal. A robot is an embodied system with sensors and actuators; it may contain an AI agent, but many robots rely mainly on conventional control software.
Model vs agent vs robot
AI model
Receives input and returns a prediction or generated output. It has no independent tool access unless software gives it one.
AI agent
Observes a state, selects an action, uses a tool, evaluates the result, and repeats within a goal and permission boundary.
Robot
Perceives the physical world through sensors and acts through motors or other actuators.
Agentic robot
Combines planning and tool selection with embodied perception, control, and physical safety constraints.
How does an AI agent work?
- Goal: the user or system defines an objective and constraints.
- Observe: the agent receives messages, files, tool results, or environmental state.
- Plan or select: a model proposes the next useful action.
- Act: software calls a permitted tool, such as search, code execution, a calendar, or a database.
- Evaluate: the result returns to the loop so the agent can continue, revise, ask for approval, or stop.
The model is only one component. Ordinary software controls credentials, allowed tools, memory, retries, timeouts, logging, confirmation steps, and final execution. That surrounding design determines much of the system’s real behavior and safety.
How does a robot work?
A robot repeatedly senses, estimates, plans, and controls. Cameras, microphones, encoders, touch sensors, or other devices measure the world. Software estimates the robot’s state, chooses a path or action, and sends commands to actuators. Feedback tells the controller whether the movement had the intended result.
Why do permissions matter?
A text error is inconvenient; the same error connected to email, money, private data, or a motor can cause real harm. Good systems grant the minimum required permission, isolate risky tools, ask before irreversible actions, record what happened, detect unexpected states, and provide a reliable stop or rollback mechanism.
Common questions
Is a chatbot an AI agent?
A basic chatbot that only returns text is usually not an agent. It becomes agentic when it can select and use tools, maintain a goal across steps, and react to results.
Are all robots powered by AI?
No. Many robots use fixed programs, classical control, and carefully engineered rules. AI may assist perception, planning, prediction, or interaction without controlling the entire system.
Can an AI agent act on its own?
Only within the tools, credentials, triggers, and policies that its software provides. “Autonomy” is designed and bounded; it is not magic.
Primary and institutional sources
Research papers explain mechanisms; institutional resources provide guidance. Manufacturer pages describe their own products and are not independent evaluations.
- ReAct: Synergizing Reasoning and Acting in Language Models (2022) — Yao et al.
- AutoGen: Multi-Agent Conversation (2023) — Wu et al.
- RT-2: Vision-Language-Action Models (2023) — Brohan et al. · Google DeepMind
- Unitree Go2: sensors, motors and computing — Unitree Robotics
- AI Risk Management Framework — NIST
- AI Research, Security and Resilience — NIST
- Recommendation on the Ethics of Artificial Intelligence — UNESCO
Continue with the companion in ChatGPT
Explore museum themes, ask questions and test your understanding with quizzes in ChatGPT. The companion and the game keep separate progress.
Discover the ChatGPT companionContent reviewed: 6 September 2026. Navigation and resources updated: 1 October 2026. Project method and limits.