Structural work with AI agents: this is how you really improve your processes

A report full of recommendations on AI is a good start, but it's not half the battle. If you only know in which process time is leaking away or in which process step there is delay, nothing will improve in the workplace. You only make a difference when you take the crucial step between insight and improvement. How do you do that? By building the right AI agents.

As we described in our previous blog post, it is important that you think in terms of processes (and not separate use cases). But how exactly do you go about it?

Example: submitting IT tickets, that's not so easy

Suppose you are at work and your laptop suddenly breaks down. In such a case, you set a process in motion. You open a ticket form and look for the category that your problem falls under. Often you run into something at the first step: your problem is not listed. That's why you select something on the gamble. Or you chat with a colleague from IT, who refers you back to that restrictive ticket form.

In the end, you submit the report with some reluctance. With an urgency checkmark and a note. You will receive an email with ticket number... and then you have to wait. After a few days you have not heard anything. And you have no idea who is handling your ticket.

After a week you go after it yourself. What turns out? The person who has to handle your ticket did not understand the problem and left it due to crowds. You then arrange for the right person to pick it up. And only then will the solution to your problem finally come.

One ticket that ends up in the wrong place is a relatively minor inconvenience at the organizational level. But what if this happens regularly? Then you are dealing with a process that is rattling. And that is therefore extremely suitable for improving with AI agents.

Which AI agents do you need?

In this case, three AI agents will go a long way. Everyone gets their own task. And together they cover the entire route from report to resolution.

1. The AI agent that maps out your problem

The first agent behaves like a chatbot: you tell in your own words what the problem is without wringing your story into tight, restrictive fields.

The AI agent then asks you questions that matter in your situation. Where necessary, the agent also retrieves information from existing systems. This way there is immediately a complete report on the table. Of course, that saves time up front. But the real gain is further down the process: every step goes better when you work with the right information from the start.
 

2. The AI agent that takes your report to the right place

What is going on? How urgent is the problem? What knowledge is needed for a solution? The second officer assesses these kinds of cases and then ensures that your report immediately ends up in the right place (and does not have to go through three colleagues first).

Does this mean that AI will make all decisions from now on? Not that. Predictable reports are largely automatic, while doubtful cases, risks and exceptions end up with an employee. People remain responsible for the considerations that actually matter and thus have more time for them.
 

3. The AI agent that makes sure it goes right the first time

If a ticket is resolved quickly, this does not say anything about the quality of the solution. That's why you need a third agent who steers on first time right. This officer ensures that the practitioner receives background information about the suspected cause in advance. This way, the person gets to work in a targeted way.

In addition, before closing a ticket, the agent will check whether the solution matches your original report.
 

Three agents, one point of contact: the orchestrator

Which officer is responsible for what? You don't want to delve into that. That is why all communication goes through one central assistant: an orchestrator.

When you make a report or request the status of your ticket (for example in a familiar environment such as Teams), you immediately get a realistic estimate of the lead time. Are you unable to work at all at the moment? Then the orchestrator will indicate that your report will be given priority. Is the problem annoying, but not blocking? Then you may see that the lead time is a bit longer. Either way, the orchestrator will keep you informed, so you won't be in the dark or have to call after it yourself.
 

The sky is the 'AI limit': there is no shortage of possibilities

Handling IT tickets is just one example of a process where AI agents can make a huge improvement.

The possibilities are endless: wherever you collect information, distribute work and check results, you can structurally work with AI agents. You don't have to put everything down at once. You start at the biggest bottleneck to lay a foundation that agents land on. You build from there. For each process step, you choose what is needed from the Microsoft AI ecosystem: sometimes a simple agent is sufficient, sometimes you need additional data or links.

How do you determine each next step? By continuing to measure. For example, do you see that an intake procedure has improved considerably, but that the turnaround time is lagging behind? Then you know exactly which pain point you need to tackle now.

And so AI does not remain a collection of separate experiments.