The first useful rung — a conversational interface that fields questions reliably — and exactly where it stops.
The Felesh team3 min
You deploy a chatbot to handle routine inquiries. Product details, policy explanations, operating hours — it resolves them reliably. Your team regains operational capacity. That is real value.
The chatbot matches each routine question to a prepared answer and sends text back; a question with no prepared answer gets nothing to match.
Then a customer asks to reschedule an appointment, update an order, or generate a custom quote. The chatbot politely explains its limitations. You hit the ceiling: the system talks, but it cannot act.
Asked to move a booking, the chatbot can only reply with text: it has no way to write to the booking system, so the booking stays as it was.
We are clear about this boundary. For a significant share of businesses, a well-built chatbot is the correct architecture. If your operation is answer-heavy and action-light — where customer needs are strictly informational — a chatbot serves that purpose. We do not advocate for agentic systems where a static conversational layer suffices.
A chatbot is enough while customers only need information, and an agent is needed once a record in the system has to change.
The shift occurs when conversation must yield to execution. It is time to move past chatbots when end-users need the system to act on their behalf — to book, update, modify, or initiate. When the interaction must produce a verifiable state change in a backend system rather than a text reply, you need an agent.
Given the same request, an agent writes the new slot into the booking system and leaves a log line anyone can check.
The material below details where the chatbot boundary lies and the architecture required to cross it:
Why we do not build chatbots — The paradigms we evaluated and rejected, and why the chatbot ceiling is a structural limitation, not a missing feature.
