Types of AI agents: the 5 classic types and the ones businesses use

There are two useful ways to sort AI agents: the classic types from AI textbooks, which describe how an agent decides, and the practical types businesses use, which describe what job an agent does. Knowing both helps you understand vendor claims and pick the right tool. New to the topic? Start with what is an AI agent.

The 5 classic types of AI agents

1. Simple reflex agents

Act only on the current input using fixed if-then rules. Example: a phone menu that routes "press 2" to billing. Fast and predictable, but helpless when something unexpected happens.

2. Model-based reflex agents

Keep an internal picture of the situation, so they can handle information that is not visible in the current input. Example: a chat agent that remembers the customer already gave their order number earlier in the conversation.

3. Goal-based agents

Choose actions that move them toward a goal, planning several steps ahead. Example: "get this caller booked for a cleaning this week": check the calendar, offer slots, confirm.

4. Utility-based agents

When several actions reach the goal, they pick the best one according to a score. Example: offering the slot that fits the customer's preference and keeps the dentist's day efficient.

5. Learning agents

Improve over time from feedback. Example: an agent whose answers are refined each week based on conversations the team flagged.

Type	Decides using	Strength	Weakness
Simple reflex	Fixed rules	Predictable	Breaks on new situations
Model-based	Rules + memory of state	Handles context	Still rule-bound
Goal-based	Plans toward a goal	Flexible	Needs good goals and limits
Utility-based	Best-scoring option	Makes trade-offs	Scoring must be right
Learning	Feedback over time	Gets better	Needs monitoring

Where today's LLM agents fit

Agents built on large language models do not map neatly to one type. They hold conversation state (model-based), work toward goals (goal-based), weigh options (utility-based) and are improved through feedback and updated instructions (learning). That mix is why they feel so much more capable than older bots. See what is an LLM agent.

Practical types of AI agents for businesses

• Receptionist agents: answer calls and messages, capture details, route requests. Learn more.
• Customer service agents: resolve common questions and order issues. Learn more.
• Sales agents: qualify leads, follow up, book meetings. Learn more.
• Scheduling agents: book, reschedule, remind.
• Voice agents: any of the above, over the phone. Learn more.
• Back-office agents: email triage, data entry, reporting.
• Coding and research agents: mostly used by technical teams.

Single agents vs multi-agent systems

Some setups use one agent for everything; others split the work between specialised agents that pass tasks to each other, for example a receptionist agent that hands a qualified lead to a quoting agent. Read multi-agent systems explained.

Each type in a front-office example

To make the textbook types concrete, here's how each would handle the same request: "I need to move my appointment."

Type	How it handles the request
Simple reflex	Matches the word "move" and replies with the reschedule phone number. Nothing more.
Model-based	Remembers the customer already gave their name, so it doesn't ask again; still follows a fixed script.
Goal-based	Aims to get the appointment moved: finds the booking, checks free slots, offers options, confirms.
Utility-based	Like goal-based, but prefers slots that suit both the customer's stated preference and the practitioner's schedule.
Learning	Notices over time that many customers ask for early mornings and suggests those first (after the team approves the change).

Other ways to classify AI agents

By autonomy

From agents that only suggest, to agents that act with approval, to agents that act within limits, to fully autonomous agents. See autonomous AI agents.

By interface

• Voice agents: phone calls and voice assistants. Read more.
• Text agents: chat, WhatsApp, SMS, email.
• Computer-use or browser agents: operate software interfaces like a person would.

By scope

• Horizontal agents: general-purpose, used across many tasks.
• Vertical agents: built for one industry or job, with domain knowledge and workflows baked in, such as a dental booking agent or an HVAC intake agent.

By organisation

• Single agent doing one job.
• Hierarchical agents: a supervisor agent directs worker agents.
• Collaborative agents: peers that pass work to each other. See multi-agent systems.

What are vertical AI agents?

Vertical AI agents are specialised for a particular industry or function. Because they come with the right questions, workflows and guardrails for that field, they usually work better out of the box than general agents. An AI receptionist set up for clinics knows to ask about new-patient status and never to give medical advice; one set up for manufacturers knows to ask about quantities, sizes and delivery deadlines. Owlena follows this approach with industry-specific setups.

Choosing the right type for the job

Job	Best fit
Route calls to departments	Simple reflex or model-based (or a basic menu)
Answer FAQs	Model-based with good knowledge
Book, order, qualify	Goal-based with tools and rules
Optimise schedules or routing	Utility-based
Improve over months	Any type plus a human-reviewed learning loop

The PEAS framework: describing any agent

AI textbooks (notably Russell and Norvig's Artificial Intelligence: A Modern Approach) describe an agent's task with PEAS: Performance measure, Environment, Actuators, Sensors. Applied to an AI receptionist:

PEAS	AI receptionist
Performance	Calls answered, bookings made, correct details, customer satisfaction
Environment	Phone lines, messaging apps, calendar, customers
Actuators (actions)	Speak or write replies, create bookings, send messages, transfer calls
Sensors (inputs)	Speech, text, calendar data, customer records

Writing a PEAS table is a surprisingly good way to specify any business agent before building or buying it.

Environment types that affect agent design

• Fully vs partially observable: can the agent see everything it needs? A phone agent can't see the caller's face or their calendar, so it must ask.
• Deterministic vs stochastic: do actions always have the same result? Booking a free slot usually does; persuading a customer doesn't.
• Static vs dynamic: does the world change while the agent thinks? Calendars change as others book, so the agent must re-check before confirming.
• Single vs multi-agent: are other agents (or people) acting in the same environment? Front desks share calendars with staff.

Customer-facing environments are partially observable, dynamic and involve people, which is exactly why goal-based agents with confirmation steps and hand-overs work best there.

Simple agents still have a place

Not every job needs a language-model agent. A simple reflex rule ("if the message says STOP, unsubscribe") is faster, cheaper and more reliable for that job. Good systems mix types: simple rules for the fixed parts, goal-based agents for conversations. See agentic AI vs traditional automation.

Which type does a small business need?

Usually a goal-based customer-facing agent with strict limits: it should pursue clear goals (answer, capture, book) but never improvise on price, promises or sensitive topics. Owlena is built that way: one assistant across your channels that follows your rules and hands decisions to your team.