Isaac · Creative Director, Aphelion
Updated October 20, 2026 · 7 min read
A chatbot helps when it answers faster than a person would and knows when to step aside. It gets in the way when it stands between the customer and a human who could have solved the problem in one sentence. The difference is design, not technology.
Scripted bot vs. AI agent
A scripted bot presents options and answers what is in its tree. Predictable, cheap, and useless the moment someone phrases a question outside the script. An AI agent reads natural language, holds context across the conversation, and can handle variation. They solve different problems and cost very different amounts.
When the scripted one wins
When the questions genuinely repeat: hours, location, whether you deliver, a base price. A menu answers those instantly and never invents anything. Paying for intelligence to answer what time you open is spending money to solve a problem you did not have.
When AI earns its cost
When the conversation is the qualification. A customer describing a project, a patient describing a symptom, a buyer weighing options: each one arrives differently, and the value comes from understanding them well enough to route or quote. That is not something a decision tree does convincingly.
The setup mistake that makes customers hate them
Trapping people. If there is no visible way to reach a human, every limitation of the bot becomes a wall. A good implementation offers the handoff early and takes it without argument. The bot should be the fast path, never the only path.
It must be allowed to say it doesn't know
An agent that improvises an answer about pricing or availability creates a problem your team has to clean up later, in front of a customer who now distrusts you. The instruction to say "I don't have that, let me connect you" is not a limitation to hide, it is the feature that makes the thing safe to deploy.
Test it as an unhappy customer
Before launch, ask it the awkward questions: the price you would rather not quote, the thing you do not offer, the complaint. How it behaves in those three cases matters more than how it handles the happy path, because those are the moments that generate reviews. You can try three agents built this way on our AI Agent service page.
Frequently asked questions
It annoys them when it traps them. If it answers quickly what it does know and offers a handoff without fighting, most people prefer it to waiting. The frustration almost always comes from not being able to get out.
A scripted bot is cheaper to build and has no per-conversation cost. An AI agent costs more to set up and consumes per message, because there's a model processing each reply.
It can, if it isn't instructed otherwise. That's why a serious implementation includes an explicit rule not to invent and to hand off when the information isn't in its context. That's something to require, not assume.
It isn't training in the technical sense, it's giving it context: services, pricing, policies and how you talk to customers. Gathering that is usually the most time-consuming part of the project, more than the technical side.
Want this handled instead of researched?
Find out in a few minutes what to fix first in your business, or tell us what you're working with and we reply within 24 hours.
Marketing agency in Tijuana
