What we deliberately do not automate
It is easy to list what a product does. It is more useful, and rarer, to be clear about what it will not do on its own. Here is a short list of things Lyrion deliberately does not automate, and the reasoning behind each.
A boundary you can name is a boundary you can trust.
Irreversible actions without a person
Anything that cannot be undone stops for a human by default. Deleting data, cancelling an account, issuing a payment above a set limit. Reversibility is the property that makes automation safe, and where it is absent, the safety has to come from a person instead.
You can lower the bar deliberately, action by action, once you trust it. But the default is that the agent does not walk through a one-way door alone.
Decisions that need a policy you have not set
An agent should not invent policy in the moment. If there is no rule for how to handle a situation, the right move is to escalate, not to guess and set a precedent nobody agreed to. The agent follows the policy you have. It does not write the policy you are missing.
This is why configuration matters more than cleverness. The agent's judgement is bounded by the rules you gave it, and where the rules run out, so does its authority.
Anything the customer explicitly wants a human for
When a person asks to speak to a person, that request is honoured, quickly, with context attached. An agent that fights to keep handling a customer who has asked for a human is optimising the wrong number and making an enemy in the process.
The measure of good automation includes how gracefully it steps aside. Insisting on handling someone who wants out is not efficiency. It is the thing people hate most about automated support.
Its own oversight
The agent does not decide what it is allowed to do, does not grant itself new access, and does not approve its own sensitive actions. The scope, the permissions and the approvals are set by people and enforced by the system, not negotiable by the thing being scoped.
This one is non-negotiable for a reason. An automation that can widen its own authority is not an automation you control. Keeping oversight outside the agent is what keeps the whole thing yours.
Where it leaves the team
The people who used to spend the morning triaging spend it on the handful of things that genuinely needed a person. The repetitive work is handled and recorded. What is left is the harder work, which is the work most teams would rather be doing anyway.