They talk about IT Operations, observability, and what changes when you turn AI loose on it. Observability means monitoring, logging, and analysing what happens in an IT environment, so you can catch problems before they escalate. That doesn't sound new. What is new: one AI agent that combines the knowledge of six experts and delivers an answer in ten minutes that would normally take days.
In this article:
- IT Ops and AI: what changes in practice?
- How did a performance problem at a client lead to an AI breakthrough?
- You built a CMDB chatbot. Why?
- The technology is there. So why do clients still hesitate?
- What will this look like in twelve months?
IT Ops and AI: what changes in practice?
Jordy: IT Operations, IT Ops for short, is about how you run and manage a complex IT environment. That complexity only keeps growing. More cloud, more layers, more specialist knowledge required. And ideally, you want to make it predictable.
That's the core of it. The advantage within IT Ops is that we have access to a lot of good, structured data. Monitoring tooling has been supplying that for years. AI Ops, applying AI to that operational data, is not a new term. But with large language models you can now take it much further.
Cees: Elastic, the observability and search platform we work with, is getting better and better at understanding the intent behind a query. You ask a question and Elastic pulls in context through vector search. What used to take three, four, five engineers each working in their own domain, Elastic now does in one go. Consistently, the same way, every time.
How did a performance problem at a client lead to an AI breakthrough?
Jordy: We were working on a complex web application with multiple layers. We had already collected server metrics and started setting up an AI agent in Elastic as a performance expert for that platform. That agent quickly told us: I'm missing the webserver access logs. It turned out the integration had the wrong path configured.
The logs were on a different disk. So the agent immediately helped us find gaps in the data. That continued iteratively, until the agent held the combined knowledge of multiple experts: infrastructure, webserver, SQL, and domain-specific application knowledge.
Normally that knowledge is spread across four, five, six consultants or teams. Now you could ask one question. The agent gave a solid answer within ten minutes that would normally take days. We went from a few hundred thousand log records per hour to one and a half million. As a human, you can no longer get through that.
Jordy: A concrete example is Entra ID, Microsoft's identity solution. Those logs are easy to forward to Elastic. We built a sign-in log agent there that simply looks at sign-in issues. Give it a username and it comes back with: this user has had sign-in problems on these applications today, these are the error messages, it's most likely caused by this setting. That's a use case every IT department recognises. And once you've done that, it becomes much easier to expand into more complex situations from there.
You built a CMDB chatbot. Why?
Jordy: A CMDB, a Configuration Management Database, is where an organisation keeps track of its IT assets: servers, applications, licences, configurations. We had already built a proposition around CMDB data and Elastic. You can ask questions about your assets: which systems run this OS, which applications sit at which locations. And you can combine that data with other sources.
Cees: What I loved: the chatbot is accessible via Semantic Search to far more people than just the system engineers. During a first presentation to a client, I immediately got the question: can I also see which licences have already expired? Am I paying for services that have been out of support for years? People no longer have to file a request with a system engineer. They look it up themselves. That gives more control and a more effective organisation. During an audit, you know what you have. No more false sense of security, but demonstrably in order.
The technology is there. So why do clients still hesitate?
Cees: In IT Ops you really run into resistance. People who have worked there for years and are afraid to hand things over. You bring an organisational change alongside a technological one. You need someone at the client who is convinced it has to work. If that person isn't there, it won't succeed.
Jordy: Many organisations have a kind of thick fog in front of them. They carry the responsibility but don't know exactly what's running. What I try to do is make choices about direction together with that organisation: these are the outcomes we want. It doesn't have to be a tightly defined project. But you do need to realise that AI in IT Ops also has consequences for how you deal with vendors and external parties. Those conversations need to happen.
Cees: You need someone to drive it, internally and at the client. Someone who follows the technology and brings the rest along. Without that person, you keep simmering without progress.
What will this look like in twelve months?
Jordy: You already see large organisations working on the full agentic story on the operations side. Smaller organisations follow quickly. But it mostly comes down to mindset. The technology is there. Deploying a new workload in a complex cloud environment can still take months today: reviewing designs, pushing through changes, aligning dependencies between teams.
Agentic AI brings that down to days. Small changes to hours. The same acceleration you already see on the software development side is going to happen on the operations side too.
Cees: Observe first. Then you automate. Once observability is in good shape, you have the data to instruct agents for remediation: agents that detect and resolve problems. Within split seconds, without an engineer stepping in. With the same interpretation, consistently, every time.
Listen to or watch the full episode of Blis Bytes on YouTube and Spotify.
FAQ
Frequently asked questions
We'd be happy to answer those right away. If your question isn't listed, please feel free to contact us.
Traditional monitoring signals that something's wrong. AIOps combines that data with AI to recognise patterns and pinpoint the cause directly, instead of just raising an alert.
No. An AI agent also works on individual logs and metrics. A CMDB does add context, such as ownership and environment, which lets the agent give sharper answers.
Anything that produces structured logs or events: webserver logs, infrastructure metrics, database logs, identity logs like Entra ID, and CMDB data. More connected sources give a more complete picture.
That depends on how it's set up. Most agents detect issues and point to the cause. Fixing things automatically, remediation, is a next step with clear boundaries.
It varies per data source. In this article the first version was up within days, then improved iteratively. A specific use case, like sign-in issues, goes faster than a broad performance agent.
