What is an AI agent?
Last reviewed 7 October 2026
An AI agent is a software program that uses an AI model to decide what to do next and then acts through tools to reach a goal it was given. AWS describes an agent as software that interacts with its environment, collects data and uses that data to perform self-directed tasks. Today most agents use a large language model as the part that reasons.
Also called: AI agents, intelligent agent, autonomous agent, software agent.
Why AI agent matters
The agent is the unit you deploy, govern and pay for. A company does not buy agentic AI in the abstract; it runs a support agent, a reconciliation agent or a research agent. Each has a job, a set of tools, a budget and someone accountable for it. Treating the agent as a named thing with an owner is what makes it possible to say what it may do, and to check afterwards what it did.
The definition also matters because older software used the same word. AI textbooks described reflex agents and goal-based agents long before language models existed, and AWS still lists those types. When a vendor says “agent” today it usually means a model-driven program with tools, but it is worth asking whether the product decides anything at all or simply runs a fixed script.
How it works
An agent has four working parts. A model reasons about the goal. Instructions tell it its role and limits. Tools let it read and change other systems. Memory holds what it has learned during the current task, and sometimes across tasks. AWS adds a planning module that breaks a goal into steps, which many agents implement as part of the model’s own reasoning.
At run time the agent receives a goal, looks at its context and either answers or asks to call a tool. The runtime executes the call, returns the result, and the agent decides again. Anthropic describes agents as systems where an LLM dynamically directs its own process and tool use, with stopping conditions such as a maximum number of iterations to keep control.
The quality of an agent depends more on its tools and instructions than people expect. Anthropic advises putting as much effort into the agent-computer interface (tool names, descriptions and parameters) as teams put into user interfaces. A vague tool description produces vague tool calls.
Example: an IT access request agent
An employee writes “I need read access to the finance reporting workspace”. The access agent checks the employee’s department in the HR system, looks up which group grants that workspace, and reads the access policy, which says finance data needs a manager’s approval.
The agent prepares the change, sends the request to the employee’s manager and waits. When the manager approves, it adds the user to the group, confirms the change took effect by reading the group membership back, and replies to the employee. If the manager rejects, it closes the request with the reason. Every tool call and the approval are recorded against the request.
How Swfte relates to it
Built in the product
Swfte Studio is built to create agents like this. You write the agent’s instructions, choose its model from those reachable through Swfte Connect, attach tools through integrations, and place it inside a workflow that can pause at an approval step for a named person. Runs record traces, logs and token cost per step.
Swfte Cortex brings agents to the desktop. Its assistants answer from on-device knowledge bases, and tool calls that send, pay, delete, push code or run shell commands are held at a high-risk gate for a person to confirm. Swfte does not offer autonomy levels as a setting today; that is designed, not built.
- Swfte Studio: Agents and workflows
- Swfte Cortex: A governed AI desktop
- AI agent builder guide
Related terms
- Agentic AI
Agentic AI is a style of AI system that works toward a goal with limited supervision, choosing its own next steps and calling tools to act on other systems.
- LLM agent
An LLM agent is an agent whose decisions are made by a large language model, which reads a goal, reasons about it and chooses actions such as tool calls in a repeated loop.
- Function calling
Function calling is a feature of language model APIs that lets a model return a structured request to run a function you defined, with arguments that match its schema, instead of only replying in text.
- Multi-agent system
A multi-agent system is a set of AI agents that work together on a task, each with its own role, instructions and tools, coordinated by a supervisor agent or by rules for passing work between them.
- Human-in-the-loop (HITL)
Human-in-the-loop (HITL) refers to a system in which a person takes an active part in the operation, supervision or decisions of an automated process at defined points.
Common questions
- What is the difference between an AI agent and a chatbot?
- A chatbot answers messages. An agent can also act: it calls tools, changes records and decides its next step from what it finds. Some chatbots now call tools, so the line is blurry. A useful test is whether the software can take an action in another system without a person copying the answer across.
- What is the difference between an AI agent and agentic AI?
- An AI agent is a specific program with a job, tools and an owner. Agentic AI is the broader description of systems that pursue goals with limited supervision, which can include several agents and fixed workflow steps. You build and deploy agents; agentic is the quality they share.
- Do AI agents learn on their own?
- Usually not in the way the phrase suggests. Most production agents do not retrain their model while running. They keep notes in memory during a task, and some store facts for later tasks. AWS lists continuous learning as a principle of agents, but in practice improvement mostly comes from people changing instructions, tools and tests.
- How many tools should one agent have?
- As few as the job needs. Each extra tool is another choice the model can get wrong and another permission to govern. If an agent needs many unrelated tools, split it into smaller agents with clear roles, or use a fixed workflow for the parts that never change. Clear tool descriptions matter more than tool count.
Sources
Definitions on this page were read on the sources below on 7 October 2026. Where sources define the term differently, the page says so. The full glossary lists more terms.