←  All terms
K
Frontend & Web Development

AI Agent

An AI agent is a software system that pursues a goal, plans its next steps, and independently uses tools or data to complete work.

An AI agent combines an AI model with context, rules, memory, and executable tools. This allows it to handle multi-step tasks, evaluate results, and respond to new information. Broader permissions increase its usefulness but require access controls, logging, and human approvals.

Read more

Agentic Workflow

Human-in-the-Loop

Governance

Audit Log

FAQ

AI Agent

When does an AI agent beat fixed automation?
An agent pays off for tasks whose path varies and that require judgment about intermediate results. For stable, rule-based processes, fixed automation is cheaper, faster, and easier to verify. In that case the agent is not progress but added complexity. The choice depends on how variable the task is, not on how modern it sounds.
What mistakes do AI agents typically make?
Typical failures are wrong assumptions when context is missing, fabricated details, needlessly broad changes, and loops in which an agent repeats the same step. These errors often look plausible and surface only during review. Logging and clear stop conditions limit the damage. The more rights an agent has, the larger the effect of a mistake.
What permissions should an AI agent be given?
An agent should receive only the rights its task specifically requires, following the principle of least privilege. Read access is often enough, and write access belongs in contained areas such as drafts or branches. Publishing, deletion, and payments should require human approval. Permissions can be widened later, whereas damage cannot be undone.
How does an AI agent work through a multi-step task?
An agent breaks the task into subtasks, performs one step, evaluates the result, and decides on the next. This loop of planning, acting, and checking runs until the goal is reached or a limit applies. Errors in early steps carry forward, which is why checkpoints matter.
What distinguishes an AI agent from a chatbot?
A chatbot answers inputs in conversation, while an AI agent pursues a goal and carries out actions to reach it. It calls tools, reads data, writes results, and checks them. The line is blurry, since many chatbots use tools. What matters is how independently steps are planned and executed.