AI agent, chatbot, or a receptionist?
The three are not competitors. They do different jobs, and most businesses end with two of them.
The scripted chatbot
Buttons and canned answers. Does: points people to pages. Fails: the moment a question isn't on the list, which is most questions. Costs: little, and earns about the same. Right for a very small site with three questions.
The AI agent
Answers from your approved facts, in the customer's language, any hour, and acts: books into the real calendar, takes an order, qualifies the enquiry and hands it over with the whole conversation. Fails: when it is allowed to guess. Ours answers only from your facts, shows the source, and says so when the answer isn't there. Costs: a two-week prototype at a flat price, then a monthly run that depends on channels and volume. Right for any business that gets the same forty questions a day or loses enquiries after hours.
The receptionist
Judgement, warmth, the ability to say "let me check" and mean it. Fails: at 11 pm, on Sunday, and on the fortieth repeat of the same question. Costs: a salary. Right for every business where a person matters at the moment of decision, which is most of them.
The combination
An agent that answers, books and qualifies at any hour, and a person who steps in the moment it matters. The agent does the repetition; the receptionist does the judgement. In practice the receptionist gets their afternoons back and the business stops losing the enquiries that arrived at night.
What to ask any vendor
Will it say something that isn't true? Can you show me where each answer came from? What happens when it doesn't know? Can my team change the facts without you? If the answers are vague, keep looking.
Tell us what you need built.
An engineer reads it and replies within minutes, in your language. Every project starts with a two-week prototype at a flat price.
Related
Why every project starts with a two-week prototype
Large contracts signed blind fail quietly. A two-week prototype at a flat price turns the decision to build into a decision made on something running.
How we make a website AI assistants can quote
Buyers now ask ChatGPT, Gemini and Perplexity who to call. What a page needs so an assistant describes you correctly and cites you: plain facts, structured data, one answer per question, and a facts page.
WebGPU or WebGL: what we ship in 2026 and why
WebGPU is in every major browser, WebGL still reaches everyone. How eORB decides per project, with a fallback so no visitor sees a blank box.