Study: AI, Overcoming Data Hurdles Key to DevOps Automation Gains

DevOps automation is delivering business benefits but data challenges and the need to leverage AI remain key obstacles organizations must overcome to reach higher levels of automation maturity, according to a new report from Dynatrace.

The 2023 DevOps Automation Pulse Report discovered that on average, 56% of end-to-end DevOps processes are automated. However, just 38% of organizations admit to having a clearly defined DevOps automation strategy.

Surveyed IT firms want to automate their DevOps work, but found that security concerns (54%), difficulty operationalizing data (54%) and toolchain complexity (53%) were preventing implementation.

“Teams are entrenched in data silos, isolated pockets of automation and reactive and manually intensive operations and security efforts,” said Dynatrace CTO Bernd Greifeneder. "They urgently need a unified, AI-backed approach to DevOps automation, or it will be impossible to accelerate innovation while maintaining software quality and security."

The report interviewed 450 IT practitioners responsible for DevOps and security automation in large organizations across the U.S., Europe, the Middle East and Asia.

Data and skill issues

In its report, Dynatrace found that automating processes in DevOps is having tangible benefits for businesses.

Related:Meet Owl: The Large Language Model that Automates IT Tasks

On average, respondents saw a 61% improvement in software quality, a 58% improvement in employee satisfaction, 57% reduction in deployment failures and a 55% reduction in IT costs from automation.

However, challenges to using data and insights to drive automation decisions remain. Dynatrace respondents said the biggest hurdles for using data to drive automation include inaccessible data (51%), siloed data (43%) and the need for data to flow through many systems to be analyzed (41%).

Some 54% of respondents said they are investing in platforms to enable easier integration of tools and collaboration between teams involved in automation projects.

However, respondents said they are relying on more than seven different tools on average, showing that disparate tools and fragmented workflows are hurdles.

Another barrier to implementing DevOps automation related to skills. Some 56% of respondents said proficiency in scripting languages is one of the biggest skill shortages hindering automation.

LLMs to the rescue

One way to improve workloads could be through large language models (LLMs). Those surveyed pointed to LLMs as helping with productivity and reducing manual effort (57%) as well as enabling teams to generate code automatically (48%).

Related:Security Teams Eye AI But May Balk at High Subscription Fees

Large language models can either be fine-tuned on existing data to improve at specific tasks, or use specially designed models for domains, like Owl, the model that automates IT tasks.

Dynatrace’s report said that DevOps teams will need to combine both large language models and data maturity to “deliver precision and prediction.”

“Data-driven automation is the key to unlocking innovation and meeting customer expectations in the cloud-native era,” said Greifeneder. “Unlike traditional AI techniques that are limited in scope and applicability, platforms that combine predictive, causal and generative techniques can excel in specific capabilities to address different DevOps automation use cases.”