Three different solutions

“Chatbot” and “agent” are often used as synonyms, and they aren’t. A conversational interface is just a way of communicating with a system. An AI agent decides which steps to take and carries them out with the tools it is authorised to use. An automation follows a fixed path: it receives a request, validates its data and records it in the right system. And sometimes what you need is an application, with screens where the team reviews and decides.

These solutions can be combined. The first step is defining what the system must be able to do, because consulting a procedure is not the same as changing a record in a management system. The product name guarantees neither.

Explore the idea

Choose by the task, not the label

Change the type of work and see which approach fits.

  1. 01Input type
  2. 02Decision needed
  3. 03Bounded action
Stable rules

Deterministic workflow

When a validated request arrives, it is recorded and assigned to the right team. Known conditions are expressed as rules.

An n8n orchestration can include AI in the step that needs it.

Variable text

AI interpretation within the workflow

A free-text message or document needs interpreting. AI proposes a category and extracts the data; the workflow validates them before continuing.

Provide a path for ambiguous or unknown categories.

Multiple possible steps

Agent with limited tools

Investigating an incident may require different queries. The agent chooses among permitted tools within a step limit.

Separate queries from actions requiring approval.

These are general criteria. Your case may have details that change the answer.

If the rules are clear, an automation

When the process follows known steps between existing systems, I would start with an automation on an orchestration platform such as n8n or Make. If the rules or integrations are unusual, with custom development.

Recording a request a person has already validated doesn’t require artificial intelligence. It requires defining which fields are updated, how to avoid duplicates and how to verify the record was saved correctly.

Where an AI agent adds value

An agent adds value when the input is free text or variable documents and interpretation is needed before acting. It can classify a request, extract its data and then pass it to a conventional rule-based flow. You don’t need to turn the whole process into an agent to benefit from that.

The more autonomy it has, the more permissions matter. A client’s message is information to interpret, never an instruction that can alter the system’s rules. For actions that commit the company, such as sending a proposal, the agent prepares and a person confirms. You also need to define what it does when data is missing or the operation has already been carried out.

If several people decide, an application

When the team needs to review pending items, correct data, assign owners and consult history, a custom application usually works better than a chain of integrations.

To compare options I always use the same cases: a correct request, an incomplete one, a service that doesn’t respond and two people editing the same record. I check the result in the target system, not just the agent’s reply, and keep those tests to rerun when the model or instructions change. Then I weigh maintenance and licence costs, and whether another professional could continue the work.

If you have a process in mind and want to decide the approach, here is how I work, or book a meeting and we’ll look at your case.