Clicked Gallery

What are AI Agents?

Highlighted from a real engineering doc. Explained by Clicked.

Used in a sentence

Engineering Notes · AI Systems

The bank is piloting AI agents that reconcile supplier invoices end to end without human review.

The reader highlighted one word in the docs. Clicked broke down the technical term “AI agents” into plain English:

Explained in three depths

Same facts, different vibe — Slang mode 😎

The Clicked way

●○○

Overview

An AI agent is an AI that acts on your behalf instead of just answering you. Give it a goal and it plans, acts with real tools such as a browser or email, checks its results, and keeps going until the job is done. A chatbot replies once; an agent runs a loop.
●○○

Overview

An AI agent doesn't just answer, it acts: goal in, then plan, do, check, repeat with real tools until the thing is done. A chatbot is a reply; an agent is a loop. The loop is the power and the entire risk. 😎

A quick take — often all you need.

●●○

Detail

What separates an AI agent from a chatbot is the loop. A chatbot produces one answer and stops; an agent takes a goal, breaks it into steps, and performs the first step with a real tool, calling an API, filling a form, running code. Then it looks at what happened, decides the next step, and repeats until the goal is met or it gets stuck. That looping is what lets an agent book travel, reconcile invoices, or research and draft a report end to end from a single instruction. It also raises the stakes compared to chat: a wrong answer wastes a moment, but a wrong action, if one happens, can compound as later steps build on it. Agents that read outside content also need protection against instructions hidden in a webpage or email that could try to hijack the loop, an attack called prompt injection. So careful deployments add guardrails: spending limits, approval steps before anything irreversible, and logs of every action taken.
●●○

Detail

Chatbot: one answer, done. Agent: give it a goal and it plans steps, actually executes them, browses, calls APIs, runs code, sends the email, looks at what happened, and picks the next step, around and around until finished. That loop is why one instruction can become booked travel or a reconciled invoice pile. It is also where the risk can live: a bad chat answer wastes a minute, while a bad action can compound, step twelve built on a slip back at step three. And an agent reading the open web needs a bodyguard, because a webpage can hide instructions that try to steer the loop, the prompt injection problem. So sensible deployments add a leash: spending caps, human sign-off before anything irreversible, and a log of every move. 😎

Want more? One click digs deeper.

●●●

Analogy

A consultant versus a contractor. Ask a consultant how to renovate a bathroom and you get an excellent answer, then everything else is on you. Hire a contractor and they take the goal itself: they measure, order the tile, schedule the plumber, hit a surprise behind the drywall, adjust the plan, and come back when there is a working bathroom. You traded effort for trust, which is exactly why contracts have budgets, checkpoints, and sign-offs before the wall gets knocked down.
Deepest level reached.
●●●

Analogy

Asking an intern to find out which venues are free is one thing; handing over the company card and saying get the offsite booked is another, and the second one is the agent. Off they go: emailing venues, comparing quotes, putting down a deposit, magic when it works. But every step taken without checking in is built on the previous one, so if they do book the wrong Springfield, you find out after the deposit and the calendar invites. Great interns get a budget cap and one rule: ask before anything you can't undo.
Deepest level reached.

Unfamiliar concept? A real-world example makes it click — fresh analogies on tap.

AI explanations may contain errors · Not professional advice

Formal definition — The same term, explained the usual way

AI agents are systems in which a language model operates in an iterative loop of planning, tool invocation, and observation to pursue a specified objective with limited human intervention. Distinguishing features include multi-step autonomy, access to external tools and environments, and state carried across steps; principal risks include error compounding, misaligned actions, and prompt injection via untrusted content, motivating guardrails such as approval gates and action logging.

Want Clicked to explain terms like “AI agents” directly in your browser — including on PDFs?

Add to Chrome — Free

50 free Explanations · No credit card required