Chatbot, assistant and agent: what’s the difference?
| Chatbot | AI assistant | AI agent | |
|---|---|---|---|
| What it does | Answers from set scripts | Understands questions, answers from documents | Plans and carries out the steps to reach a goal |
| Access to systems | None or limited | Usually read-only | Reading plus defined write actions |
| Example | “What are your opening hours?” | “What does the contract say about returns?” | “Read this enquiry, draft the quote and send it for approval.” |
How does an AI agent work?
- Perceive: reads an input such as an email, form, message or system event.
- Plan: decides which steps are needed to reach the goal.
- Use tools: calls defined tools such as a CRM search, a stock lookup or a calendar booking.
- Check: tests the result against the rules, and asks a person when it is unsure or an action is outside its permissions.
- Report: records what it did and why.
One open standard that is spreading for connecting agents to tools and data is the Model Context Protocol (MCP). With MCP, the same tool can be used with different models.
How companies use AI agents
- Sales: reading a price enquiry, drafting the quote and sending it for approval.
- Customer service: finding an order’s status in the system and replying; passing anything it can’t resolve to the team.
- Operations: comparing delivery dates in supplier emails with order records and reporting delays.
- Finance: matching incoming invoices with orders and flagging differences.
- Website: a site assistant that answers visitors’ questions from company information and points enquiries to the contact form.
When do you need an agent, and when simple automation?
For work whose steps are the same every time (say, creating a CRM record from a form), rule-based automation is enough and cheaper. An agent earns its place when the input is free text, the steps change with the situation and judgement is needed. Most projects use both: the agent interprets, and a rule-based flow carries out the action.
Risks and how to control them
- Permission limits: the agent should use only defined tools, within a defined scope.
- Human approval: money, prices, contracts and anything sent to customers should go for approval.
- Logging: every step should be traceable.
- Cost limits: set a daily or monthly cap on model usage.
- Wrong output: models can make mistakes; check critical facts against the source.
Frequently asked questions
Will AI agents replace employees?
Agents take on repetitive preparation work; decisions, approvals and customer relationships stay with people. Set up well, a team gets more done in the same time.
Can we run an agent on our own servers?
Yes. With open-source models, agents can run on the company’s own servers, and in that case the data never leaves the company.
What is MCP?
The Model Context Protocol (MCP) is an open protocol that gives AI models a standard way to connect to external tools and data sources.
Sources
This guide was prepared by the dipro.uk team and last updated on 16 November 2026. Questions: [email protected].
