QAIYU Agents

An agent is not a chatbot with a longer memory. A chatbot answers; an agent acts. It reads an incoming request, checks it against your rules, does something in one of your systems, and reports back what it did.

That difference decides where an agent is worth building. Work that repeats in the same shape every week, with rules you can write down, is where it pays off. Work that needs judgement about a person, a price or a risk is where it does not.

AI agents for business, not for demos

An agent that impresses in a demo and an agent that survives a quarter in your operation are different builds. The first needs a happy path; the second needs the exceptions written down, a boundary that holds, and someone who notices when it drifts.

What that means concretely per business function is on the solutions pages, and how an agent is put together is under platform.

Where does an agent take over, and where does it stop?

The boundary is the part that takes the longest to agree on and the part that matters most. An agent that may do anything creates work instead of removing it, because everything it touches has to be checked afterwards.

We write the boundary down before we build: which actions it performs on its own, which ones it prepares for a human, and which ones it never touches. That document is the deliverable, not a side note.

What we need from you

The process as it runs today, including the exceptions. Most of the value sits in the exceptions: they are the reason a process cannot simply be automated with a rule in your existing software.

What an agent costs to keep correct

The build is the cheap part. What follows is not, and it is the number most proposals leave out.

Processes drift. A supplier changes an invoice format, a rule changes, an exception that used to be rare becomes normal. An agent built once and left alone does not fail loudly when that happens: it keeps running and starts being wrong, and the gap between those two states can last months before anyone notices.

That is why this is priced as a subscription rather than as a project. Not because it is a better business model, but because an agent without maintenance is a liability with a good first quarter. Any calculation that compares a one-off build against an ongoing salary is comparing the wrong two things.

Why one agent per process

The tempting design is one agent that handles everything, because it sounds simpler and demos beautifully. It is the wrong shape.

An agent covering four processes has a boundary that is the union of four sets of rules, which nobody can hold in their head or check. When it makes a mistake you cannot tell which part of its remit was at fault, and every change to one process risks the other three. One agent per process keeps the boundary small enough to be read in one sitting, and small enough to be changed without a regression somewhere else.

It also means you can stop one without stopping the rest, which matters more than it sounds when something goes wrong on a Friday.

What it connects to

Agents work against the systems you already run. Replacing your stack to enable automation is a larger project than the automation itself and is almost never necessary; the systems you have contain the data and the history, and moving that is where projects die.

What matters is not which tools you use but whether they can be reached and whether their data is consistent enough to act on. That second question is the one that usually needs work first, and it comes out of the mapping stage rather than out of a sales conversation.

The log is not a technical detail

Every action an agent takes is recorded: what triggered it, what it did, what it decided not to do. That record exists for three separate reasons, and only one of them is debugging.

The second is accountability. When a customer asks why something happened, “the system did it” is not an answer anyone accepts, and it is increasingly not an answer regulators accept either. The third is improvement: the log is where you see which exceptions keep recurring, which is what tells you whether the boundary needs widening or the process needs fixing.

Before you commission one

The test for whether a process is worth handing to an agent is set out in where AI agents actually save time, and it is worth applying before asking for a quote. Every engagement runs through the same four stages, described under how a project runs; the first two produce documents rather than software.

Scope is set by how many processes move across rather than by which features you get. Book a demo to see an agent running against a process like yours before anything is committed.