If you have spent any time around AI lately, you have probably heard the word agent thrown around constantly. AI agents are suddenly everywhere. One company says their agent can research and write reports. Another says their agents can book meetings, update CRM records, and send follow-ups. Meanwhile, plenty of products that are really just chatbots or automation tools are also being marketed as agents.
So what actually counts as an AI agent? And what are the different types of AI agents people are talking about?
What is an AI agent?
An AI agent is a software system that can understand a goal, perceive context, reason about what to do, and take actions to move toward that goal. That is what separates an agent from a basic chatbot.
A chatbot mostly responds to prompts. An agent can take action.
The strongest current definitions all point in roughly the same direction:
- Anthropic distinguishes between workflows, where code determines the path, and agents, where the model dynamically directs its own process and tool use.
- Google Cloud defines AI agents as systems that pursue goals on behalf of users and show reasoning, planning, memory, and some autonomy.
- Microsoft describes an agent as a system that perceives its environment through inputs and takes actions to achieve a defined objective.
- IBM defines AI agents as systems capable of autonomously performing tasks on behalf of a user or another system.
Put simply: An AI agent is not just something that talks. It is something that can decide and do.
AI agent vs chatbot vs automation
This is where a lot of confusion comes from.
Chatbot
A chatbot is mainly designed to respond to user input in conversation. Examples:
- Answering questions
- Summarizing text
- Drafting messages
- Explaining concepts
A chatbot may feel smart, but it usually stops at the response.
Automation
Automation follows predefined rules. Examples:
- If a form is submitted, send an email
- If a lead changes stage, assign a task
- If a meeting is booked, create a calendar event
Automation is powerful, but it is not intelligent in the same way. It typically does not reason through a situation in real time.
AI agent
An AI agent combines intelligence with action. Examples:
- Research a company, summarize findings, and draft a tailored outreach email
- Read an incoming email, decide whether it is urgent, and route it appropriately
- Review a job posting, compare it to your background, and recommend whether it is worth pursuing
The important distinction is that an agent is working toward an objective, not just firing a rule or replying with text.
How do AI agents work?
Most modern AI agents follow some version of a loop like this:
- Observe the current situation or inputs
- Reason about what matters
- Plan the next step or sequence of steps
- Act using available tools
- Review the result and adjust if needed
Microsoft describes AI agents as following a perception-reasoning-action loop, and Google highlights reasoning, acting, observing, and planning as core features. More advanced systems also add memory, collaboration, and self-refinement.
In plain English: an agent looks at the situation, decides what to do, does it, sees what happened, and keeps going if needed.
The classic 5 types of AI agents
If you search for “different types of AI agents,” you will often find the classic AI taxonomy. IBM summarizes five main types:
1. Simple reflex agents
These are the most basic. They respond directly to current conditions using fixed rules. Examples:
- A thermostat turning heat on or off based on temperature
- A basic spam filter blocking messages based on predefined patterns
These systems are useful in predictable environments, but they do not remember the past or plan for the future.
2. Model-based reflex agents
These agents still use rules, but they also maintain an internal model of the world. That means they can track state, not just immediate input. Examples:
- A robot vacuum that remembers which areas it has already cleaned
- A navigation system that tracks known obstacles and room layout
Best for: stateful environments where memory matters.
3. Goal-based agents
These agents evaluate actions based on whether they help achieve a goal. Examples:
- A route planner trying to find the best way to get from point A to point B
- A sales assistant agent trying to move a lead toward a booked demo
Goal-based agents are more flexible because they are not only reacting. They are choosing actions in service of an objective.
4. Utility-based agents
These agents do not just ask, “Does this help achieve the goal?” They ask, “Which option is best among competing tradeoffs?” Examples:
- A delivery system balancing speed, fuel cost, and reliability
- An AI assistant deciding whether to interrupt you now or wait until later based on urgency and context
Best for: optimization under competing priorities.
5. Learning agents
These agents improve over time based on experience and feedback. Examples:
- Recommendation systems that improve as users interact with them
- Fraud detection systems that get better as they see more examples
- Assistants that learn which actions are most useful in a given workflow
A learning agent is not stuck with the same rules forever. It adapts.
The modern types people actually mean
The classic taxonomy is still useful, but in practice most people talking about AI agents today mean a more modern set of categories. Here is a simpler practical breakdown.
1. Prompt-and-response agents
The lightest version. They take in a prompt and produce a useful result, sometimes with some tool use, but they are still fairly session-based. Examples:
- A writing assistant
- A meeting-note summarizer
- A customer-support responder
2. Workflow agents
These operate within predefined paths. Anthropic makes an important distinction here: workflows are agentic systems where the LLM and tools are orchestrated through predefined code paths, while true agents dynamically direct their own process. Examples:
- A support flow that always classifies, drafts, escalates, and logs in the same order
- A hiring flow that always reviews a resume, scores fit, and drafts follow-up questions
Workflow agents can be very effective and often more predictable than open-ended agents.
3. Tool-using agents
These are the systems most people picture when they hear “AI agent.” They can call tools like web search, APIs, calendars, CRMs, file systems, and internal databases. Examples:
- An agent that researches prospects and updates a CRM
- An agent that reads a contract and flags unusual clauses
- An agent that analyzes job descriptions and drafts tailored outreach
4. Multi-agent systems
These systems use multiple agents that each specialize in a different part of a task. One agent might research. Another might write. Another might validate findings. Examples:
- A design tool where multiple agents build different parts of a prototype
- A research system where several agents gather evidence in parallel and one synthesizes the answer
5. Persistent agents
These agents work over time, not just in one session. They remember the goal, retain state, and carry work forward across days or weeks. Examples:
- An agent managing an ongoing job search
- An agent helping a founder prepare for fundraising over several weeks
- An agent guiding a customer-success process from onboarding through renewal
6. Adaptive agents
These agents do not just continue over time. They also change strategy based on what is happening:
- An agent that notices outreach is not getting replies and shifts the playbook
- An agent that sees interview traction but no offers and pivots toward interview coaching
- An agent that reprioritizes based on urgency, feedback, or changing constraints
This is where AI starts to feel more strategic.
What makes an AI agent good?
Not every AI agent is good just because it can call tools. The most useful agents usually have a few things in common:
- Clear goals - If the objective is vague, the output often gets vague too.
- The right tools - An agent is only as useful as the systems it can access.
- Strong context - Agents need the right information at the right time.
- Memory or state - Many real-world tasks require continuity.
- Feedback loops - The agent should be able to learn what worked.
- Guardrails - Human oversight, permissions, and limits matter a lot.
Anthropic explicitly recommends starting with the simplest possible solution and only increasing complexity when needed, because more agentic systems often trade predictability, latency, and cost for flexibility.
When should you use an AI agent?
Use an AI agent when the job:
- Has multiple steps
- Benefits from reasoning and tool use
- Changes based on context
- Requires flexibility rather than rigid rules
- Has a clear enough objective to guide decisions
Examples: sales prospecting, customer support triage, recruiting coordination, financial analysis, research and reporting, job search management.
Do not assume you need an agent when a simple prompt, retrieval system, or workflow automation would do the job more reliably and cheaply.
Why AI agents matter
AI agents matter because they shift software from passive tools to active systems. Instead of just helping you think, they can help you move.
That does not mean every agent should be fully autonomous. But it does mean software is starting to move from:
- Answering
- To assisting
- To acting
- To adapting
That is a major shift.
Final takeaway
If you remember one thing, make it this:
A chatbot talks. Automation follows rules. An AI agent works toward a goal.
And as AI keeps improving, the most important differences between agents will come down to a few key questions:
- Can it take action?
- Can it use tools?
- Can it remember context?
- Can it adapt?
- Can it coordinate multiple steps or agents?
- Can it improve over time?
Those are the questions that separate a flashy demo from a genuinely useful system. At Offboard, we think that distinction matters a lot - especially in areas like job search, where people do not just need answers. They need systems that can help them manage a mission over time.