A8gent
GlossaryAgent concepts

What is AI agent?

Also called: autonomous agent, LLM agent

An AI agent is a language model wired to tools and a goal, so it can decide which actions to take, take them, read the results, and keep going until the job is done or it hands off to a human.

Written by A8gent
Contents
  1. At a glance
  2. What an AI agent is
  3. How the agent loop actually works
  4. What agents cost to run
  5. What you should know before you build one
  6. FAQ
  7. Related terms

At a glance

Category
Core agent concept
Minimum ingredients
A model, a goal, tools it may call, and a loop with a stop condition
Differs from a chatbot by
Taking actions in other systems rather than only producing text
Differs from a workflow by
Choosing the sequence of steps at runtime instead of following a fixed path
Typical cost driver
Several model calls per run, each re-reading a growing context
Needs
Guardrails, a step cap, logging, and a human checkpoint on expensive actions

What an AI agent is

An AI agent is a large language model placed inside a loop and given three things a chatbot does not have: a goal, a set of tools it is allowed to call, and permission to keep acting until the goal is met. Ask a chatbot to chase an unanswered quote and it writes you a draft. Ask an agent and it looks up the quote in the CRM, checks whether the customer replied, drafts the follow-up in the right tone, sends it, and logs the activity - because each of those is a tool call it is allowed to make.

The distinction that matters commercially is autonomy over which steps to take, not intelligence. A traditional workflow automation follows a path you drew in advance. An agent is handed a goal and picks the path at runtime. That flexibility is the whole value and the whole risk, which is why every serious build pairs it with guardrails and a human in the loop at the points where a mistake is expensive.

How the agent loop actually works

Strip away the frameworks and every agent runs the same four-beat loop. The model receives the goal, the conversation so far, and a machine-readable list of the tools available. It emits either a final answer or a request to call a tool with specific arguments. Your code runs that tool - a CRM lookup, a calendar write, an SMS send - and appends the result to the conversation. The model reads the result and decides the next move. The loop ends when the model stops asking for tools, or when a step limit you set is reached.

Everything else is packaging. ReAct is a prompting pattern for that loop. A planner-executor splits it into a planning model and a doing model. A multi-agent system runs several loops that hand work to each other. The primitive underneath is unchanged, which is why an agent built on n8n and an agent built in Python behave the same way when they go wrong.

What agents cost to run

An agent costs more per job than a fixed workflow because it makes several model calls per run instead of zero or one, and each call re-reads the whole conversation so far. A five-step agent is not five times the cost of a one-step call - it is closer to fifteen, because the context grows with every step. Budget in tokens, cap the number of steps, and check whether the decision the agent is making is genuinely variable. If the path is always the same, a plain workflow is cheaper, faster and easier to debug.

The other cost is supervision. Agents need observability - a record of every prompt, tool call and result - or you cannot answer the question "why did it do that" when a customer complains. Building that in from day one is cheaper than retrofitting it after the first incident.

What you should know before you build one

Most jobs businesses want automated do not need an agent. Missed-call text-back, appointment reminders and quote follow-up are deterministic sequences; a workflow does them more reliably for less money. Agents earn their keep where the input is messy and the right next step genuinely depends on what the input says - triaging inbound email, qualifying a lead from a free-text form, reconciling a supplier invoice against a purchase order.

Start with one workflow, define done precisely, and instrument it before you widen its permissions. If you want that scoped for you, the free AI automation audit maps which of your processes are agent-shaped and which are better off as plain automations.

AI agent FAQ

What is the difference between an AI agent and a chatbot?

A chatbot produces text and stops. An AI agent produces text and also calls tools - it can read your CRM, book a slot, send a message, update a record. The chatbot tells you what to do next; the agent does it. Practically, that means an agent needs credentials, permissions and an audit trail that a chatbot never required.

Do I actually need an AI agent, or just an automation?

If the steps are the same every time, you need an automation - it is cheaper, faster and never improvises. You need an agent when the right next step depends on interpreting messy input, like an inbound email that might be a complaint, a booking request or spam. A good rule: if you can draw the flowchart without any diamond shapes labelled 'it depends', build the workflow.

How much does it cost to run an AI agent?

The model bill is usually the smaller half. A single agent run makes several model calls and each one re-sends the growing conversation, so a multi-step run can cost ten to twenty times a one-shot call on the same model. On cheap models that is still fractions of a cent per run. The larger cost is the platform, the integrations and the time spent supervising it.

Can an AI agent work without any code?

Yes. n8n, Make, Zapier and several agent platforms expose the agent loop as a node you configure rather than code. You still make every real decision - which tools it may call, what the system prompt says, where a human approves - so the skill required is design, not syntax. Code becomes worth it when you need custom tools or tight control over retries.

What stops an AI agent doing something stupid?

Nothing, unless you build it. The standard controls are a hard cap on loop steps, a narrow tool list with read-only defaults, validation on every argument before a tool runs, a human approval gate on anything that spends money or contacts a customer, and full logging so you can reconstruct any run. Treat the model as a capable new hire with no context, not as a guarantee.

What is the easiest first AI agent for a small business?

Inbound triage. Point it at a shared inbox or web form, have it classify each message, extract the useful fields, and write a structured record into your CRM with a suggested reply for a human to approve. It is high volume, low risk because nothing sends without approval, and it produces an immediate measurable saving in admin minutes.

Primary source

OpenAI - Agents guide

Checked August 2026. Anything priced or rate-limited moves, so verify against the source before you budget.

Want this built rather than explained?

Knowing what ai agent means is one thing; having it running in your business is another. We scope one workflow at a time, quote a fixed price in writing after a free consult, build it, and hand it over with documentation you own.