Business ·
A chatbot answers. An assistant helps. An agent acts within boundaries
Chatbots, AI assistants and AI agents often appear in the same proposal despite carrying different levels of responsibility. One conducts a conversation, another prepares work for a person, and the third can perform a sequence of actions in operational systems. The choice begins with permission to act rather than technology.
A chatbot follows a known conversation route
A chatbot receives a message, identifies a topic, asks clarifying questions and either answers or routes the enquiry to a person. It suits common questions, initial qualification, booking and service navigation.
Even when powered by a language model, it remains a communication channel. If it does not change CRM records, create documents or decide on behalf of the company, its risk is mainly in response quality and routing.
An AI assistant prepares work for a person
An assistant can retrieve knowledge, summarise correspondence, propose a reply, draft a CRM record or compare documents. A person sees the result, checks it and decides what to send or save.
This mode fits tasks that benefit from context and speed but still require professional judgement. It reduces preparation without taking the final decision away from the employee.
An AI agent performs an action
An agent receives a goal, uses tools and changes system state: it creates a task, updates CRM, requests a document or starts an approved process. It needs permissions, an action log, limits and stopping conditions.
Autonomy does not need to be maximal. A useful agent has a narrow role, known inputs and an allowed set of operations. Irreversible actions remain with a person or require an additional confirmation.
Error cost determines the right mode
When a wrong answer is easy to correct, conversation can be automated. When the output affects a client, contract or money, an assistant with review is safer. When a process is formalised, reversible and observable, some actions may move to an agent.
Assess consequences rather than the impact of a demo: what happens after a wrong classification, stale source, unavailable integration or unusual request?
One system may combine all three modes
A customer chatbot gathers initial facts, an internal assistant proposes a decision, and an agent creates a task after approval. The boundaries between modes should remain visible in the interface and action log.
This composition is often more reliable than one supposedly universal agent. Each stage receives its own quality criterion, access rights and path back to a person.
Begin with a short, controlled segment
Choose a repeated operation, reliable sources and a measurable result such as response time, correct routing or fewer manual transfers. Test exceptions before extending permissions.
Within the Method, I define an AI role as I would define an employee’s role: task, tools, boundaries, responsibility and supervisor. Automation then becomes part of the organisation rather than an uncontrolled promise of autonomy.
AI for a specific process
If you are considering AI implementation, we can start with the process, available data, automation boundaries and a way to assess the result.