Originally reported by ZDNet (June 2018). This article summarises and expands on CA Technologies’ AI-powered DevOps strategy as described by DevOps General Manager Ashok Reddy at the CA AgileOps Summit.
Background: CA Technologies and the Push for Reinvention
Founded in 1976, CA Technologies built its reputation as one of the largest and most traditional enterprise software companies in the world. Its product line traces the arc of IT itself: mainframes, client-server architecture, multi-tier applications, cloud migration, and now data-driven operations. While the technology changed dramatically over four decades, CA’s stated mission—eliminating the barriers between ideas and business outcomes—remained largely constant.
By 2018, however, the company was facing a strategic inflection point. Ashok Reddy, CA’s DevOps General Manager, used the CA AgileOps Summit keynote to lay out an ambitious vision: applying machine learning and AI across the entire DevOps lifecycle, a concept he called continuous AIOps.
The SaaS Differentiation Problem
One of the central questions Reddy posed is deceptively simple: if every enterprise is buying the same SaaS products, where does competitive differentiation come from?
His answer points to a fundamental shift in how software itself must be designed. Traditional enterprise software was deterministic—developers defined rules, those rules were encoded and deployed, and the application behaved predictably within those boundaries. In a world where every competitor runs the same SaaS platform, that model offers no edge.
Data-driven approaches change that equation. Rather than following hard-coded rules, software can adapt its behaviour based on what it learns from its operating environment. As Reddy put it:
“Software used to be deterministic — you developed it, deployed it, focused on quality, and made sure it satisfies user requirements, and that was it. Now, with data-driven approaches, software needs to incorporate what can be learned from the environment.”
Ashok Reddy, DevOps General Manager, CA Technologies
This shift from deterministic to non-deterministic behaviour is significant—and not without risk. Non-deterministic systems are harder to reason about, audit, and secure. Reddy acknowledged these concerns directly:
“Non-deterministic means you have to ensure that data used to inform the software is the right data. It must not be biased, you have to look at issues like security and privacy. And what kind of actions will the software take? Will the actions be fully automated, or will there be human involvement?”
Ashok Reddy, CA Technologies
Applying AI to DevOps Itself
The more operationally significant part of Reddy’s vision was not just making end-user applications smarter—it was applying the same data-driven, ML-powered approach to the software used to build software: the DevOps toolchain itself.
In a continuous integration and delivery pipeline, enormous amounts of signal are generated but rarely fully exploited: which test cases pass or fail, which deployments succeed or need rollback, which code paths are adequately covered. Reddy argued this data could be used to make the pipeline itself adaptive:
“Do tests reach all paths? Which test cases fail, and which ones pass? Which deployments work, and which ones need to be rolled back? All these are things we can learn, and based on this, we can change our DevOps approach. Companies can use data to make better decisions, but software itself can also learn what to deploy and when.
With AI, we can move to risk-based testing. We can apply lookup patterns: If something changes in the code, why not look at the test cases affected by that to identify the potential risk for the business?”
Ashok Reddy, CA Technologies
Risk-Based Testing: A Practical Example
Risk-based testing is one of the more concrete applications of this vision. Rather than running an entire test suite on every code change—a costly and time-consuming process—an ML model can analyse which test cases are most likely to be affected by a specific code change, and prioritise those. This reduces cycle time while concentrating coverage where the risk is highest.
This type of test impact analysis was not a new concept even in 2018, but framing it as part of a broader AI-driven DevOps lifecycle—covering testing, deployment decisions, rollback triggers, and operational monitoring—represented a more ambitious integration of the idea.
Self-Driving Operations: The Automation Progression
Reddy used the metaphor of self-driving cars to describe where he believed DevOps automation was heading. Just as vehicle autonomy spans a spectrum from manual driving to full self-driving, operational automation spans a spectrum from fully manual to fully autonomous. The near-term goal, in his framing, was not to remove humans from the loop entirely, but to free operations teams from repetitive, low-value tasks so they could focus on higher-order decisions:
“I talk about self-driving operations. The reality is that operations needs to focus on more added-value things rather than getting lost in the day-to-day chores… We want to give them a way to predict things before they happen and to automate, so they can focus on the big picture.”
Ashok Reddy, CA Technologies
This framing aligns with what has since become a widely used concept in the industry: AIOps (Artificial Intelligence for IT Operations), which applies ML to correlate events, detect anomalies, and automate responses across complex operational environments.
How CA Built This Capability: Acquisitions and Portfolio Evolution
A natural question is how a company with CA’s legacy—built on mainframes and on-premise enterprise software—developed the capability to execute on this kind of strategy. A significant part of the answer lies in its acquisition history. In the years preceding this announcement, CA acquired companies operating in:
- API management and monitoring
- Application release automation
- Test data management
- Automated test design and optimisation
- Business process automation
Notable assets in its portfolio at the time included BlazeMeter (performance and load testing) and Veracode (application security testing), both of which were SaaS-native products.
Reddy also described a SaaS-first transition for CA’s broader product portfolio, with a stated goal of making at least 50% of offerings subscription-based, and moving all products to SaaS delivery across AWS, Azure, and IBM Cloud—while retaining on-premise options for customers that required them.
Telemetry played a key role in this transition: by instrumenting its own applications and collecting operational data, CA intended to feed that signal back into training models that would make testing and development tooling progressively smarter.
Industry Context: CA Was Not Alone
CA’s push into AI-assisted DevOps was part of a broader industry movement in 2018. The use of analytics to connect development and operations teams, anomaly detection in production systems, and automated tuning of infrastructure were themes appearing across multiple vendors. Around the same time, Splunk announced its own acquisition aimed at expanding into DevOps automation—signalling that the market was converging around the idea that operational intelligence should be built directly into the delivery pipeline.
The critical open question at the time—and one that remains relevant when evaluating any AIOps platform—was whether the individual capabilities acquired or developed could be integrated into a genuinely coherent workflow, or whether they would remain loosely connected point solutions.
Key Takeaways for DevOps Teams
- Deterministic vs. adaptive software: As ML is embedded deeper into applications and tooling, teams need governance frameworks to manage non-deterministic behaviour—including data quality, bias controls, and human-in-the-loop checkpoints.
- Risk-based testing is a practical near-term win: Using code change analysis to prioritise test execution reduces pipeline duration and concentrates coverage where it matters most.
- AIOps is about augmentation, not replacement: The realistic near-term value is freeing engineers from toil, not removing them from operational decisions.
- Telemetry is a prerequisite: You cannot train adaptive systems without instrumentation. Building observability into every layer of the stack is foundational to this approach.
- Evaluate integration depth, not just feature breadth: A portfolio of acquired point tools is not the same as an integrated platform. Scrutinise how data flows between components before committing to a vendor’s AIOps story.
Note: CA Technologies was acquired by Broadcom in 2018, shortly after this announcement. How CA’s AI-powered DevOps roadmap evolved under Broadcom ownership is a separate consideration for teams evaluating these products today.