AI Agents

A practical guide to understanding what AI agents are and how they differ from standard AI features.

What AI agents are, how they work, and what product and design teams need to understand when building or evaluating agentic AI features.

22 May 20265 min read

What it is

An AI agent is a glossarySystemA system is a collection of interconnected components that work together to achieve a specific function or outcome.Open glossary term that uses a language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term to plan and execute a sequence of actions to complete a goal, rather than simply responding to a single glossaryPromptA prompt is the input or instruction given to an AI system to guide its output or response.Open glossary term.

Where a standard AI glossaryInteractionInteraction refers to any action a user takes within a product and how the system responds. It includes clicks, taps, gestures, and inputs that drive the user experience.Open glossary term involves a user asking a question and the glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term providing an answer, an agent can take that goal and break it down into steps, use tools to gather information or take actions, and iterate until the task is complete.

An agent might glossarySearchSearch is the functionality that allows users to find content or information by entering queries. It relies on indexing, metadata, and relevance algorithms to return useful results.Open glossary term the web, read a document, write and run code, fill in a form, send a message, or call an external glossaryServiceA service is a component or function that performs a specific task within a system.Open glossary term — all as part of completing a single instruction.

The key distinction is autonomy. An agent acts, rather than just responding.

This makes agents powerful for complex, multi-step tasks. It also introduces new design challenges around glossaryTrustUser confidence that a product, service, or organisation will do what it promises.Open glossary term, oversight, and failure handling that do not exist in simpler AI glossaryInteractionInteraction refers to any action a user takes within a product and how the system responds. It includes clicks, taps, gestures, and inputs that drive the user experience.Open glossary term.

When to use it

Understand when agentic AI is appropriate.

It is most relevant when:

A task involves multiple steps that would otherwise require human coordination

The task is repeated frequently enough to justify automation

The AI needs to interact with external tools, data, or systems

Human effort on the task is high relative to the value of each instance

It is less relevant when:

A task can be completed with a single prompt and response

The task involves high-stakes, irreversible actions that need human sign-off

The workflow is unpredictable enough that structured automation is unreliable

Key takeaway

Agents are powerful when tasks are structured, repetitive, and recoverable. They are risky when tasks are ambiguous, high-stakes, or difficult to reverse.

How it works

Understand the basic mechanism. An AI agent uses a language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term as its reasoning engine. Given a goal, the model decides what steps to take, calls the appropriate tools, observes the results, and decides what to do next.

This loop — observe, plan, act, observe — continues until the goal is achieved or the agent reaches a point where it cannot proceed.

Agents require access to tools, which might be web glossarySearchSearch is the functionality that allows users to find content or information by entering queries. It relies on indexing, metadata, and relevance algorithms to return useful results.Open glossary term, code execution, file reading, API calls, or any other glossaryCapabilityCapability refers to an organisation’s ability to perform a specific function or deliver a particular outcome.Open glossary term defined by the glossarySystemA system is a collection of interconnected components that work together to achieve a specific function or outcome.Open glossary term. The agent decides which tools to use and when.

What this means for designers and product teams. Designing an AI agent means designing the glossaryWorkflowA workflow is a defined sequence of tasks or steps required to complete a process.Open glossary term, the tools available, the glossaryConstraintsConstraints are limitations or restrictions that impact how a product or solution can be designed or built.Open glossary term on what the agent can do, and the points at which a human should be involved.

Agents can fail in novel ways. They can take wrong turns mid-task, use tools incorrectly, or produce errors that compound across steps. Unlike a single wrong answer, a wrong action taken early in an agentic glossaryWorkflowA workflow is a defined sequence of tasks or steps required to complete a process.Open glossary term can create problems that are difficult to unwind.

What to look for

Focus on:

Task boundaries — whether the agent's scope is clearly defined

Tool access — whether the agent has access only to what it needs

Failure handling — what happens when a step fails or produces unexpected results

Human checkpoints — where human review or approval is required

Reversibility — whether actions taken by the agent can be undone if needed

Where it goes wrong

Most issues come from: Giving an agent broad access and minimal oversight is not a design decision — it is a risk.

Poorly defined task boundaries leading to unpredictable behaviour

Excessive tool access beyond what the task requires

No checkpoints for human review at high-risk steps

Insufficient testing across edge cases and failure modes

Assuming agents will handle ambiguous situations gracefully

What you get from it

Understanding AI agents gives you:

A clearer framework for deciding when agentic AI is appropriate

Better ability to brief and evaluate agentic features

A foundation for designing safe, well-bounded autonomous workflows

More informed conversations with engineers about agent architecture and risk

Key takeaway

AI agents are not just smarter chatbots. They introduce a fundamentally different design challenge that requires thinking carefully about what the agent can do, what it cannot, and what happens when things go wrong.

FAQ

Common questions

A few practical answers to the questions that usually come up around this method.

What is an AI agent?

An AI agent is a glossarySystemA system is a collection of interconnected components that work together to achieve a specific function or outcome.Open glossary term that uses a language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term to plan and execute a sequence of actions to complete a goal. Rather than answering a single question, an agent can take that goal, break it into steps, use tools, and act on the results — iterating until the task is done.

How is an AI agent different from a chatbot?

A chatbot responds to glossaryPromptA prompt is the input or instruction given to an AI system to guide its output or response.Open glossary term with text. An agent acts. It can take real actions — searching the web, running code, calling APIs, writing files — as part of completing a task. The difference is between answering a question and completing a job.

What kind of tools can an AI agent use?

It depends on what tools have been connected to it. Common examples include web glossarySearchSearch is the functionality that allows users to find content or information by entering queries. It relies on indexing, metadata, and relevance algorithms to return useful results.Open glossary term, code execution, file reading and writing, API calls to external glossaryServiceA service is a component or function that performs a specific task within a system.Open glossary term, calendar and email access, and database queries. The agent's glossaryCapabilityCapability refers to an organisation’s ability to perform a specific function or deliver a particular outcome.Open glossary term are defined by the tools it has been given access to.

Are AI agents safe?

They can be, when designed well. The key risks are that agents can take actions that are difficult to reverse, that errors can compound across steps, and that poorly constrained agents may take actions outside their intended scope. Good agent design involves clear task boundaries, appropriate human oversight, and extensive testing.

Do I need technical knowledge to work on AI agent products?

Not deeply technical knowledge, but you need to understand the principles. Knowing what tools the agent has access to, where human oversight is needed, how failure modes are handled, and what the intended scope of the agent is are all design and product decisions — not just engineering ones.

Quick take

AI agents go beyond answering questions — they take actions, use tools, and complete tasks on your behalf, which changes what is possible and what can go wrong.

Related Services

Artificial Intelligence