“We need ten times fewer lines of infrastructure code than two years ago” Laurent Sorber, Superlinear CTO

Almost everyone agrees: AIOps has a bright future ahead. But what about AIOps today?

Who uses AIOps tools and is already reaping the benefits?

Professional IT magazine SAI interviewed Superlinear CTO Laurent Sorber, AIOps pioneer in Belgium. Let’s dive into the concrete uses of AIOps.

Interviewer (I): Superlinear is an AI agency. What does that mean exactly?

Laurent Sorber (LS): At Superlinear we build AI solutions across industries and AI domains. For example, optimizing the production planning of Atlas Copco’s air compressors, or accelerating vaccine development for pharma giant GSK by automating the counting of bacterial colonies in Petri dishes. In every project we do, we aim to create a measurable difference in value for our clients.

I: How does AIOps fit into your philosophy that AI should make a difference?

LS: As we see it, AIOps is a next step in the automation of the software development life cycle. With DevOps, the industry already took the first steps in that automation. AIOps represents the same principle but specialized further for AI applications. AIOps makes setting up the core tasks of any AI solution a lot easier: just think of data pipelines, training, and inference. For us, this is where the initial value of AIOps lies, but by extension, there is also the possibility to apply AI to IT operations itself.

I: Can you give a concrete example of a project where you have applied AIOps?

LS: At VDAB, we helped build AI into their job-seeking platform, Jobnet. Jobnet enables VDAB to make recommendations to employers and employees on the basis of their predicted mutual interest, rather than purely on the basis of keyword matching. For this application, we built an AIOps setup to automate deployment, training and inference.

This gives an enormous boost to the continuous integration of new components, with automatic checking for code errors and provisioning of the necessary infrastructure and processes.

I: There are still relatively few organizations that apply AIOps in Belgium. Is this due to the limited maturity of the technology?

LS: In the past two years, we have seen a marked evolution in the availability of MLOps tooling. Two years ago we still had to develop AIOps with DevOps tools; today there are ready-made tools with which you can get started. For AWS we have Sagemaker, for Azure there is now Azure ML in combination with Azure DevOps. These tools make life as an ML Engineer a lot easier.

Take for example the Jobnet project at VDAB: there, we had to set up the AIOps completely by hand for the entire network topology and compute infrastructure. In total, we wrote almost 3,000 lines of infrastructure code. In a recent analysis, we estimated that the same functionality would only require about 300 lines today, thanks to the tremendous advances in both DevOps and AIOps tools. That’s the difference between one day and a few weeks worth of work!

I: If AIOps can automate the entire DevOps process, do you still need DevOps experts?

LS: DevOps experts needn’t worry. It’s true that there is much less manual work, but even if you further automate DevOps with new technology, it still helps to understand how it all works under the hood. We will have fewer DevOps engineers, but their expertise will still be very valuable.

I: In the long term, everyone predicts a bright future for AIOps. What about the near future?

LS: It looks rosy there too. More and more customers are asking for AIOps as a condition for starting a project because they know that it can drastically reduce the total cost of the project. 

We are happy to comply because it’s a strong differentiator for Superlinear for the time being.

This article first appeared in the professional IT magazine SAI Update in September 2020 https://sai.be/tijdschrift.