Skip to content

Glossary

What are AI agents?

AI agents are software systems that use a language model to decide their next step and act through tools, such as search, a CRM or email, until a task is done.

Written and reviewed by Solorai

In short

An AI agent receives a goal, chooses an action, uses a tool, reads the result and repeats until the task is finished or a person needs to decide. Anthropic draws the line this way: in a workflow, models and tools follow paths written in code; in an agent, the model directs its own process and tool use.

In companies, agents take on multi-step work such as preparing an account brief, triaging tickets or matching invoices, with a person approving the steps where a mistake would be costly.

How an AI agent works

Every agent runs the same loop. It starts from a goal written in plain language. The model decides the next step and, when it needs information or has to act, calls a tool: a search, a database query, a CRM update, an email draft. The result comes back to the model, which decides again. The loop ends when the goal is met, when a limit is reached or when the agent hands over to a person.

Four things decide how well an agent works:

  • The model, which plans, reads results and decides.
  • The tools, each with a clear description, often provided through MCP servers.
  • The context: instructions, examples and the knowledge the agent can look up.
  • The guardrails: limits on steps and spend, approvals for sensitive actions, and a log of everything it did.

AI agents, chatbots and automation

The three are often confused. The difference lies in who decides the steps and whether the system acts in other software.

ChatbotWorkflow automationAI agent
Who decides the stepsNo steps: it answers in the conversationRules written in advanceThe model, while it works
Acts in other systemsRarelyYes, along fixed pathsYes, through the tools it chooses
Handles exceptionsIt can explain them, not resolve themIt stops on cases the rules did not foreseeIt adapts, within its limits
Fits bestQuestions and answersHigh-volume, predictable processesVaried tasks with several steps

Good systems often combine them: a workflow handles the predictable part and calls an agent only where judgement is needed.

Types of AI agents

Textbooks classify agents by how they choose actions, from simple reflex agents to goal-based, utility-based and learning agents. In business, a more useful split is by the job they do:

  • Task agents run one workflow end to end, such as triaging support tickets or reconciling invoices.
  • Assistant agents work for one person inside their tools, like Claude Cowork on a desktop, preparing documents, analysis and follow-ups.
  • Coding agents, such as Claude Code, read a codebase, make changes and run the tests.
  • Voice agents answer and make phone calls, take bookings and hand over to a person.
  • Multi-agent systems split a goal across several agents coordinated by an orchestrator. See agentic AI.

Where companies use AI agents

  • Customer service: answer, classify and route requests, and hand over to a person with a summary.
  • Sales: research accounts before a meeting and update the CRM after it.
  • Finance: match invoices to orders and draft follow-ups on exceptions.
  • Legal: a first review of contracts against the company’s playbook, for a lawyer to check.
  • IT and HR: triage tickets and answer policy questions from the company’s own documents.

The common thread is work with several steps, information spread across systems and a clear definition of done.

How to deploy AI agents safely

Agents act, so the controls matter as much as the model. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, because of rising costs, unclear business value or inadequate risk controls. Most of that can be avoided at the design stage.

Under the EU AI Act, an agent that talks to people must make clear that they are dealing with AI, unless it is obvious, since 2 August 2026. Agents used for high-risk purposes listed in Annex III, such as recruitment or credit scoring, carry high-risk obligations from 2 December 2027. See the EU AI Act checklist.

What the research says

  • Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of 2026, up from less than 5% in 2025.
  • In McKinsey’s State of AI 2026 survey of 1,719 respondents, about two in ten said their organisation had reached the scaling phase with AI agents. Among companies with more than $1 billion in revenue, the share was 40%.

Sources

FAQ

Questions about AI agents

What is an AI agent in simple terms?

Software that is given a goal and works towards it on its own: it decides the next step, uses tools such as search, email or a CRM, checks the result and continues until the task is done or a person needs to decide.

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

A chatbot answers questions in a conversation. An AI agent acts: it uses tools to fetch information and make changes in other systems, across several steps, towards a goal.

Are AI agents the same as agentic AI?

Agentic AI is the broader approach of building AI systems that pursue goals with some autonomy. An AI agent is the unit that does the work, and agentic systems often combine several agents.

Do AI agents fall under the EU AI Act?

Yes. Agents that interact with people must make clear that they are AI, since 2 August 2026, and agents used for high-risk purposes listed in Annex III, such as recruitment or credit scoring, carry high-risk obligations from 2 December 2027.

How long does an AI agent pilot take?

At Solorai, an AI agent pilot takes 6 to 8 weeks: one agent on one workflow, taken into production.

Related reading

Put an agent to work. Start with a 30-minute briefing.

Tell us which workflow you have in mind. We will reply with a clear next step.

Book a briefingEmail us