When we speak of AI Operations (AIOps), we’re talking about the patterns, best practices, and enabling tools used to develop, tune, test, incrementally improve, and productionize these types of AI workloads and the models themselves in a scalable way, automating where possible to achieve efficiencies and reduce human error. In this way, AIOps is itself a sub-discipline within DevOps, and mirrors many of the patterns (e.g., CI/CD), best practices (e.g., automate where possible), and enabling tools (e.g., Azure DevOps) found in “traditional” DevOps.

Just as AIOps is a sub-discipline of DevOps, so too is Machine Learning Operations (MLOps) a component of AIOps. It’s confusing, so let’s set the record straight.

It's easy enough to use the terms “Artificial Intelligence” (AI) and “Machine Learning” (ML) interchangeably, so let’s clearly define the difference.

AI is a broad and evolving field wherein the technology mimics, and in some cases surpasses, the cognitive abilities of humans. This broad category encompasses everything from “if this, then that” scenarios where the seeming “intelligence” is the product of pre-determined decision trees and patterns that humans have themselves created, all the way to “generative AI” where the technology is able to generate bespoke answers to questions, images, insights, and other responses based on its index of accumulated knowledge.

ML is a sub-discipline within the broad category of AI, wherein the machine “learns” and refines its own capabilities based on the information and feedback it encounters, and based on the tuning that ML engineers apply over time.

Implementing and maturing AIOps will, for many organizations, involve a simultaneous maturing of their DevOps capabilities to be relevant in the age of AI, as well as maturing their MLOps to support their machine learning models.