devopsy.kinsta.cloud

Tag: ai

The Era of AI-Native Software: Why Retrofitting AI Won’t Work And How DevOps Must Keep Up

Integrating artificial intelligence (AI) onto legacy software is like trying to get an old flip phone to run the latest augmented reality (AR) apps. Yet, enterprises continue to take this approach, in the process encountering compatibility issues, sluggish performance and AI behavior which deviates from expectations.

We live in an AI-centric age, where organizations have AI integrated into at least one business function. Statista reports that global AI adoption soared to 72% in 2024 from 55% in 2023. Hence, it is no secret that AI is quickly becoming a mainstream business tool instead of remaining an emerging trend. However, Boston Consulting Group’s report reveals that 74% of companies are yet to gain a tangible return on investment from AI.

So, how do we reap the obvious benefits of AI, which the 4% of the companies that have built cutting-edge AI capabilities across functions are reaping? The answer is simple — AI-native software.

Understanding AI-Native Software

Building AI-native software entails leveraging AI and machine learning (ML) as core components in applications and systems from the start, rather than considering it as an afterthought in the architecture. Such software can learn and improve its performance over time.

What sets AI-native software apart is its ability to combine the trifecta of deep research, reasoning and Agentic AI. This combination enables systems to analyze vast datasets and reason through complex scenarios to determine optimal and personalized outcomes. It then collaborates dynamically with AI agents to enhance research and decision-making.

The key difference lies in the structure. Older systems are about retaining stability and depend on fixed, predefined workflows that produce expected results. AI-native apps refine and continuously improve to adapt based on learned information. This shift in design philosophy demands an entirely new DevOps approach, one that aligns with the changing nature of AI models.

Challenging Traditional DevOps

DevOps pipelines were automated to improve the integration, testing and deployment of code in the software development lifecycle. This set of workflows meets the needs of deterministic applications where updates follow a set and predictable plan. Nevertheless, AI-native applications need more than a simple code change. They require continual retraining of models, handling of complex data evolution, ensuring real-time performance and validating outcomes across dynamic environments. Without a holistic approach, these challenges can disrupt established pipelines and lead to inconsistent results.

The complexity of AI-native applications requires seamless cross-functional collaboration. Quality engineering ensures model validation and addressing issues like bias and accuracy, while delivery excellence focuses on smooth integration, testing and deployment. Data science optimizes model performance, and security teams safeguard models from threats while ensuring compliance. Business analysts translate AI insights into actionable strategies, and product managers ensure alignment with business goals. Together, these teams work to maintain model performance, scalability and reliability throughout the lifecycle.

At Persistent, we have witnessed how organizations struggle when retrofitting AI into legacy systems — dealing with outdated infrastructure that isn’t AI-ready, fragmented data silos and the difficulty of adapting existing DevOps pipelines to handle AI models. By rethinking software architecture with AI as a foundational element, we can help them navigate these hurdles seamlessly. This approach ensures that data is clean, integrated and accessible, enabling AI to perform its best. Incorporating feedback loops, data validation and continuous model evolution into the development lifecycle allows AI systems to remain reliable and adaptive as the application evolves. This shift has empowered clients to move beyond one-off implementations, enabling smarter, more agile decision-making while keeping systems flexible, scalable and future-proof.

These complexities of AI-native applications demand an advanced DevOps strategy — one that incorporates an agile approach to AI for operations (AIOps) and ML for operations (MLOps) to ensure that applications are built and released in an optimized, dependable state at all times.

Leveraging AIOps and MLOps

AIOps introduces real-time monitoring, anomaly detection and automated issue resolution. It ensures that AI-powered applications evolve correctly. For example, if an AI chatbot begins generating biased responses due to skewed training data, AIOps steps in to flag and correct the anomaly before it disrupts customer interactions.

Meanwhile, MLOps manages a model’s life cycle, including version control, automated model retraining and performance audits. Without MLOps, AI models risk stagnation, leading to degraded performance and ethical concerns such as data drift, bias or compliance violations.

Together, AIOps and MLOps form a strategic framework that oversees the entire lifecycle and integrity of AI models, ensuring that every decision and action are accurate, unbiased and continuously improving.

Rethinking CI/CD for AI

Traditional CI/CD pipelines need to be redesigned to focus on data and model performance rather than just the code. Unlike AIOps and MLOps, which maintain the ongoing health and evolution of the models themselves, the modern CI/CD evolution centers on the deployment pipeline, which ensures that as models are updated, the process remains agile and responsive to new data.

Automating retraining and robust validation within the deployment pipeline can enable businesses to keep their models as agile as their environments. The approach goes further by shifting the focus of continuous monitoring from simplistic uptime checks to real-time evaluation of accuracy, bias and other performance factors to ensure swift, well-informed action can be taken when needed.

At Persistent, we encourage this shift — moving from code-centric to data-centric pipelines. We guide organizations to integrate model training and validation into CI/CD, implement continuous monitoring and feedback loops to eliminate biases and embed explainability and interpretability into every layer of the stack.

This evolution in CI/CD is less about software deployment and more about intelligence deployment, where AI models are refined and optimized just as frequently as the underlying application code. This evolved approach safeguards the integrity of AI models and reinforces their ability to drive impactful, reliable decisions in a world that waits for no one.

Securing AI-Native Applications

Adopting AI-native software is not just a technological change; it is a redefinition of how companies tackle software engineering, security and organizational workflows. To have value-based AI-native systems, businesses need to adopt cloud-first, containerized environments for scalable AI workloads, migrate from traditional databases to deep learning and predictive data lakes and balance regulatory compliance with ethical AI development to ensure unbiased and transparent decision-making.

Security is another essential aspect to consider in AI-native applications. The presence of adversarial attacks, model manipulation and data poisoning necessitates an integrated effort from security, DevOps and data science teams. The responsibility of making sure AI models function well while ensuring fairness, security and legal compliance is swiftly becoming an issue that companies must deal with.

At Persistent, we are committed to developing AI-native software that not only meets the highest standards of quality and performance but also addresses the unique security, compliance and ethical concerns associated with these applications. Our delivery excellence and quality engineering center of excellence (COE) play a critical role in ensuring that our AI-native software is developed with security, compliance and ethics in mind. Toward this, they conduct threat modeling & risk assessments, outline & implement secure coding standards & practices, develop & implement compliance frameworks, collaborate with regulatory bodies and finally conduct regular audits and assessments.

Paving the Road to AI-Native Systems

Software development is rapidly evolving — from integrating AI to being fundamentally shaped by it. The next wave of innovation requires companies to invest in AI architects who can code and embed intelligence deep within software.

At the leadership level, fostering an agile and adaptive mindset is essential. We focus on strategic guidance — setting ethical and business boundaries while empowering AI systems to self-optimize. Success indicators will evolve from counting detected flaws or delivered functions to measuring autonomously enhanced outcomes, such as AI-driven engagement models reducing customer churn.

The future belongs to those who lead, not just adapt. AI-native development isn’t merely an advantage — it is the foundation of the next era of technology. Companies that move decisively today won’t just optimize efficiency; they’ll create new business models, revolutionize customer experiences and set the benchmark for intelligent software. As AI evolves, the true differentiators will be those who innovate boldly, embed AI as a core capability and shape a future where software doesn’t just enable businesses — it drives entirely new possibilities.

May 22, 2025
Why You Shouldn’t Forget Workflows With Agentic AI Systems

As Matthew McConaughey runs around an airport wishing he had used an AI agent to book his flight, we could be forgiven for assuming that the technology is now a well-understood and integral part of everyday life for most people.

Of course, that’s not true. In reality, it’s quite far from the truth, even for businesses, but that is not to say that things aren’t going that way. In the future, many of our decisions will be handled by agents, or at least heavily informed by them. But in areas where we want more control, we will likely retain decision-making power, following a programmatic approach or workflow.

When booking a vacation, for example, we could consider different countries, cities and situations, the cuisine, the weather and the culture. Or we could just get agents to put together some options based on our preferences. The third option would be a mix of the two, such as choosing where you want to go and allowing agents to choose and book activities — again, based on our preferences and hobbies.

The choice will come down to how much autonomy we want to retain. And it’s the same with AI systems. In this article, I’ll explain the move from one decision-making method to another and show why both are still valid and complementary.

How Models Think

Traditional models have limited knowledge. They are trained on a finite amount of data, which puts some restrictions on the kinds of problems they can solve. For example, the first public iteration of ChatGPT could only deliver answers to prompts based on the information available up to 2022, as its underlying LLM, GPT-3.5, was trained on data up to that point. This is also why you will often hear or read about models that have been fine-tuned for specific use cases, such as HR or legal, as these use cases require additional steps and data to tailor them to the use case.

These models can be used to underpin generative AI applications within an organisation. They can therefore broadly answer queries relating to the models that have been fine-tuned, such as general HR policies or legal practices, but cannot answer specific questions employees might have that relate to their circumstances or organisational data. To unlock this ability, we can create a workflow that enables models to access and retrieve data from proprietary systems and internal corpora of data. This is what is known as a Retrieval-augmented generation (RAG) system, but one that is programmatic, i.e., a human has built in the capabilities that enable the model to access systems outside of its training data to extend its capabilities.

With tool-augmented LLMs, it is therefore possible to create advanced workflows that are limited only by the modular elements developers choose to include, meaning the control logic is determined by a human.

An agentic system uses an LLM’s ability to reason to determine the steps required to complete a task or fulfill a request. The difference between this approach and programmatic workflows boils down to the amount of autonomy we want our AI systems to have.

Structured workflows involve a linear progression to deal with problems. In many instances, this is the most appropriate approach, as it guarantees that the system won’t go beyond what it has been programmed to do. However, as the reasoning capabilities of LLMs have continued to develop, it is now possible to hand the control logic of compound AI systems over to the LLMs themselves. This is thanks to “chain-of-thought” reasoning that enables an LLM to break down problems, e.g., prompts, into logical steps, taking a little longer to consider each stage so that a more reliable and accurate output is delivered.

Orchestrating Agentic Autonomy

Orchestration frameworks, such as LangGraph, are used to build in the stages of reasoning that agents use to complete their tasks. This can include a stage in which the agent considers its initial answer and determines whether to proceed with or change its output. Orchestration frameworks also build in the tools that agents can use to complete their tasks. A simple example might be providing a specific tool that retrieves live meteorological information. By giving the agent access to this tool, it can bring in this live information for the use case that is being served, such as booking activities on vacation. No one wants to be struggling through outdoor activities when there’s a storm raging, which is why an agent who can reason on the information they get from calling a weather tool is particularly useful.

These agents are built in line with a framework known as ReAct (‘Re’ for ‘reasoning’ and ‘Act’ for ‘acting’), which ensures the agent follows the Thought-Action-Observation reasoning paradigm. These stages mean the agent will first consider the tools it has at its disposal to return a satisfactory response, such as a weather app, then take action by retrieving the information, observing the situation from the raw data and then presenting the information in a user-friendly format following a further reasoning stage.

Agents are the new kid on the block, but it doesn’t mean they’re here to usurp traditional workflows. The two work rather well in conjunction. Workflows can hand off to an agent, or team of agents, to complete tasks, just as with the use case I outlined at the beginning of this article — and it’s no different in a business context. The reverse can also be true, with agents handing off to pre-defined workflows once a course of action has been suggested. This can then lead to further feedback loops to iterate and ensure accuracy and better decision making.

Ultimately, the amount of autonomy we build into a system will always be determined by the use case, which will, in turn, be determined by the exact amount of control that is required.

May 21, 2025
JFrog and Dynatrace Extend DevOps Reach to NVIDIA Enterprise AI Factory

NVIDIA in partnership with JFrog and Dynatrace are working toward integrating DevOps tools with a platform for building artificial intelligence (AI) agents.

At the COMPUTEX conference in Taiwan this week, NVIDIA added a validated design for building AI agents to the NVIDIA Enterprise AI Factory, an application development platform based on the NVIDIA Inference Microservices (NIM) framework that leverages containers to provide access to reusable services.

Additionally, NVIDIA also added NVIDIA AI Blueprints to make it simpler for developers to build avatars, otherwise known as smart AI teammates.

At the same time, JFrog revealed that its platform for managing software artifacts will be included in the NVIDIA Enterprise AI Factory platform.

JFrog CTO Yoav Landman said the Software Supply Chain Platform developed by JFrog will now serve as a single source of record for keeping track of the provenance of the software artifacts and AI models used to build AI applications. The JFrog Software Supply Chain Platform provides the ability to track the pulling, uploading and hosting of AI models and datasets, AI containers, Docker containers, and dependencies across NVIDIA’s AI Factory platform. All told now, the JFrog artifact repository can now support more than 40 different types of software packages, including NVIDIA NIM.

Additionally, DevSecOps teams building these applications will be able to use JFrog tools to scan for vulnerabilities and other security-related issues, he added.

JFrog also revealed the core JFrog Platform can now also run natively on NVIDIA’s Grace Blackwell graphical processor units (GPUs) to improve overall performance.

Dynatrace, meanwhile, announced it is integrating its observability platform with NVIDIA Enterprise AI Factory to enable DevOps teams to troubleshoot the NVIDIA application development platform.

It’s not clear how widely adopted the NVIDIA Enterprise AI Factory has been adopted, but as more AI applications are built, the need to integrate these types of platforms into DevOps workflows is becoming more pressing. On one level, AI models are just another type of artifact being used to build an application. However, the way existing artifacts are managed using version control systems doesn’t as easily apply to AI models. Each new version of an AI model, for instance, isn’t backwards compatible with the previous instance of the model, noted Landman.

Regardless of how AI application development becomes integrated with DevOps tooling and platforms, the pace at which these applications and associated AI agents are being built will increase dramatically in the age of AI.

The challenge and the opportunity is integrating a set of application development platforms that, in many instances, are adopted by data science teams rather than traditional application developers. In effect, data scientists are emerging as a new type of end user that DevOps teams will need to support.

Undoubtedly there will be culture clashes in the months ahead as organizations look to extend their existing DevOps workflows to govern the development of AI applications. In the meantime, however, DevOps teams would be well-advised to investigate how their organizations may be building AI applications today in a way that probably isn’t going to scale as easily as it should.

May 20, 2025
GitLab Extends Scope and Reach of Core CI/CD Platform

GitLab Inc. has updated its core continuous integration/continuous delivery (CI/CD) platform to natively provide more artificial intelligence (AI) capabilities, along with built-in software artifact management capabilities and more robust security tools.

Justin Farris, vice president of product management for GitLab, said GitLab 18 makes available AI capabilities, such as code suggestion and completion, to all customers rather than requiring a separate GitLab Duo subscription.

There are still GitLab Duo capabilities that are not available in the core platform, but as DevOps continues to evolve, it is becoming more apparent which capabilities are going to be required by most customers compared to others that only a subset of GitLab customers may require, he added.

Additionally, GitLab is also moving to make GitLab Duo itself more widely accessible by no longer requiring customers to upgrade to GitLab Ultimate, the highest level of service that GitLab makes available on its software-as-a-service (SaaS) platform. Instead, organizations that have a GitLab Premium subscription can now also license GitLab Duo.

GitLab is also making it simpler to adopt best DevSecOps practices by providing built-in artifact management tools such as immutable tags and a virtual registry for Maven, open source project management software from the Apache Software Foundation (ASF).

There is also now support for structured inputs and enhanced pipeline management tools to manage artifacts and detect any changes made to them. Those capabilities promise to make it simpler to both securely parent/child pipelines and optimize pipeline execution. GitLab Query Language also now enables users to find, filter and embed content from anywhere in the GitLab platform.

GitLab has also added impact assessment tools for changes made to security policies, additional vulnerability dashboards, a reachability analysis capability for dependencies, support for passkey as defined by the FIDO Alliance and an ability to customize the detection logic it makes available in its static application security testing (SAST) tool.

Finally, GitLab 18 added more controls for complying with additional security frameworks, including the ISO 27001 framework for implementing information security management systems and benchmarks defined by the Center for Internet Security (CIS).

Collectively, these additional AI, artifact management and security capabilities ultimately serve to further reduce the total cost of DevOps by reducing the need for third-party offerings, while at the same time streamlining workflows, noted Farris.

It’s not clear how aggressively DevOps teams are streamlining workflows, but as more of them adopt platform engineering as a methodology for managing pipelines at scale, many of them are at the very least evaluating their options. At the same time, the rise of AI is also driving organizations toward platforms that make it simpler to extend the accuracy of a model by, for example, exposing it to the telemetry data they collect.

Regardless of approach, the one certain thing is that DevOps will continue to become more automated. The challenge and the opportunity now is determining which manual tasks performed today by a DevOps engineer lend themselves better to being assigned to an AI agent that has been specifically trained to complete them.

May 19, 2025
Windsurf Launches SWE-1: AI Models Built for the Entire Software Engineering Process
Windsurf has unveiled its first family of specialized models designed to transform developers’ work in a significant development for AI-assisted software engineering. The SWE-1 family, announced on May 15, 2025, represents a fundamental shift in AI assistance for developers, moving beyond mere code generation to encompass the entire software engineering workflow.

Beyond Just Writing Code

While recent years have seen remarkable improvements in AI coding capabilities, Windsurf recognized a critical limitation in current approaches: Software development involves much more than just writing code.

“Why build SWE-1? Simply put, our goal is to accelerate software development by 99%. Writing code is only a fraction of what you do. A ‘coding-capable’ model won’t cut it,” explains the Windsurf team in their announcement.

The SWE-1 family includes three distinct models:
- SWE-1: The flagship model, comparable to Claude 3.5 Sonnet in tool-call reasoning capabilities, while being more cost-efficient to run. It’s temporarily available to paid users at zero credits per prompt.
- SWE-1-lite: A mid-sized version that replaces Windsurf’s previous Cascade Base model, offering all users improved quality and unlimited use.
- SWE-1-mini: A small, ultra-fast model powering Windsurf Tab’s passive experience for all users.
Flow Awareness: The Key Innovation

What sets Windsurf’s approach apart is its concept of “flow awareness”—the ability for AI systems to understand and operate within the complete, shared timeline of development work. This insight came from the company’s popular Windsurf Editor, which enables seamless collaboration between humans and AI.

This flow awareness allows the models to understand incomplete work states and switch naturally between AI and human contributions. If a model makes an error, the human can jump in to correct it, and the model can then continue working based on those corrections, creating a truly collaborative workflow.

“It will be a while before any SWE model can truly do everything independently,” acknowledges Windsurf. “Flow awareness enables the right form of interaction during this intermediate period.”

Impressive Benchmark Performance

According to Windsurf’s evaluation data, SWE-1 performs comparably to frontier models from major AI labs and significantly outperforms mid-sized and open-weight alternatives. The company uses two primary benchmarks:
1. Conversational SWE Task Benchmark: Testing how well a model can address the following user query in the middle of an existing session with a half-finished task.
2. End-to-End SWE Task Benchmark: Evaluating a model’s ability to solve a problem independently of beginning to end.
In production experiments with real users, SWE-1 demonstrated strong performance in metrics like “Daily Lines Contributed per User” and “Cascade Contribution Rate,” reflecting both the quality of its suggestions and users’ willingness to adopt them.

A DevOps Perspective

These developments hold particular promise for DevOps professionals. The SWE-1 models’ ability to work across multiple surfaces—including the terminal, text editor, and browser—aligns perfectly with the integrated nature of modern DevOps workflows.

The models can:
- Incorporate terminal outputs and understand errors
- Seamlessly transition between text editing and debugging
- Maintain awareness of terminal commands and IDE actions
- Process user feedback and testing results
These capabilities could significantly streamline the often complex handoffs between development and operations phases that DevOps teams manage daily.

“Windsurf’s SWE-1 announcement is a clear indicator that the future of software development is rapidly becoming AI-driven, extending far beyond simple code generation,” said Mitch Ashley, VP and practice lead, DevOps and application development at Futurum. “I applauded Windsurf’s ambition to address the entire development process with AI, integrating human-AI interaction at a fundamental level. This raises the bar for all vendors in the space, pushing them to deliver more holistic, contextually aware, and truly agentic capabilities to developers.”

Building a Software Engineering Flywheel

Windsurf’s approach represents a promising flywheel effect: As users interact with their tools, the company gains valuable insights into where models need improvement, enabling it to enhance model capabilities continuously.

“We always know, at scale, exactly what our users want us to improve with our models next,” notes the Windsurf team. “That’s how we’ve rapidly built our model to the level it has achieved in today’s SWE-1 state.”

What’s Next for Windsurf

While the company is proud of its initial results, it emphasizes that SWE-1 is just the beginning. Windsurf plans to invest significantly further in its model development, with the ambitious goal of not just matching but exceeding the performance of frontier models from major research labs within the software engineering domain.

For the growing number of DevOps teams integrating AI tools into their workflows, Windsurf’s focus on the complete engineering process rather than just coding tasks represents a promising evolution that could help bridge the traditional gaps between development and operations.

As software teams continue exploring how AI can enhance their productivity without sacrificing quality or maintainability, Windsurf’s “flow-aware” approach offers an intriguing model for human-AI collaboration that respects the complex, iterative nature of modern software development.
May 16, 2025
Agentic Process Automation This Way Comes to Software Development

The shift from robotic process automation (RPA) to agentic process automation (APA), prominently on display at Automation Anywhere’s Imagine conference in Orlando this week, promises significant implications for the software development community—particularly for low-code/no-code and citizen developers. Automation Anywhere, a provider of automation platforms and framework, articulated how APA, powered by agentic AI, will redefine the development landscape by providing unprecedented flexibility, intelligence and ease-of-use in application development. More than that, they are going all in on APA.

Historically, RPA has been widely employed to automate a range of routine and repetitive tasks. However, its limitations were around RPA requiring highly structured environments and precisely defined logic paths. Small deviations or variations in data inputs often meant manual intervention, limiting its scalability and efficiency in dynamic business processes.

APA, with its integration of advanced AI capabilities, overcomes these limitations. It offers a significant advancement—automation agents capable of understanding context, learning from experiences and adapting workflows dynamically. This intelligence liberates automation from rigid script-bound processes, providing immense potential to enhance productivity, especially in software development environments dominated by low-code/no-code tools.

One of APA’s most transformative impacts will be felt by citizen developers—individuals with limited or no formal coding training who build applications primarily through visual interfaces and drag-and-drop components. Historically, citizen developers faced significant hurdles as their applications became more complex, often necessitating the involvement of professional developers to handle the logic-intensive, conditional workflows beyond their skill sets.

With APA, citizen developers have access to a new class of tools—intelligent agents capable of automatically handling complex backend logic and decision-making processes. These agents don’t just automate routine tasks; they actively assist in managing more sophisticated operations typically reserved for professional developers. As a result, citizen developers can confidently create more robust, intelligent, and dynamic applications without becoming overwhelmed by complexity.

APA’s impact on low-code/no-code platforms will also be profound. These platforms have democratized application development, empowering non-technical users to address their unique business challenges rapidly. Integrating agentic automation into these platforms amplifies their capability significantly. APA agents can proactively identify patterns, suggest optimal workflows, troubleshoot issues autonomously, and dynamically adapt processes based on evolving data inputs.

Consider a typical scenario where a citizen developer creates a customer-facing app to streamline order processing. Traditionally, even low-code platforms require significant configuration to handle variations in customer behavior or unexpected data formats. APA agents, by contrast, autonomously manage these exceptions in real-time, learning from every interaction and continuously refining the process. This means less manual tweaking, fewer errors, and dramatically improved operational efficiency.

Moreover, the Imagine conference emphasized the importance of orchestration—a crucial consideration in the APA era. Organizations increasingly utilize diverse automation tools and multiple intelligent agents from different vendors. Effective orchestration ensures these various agents collaborate seamlessly, complementing rather than competing against each other. For citizen developers, simplified orchestration interfaces integrated directly within low-code/no-code platforms will be a game changer, dramatically reducing complexity and improving user experience.

This orchestrated approach mirrors strategic insights from industry giants like Salesforce and ServiceNow, who recognize orchestration as key to leveraging multi-vendor, multi-agent environments successfully. Automation Anywhere’s commitment to robust orchestration tools positions them well to lead this transformative shift.

Yet, as with any technological shift, the transition from RPA to APA must be handled strategically. Automation Anywhere’s 20 years of industry experience provides a significant advantage, enabling the company to anticipate potential pitfalls and smoothly manage the transition. Their established market credibility and deep industry relationships offer solid assurance to businesses hesitant about adopting radically new technologies.

From a broader AppDev perspective, APA will significantly influence software quality, delivery speed, and development agility. Intelligent automation agents support developers by rapidly handling mundane, repetitive tasks, freeing them to focus on more innovative, strategic initiatives. Consequently, APA facilitates faster innovation cycles, improved customer satisfaction and enhanced business agility.

As businesses adopt APA-driven low-code/no-code solutions, they’ll likely witness an acceleration in their digital transformation journeys. Citizen developers, empowered with intelligent APA agents, will increasingly contribute to critical business solutions without relying heavily on IT departments. This not only democratizes software development but also enables organizations to respond more rapidly and effectively to market demands.

In conclusion, Automation Anywhere’s strategic pivot from RPA to APA doesn’t merely represent an incremental upgrade—it heralds a new era of intelligent, flexible and powerful automation that is poised to revolutionize the AppDev landscape, empowering citizen developers and accelerating innovation.

May 16, 2025
If Planes Can Fly Themselves Then Why Can’t IT Management be Autonomous?

When it comes to discussing how AI will impact the future of software development and IT management, most vendors hold on to the objective that it’s important to keep the human in the loop. They are afraid to publicly acknowledge what’s always been true — that machines are better than humans at many things. Meanwhile, that list of things continues to grow.

When it comes to observability and IT operations, our goal should be to get humans out of the loop as much as possible. In a world run by software, where SLAs guarantee 99.999% uptime, even one incident in a month — which requires human intervention — is enough downtime to violate your commitments to your customers.

With the rise of distributed, complex applications exacerbated by more and more AI-generated code, it is difficult for traditional IT operations to keep up across troubleshooting, provisioning infrastructure, performance monitoring, security, etc. Developers now deploy code 70% faster using AI, according to the latest DORA report — you can’t have enough SREs and security personnel to keep up with this kind of activity. The companies that succeed will leverage technology to get humans out of the way as quickly as possible.

Failed Promises of AIOps

It is worth stating the obvious that there’s a difference between where things are today and where they are heading in the future. As someone who has been in IT management for decades, I have seen all the buzzwords as well as both marketing nonsense and genuine progress.

The term ‘AIOps’ was coined by Gartner in 2016 to address the increasing complexity and data volume in IT environments, aiming to automate processes such as event correlation, anomaly detection and causality determination. But as it turned out, many of the vendors who claimed to offer AIOps were nothing more than empty shells when you looked under the hood.

It was essentially the same AI/ML that had been used for a decade beforehand, branded in a new way and making outsized claims that didn’t map to reality. We saw tech giants making acquisitions of point solutions, and then bundling them under the ‘AIOps’ category because it was trendy to do so and they had nowhere else better to put them.

But I would argue that many of the companies that continued to trumpet these claims without actually delivering on the promises eventually suffered a hit to their reputation. We still aren’t at that stage where machines are acting autonomously to solve complex problems across performance, reliability and security. But they are doing more than ever before, and there’s every reason to believe we will reach that goal in the future.

The Future of Autonomous Reliability

Autonomous reliability platforms of the future will not only surface actionable insights, but they will also be competent enough to make autonomous decisions without human intervention. And why should this be impossible? If planes can mostly fly themselves, why can’t IT management become autonomous?

The decades-long trend of collecting more and more data in the name of observability isn’t rendering autonomous service reliability. Nor is it feeding that data into a machine and magically hoping for answers. Machines trained on yesterday’s patterns might explain what went wrong in the past, but they can’t make the real-time decisions required to keep systems running, especially in dynamic, cloud-native environments.

What we need is a paradigm shift from data collection to causal understanding. By capturing causal knowledge as part of an ontology, we can reason about cause-and-effect in complex, ever-changing systems — the key is to move beyond reactive alerting into autonomous reliability. Unlike the growing industry trend of offloading alerts to LLMs, causal reasoning gives us the context and clarity needed to take real control.

May 16, 2025
Survey Surfaces AI Contest for the Hearts and Minds of DevOps Engineers

A report published by the Futurum Group finds organizations are exploring multiple paths to employing artificial intelligence (AI) capabilities across the software development lifecycle (SDLC).

Over the next 12 to 18 months organizations plan to increase spending on not only on AI code generation (83%) and agentic AI technologies (76%), but also existing familiar tools that have been augmented with AI, according to a survey of 855 IT leaders conducted by the Futurum Group.

While many DevOps teams will undoubtedly employ a mix of these approaches, it’s clear there is a battle underway for the hearts and minds of software engineers, said Mitch Ashley, vice president and practice lead for DevOps and application development at the Futurum Group.

Providers of application development tools and platforms have reached a pivotal AI moment as DevOps teams weigh the merits of switching platforms versus relying on incumbent vendors to bolt on additional AI capabilities, he added. Ultimately, success will hinge on lowering barriers to entry, go-to-market models and demonstrating tangible value at a time when the way software is being created is fundamentally changing, noted Ashley.

It’s still early days so far as adoption of AI is concerned but application development teams are clearly at the forefront. Previously, research from the Futurum Group revealed that 41% of the respondents it surveyed expect generative AI tools and platforms will be used to generate, review and test code.

Less clear is how much faith application development teams will have in those tools and platforms. It’s one thing to rely on a generative AI tool to explain how an existing code base is structured, but the quality of the code that might be generated can vary substantially. In some instances, such as generating a script for invoking an application programming interface (API), the quality of the code might not matter. However, the quality of the code being used to create business logic is critical, especially if that code might contain known vulnerabilities that were never properly scanned.

Of course, depending on the experience of the application developer, there are also plenty of instances where the code being generated might be better than the code that a human developer might have created on their own. As a result, the return on investment in AI tools and platforms will tend to vary widely from one DevOps team to the next.

Ultimately, advances in AI should bring about more positive than negative changes, especially as more AI agents that have been narrowly trained to automate a specific set of tasks become more widely employed. The simple fact is that there are still far too many manual tasks that make building and deploying software more tedious than anyone really enjoys, which only leads to higher turnover rates as software engineers inevitably become burnt out.

Conversely, however, DevOps teams should also be wary of becoming overly dependent on AI. After all, the one golden rule of application development that still applies in the age of AI is that no code should be deployed in a production environment that one or more humans don’t thoroughly understand.

May 15, 2025
Will Policy-as-Code Still Matter if AI Generates Most Code?

As artificial intelligence (AI) code-generation tools like GitHub Copilot and ChatGPT revolutionize software development, a critical question emerges: If AI writes most code, do we still need policy-as-code (PaC) to govern it? The answer isn’t just ‘“yes’” — it is a resounding ‘“now more than ever’.” To understand why, we must explore PaC’s evolution, its role in modern infrastructure and how it will anchor trust in the AI-driven future.

What is Policy as Code? A Governance Layer Born From Chaos

PaC is the practice of defining security, compliance and operational rules in machine-readable formats, enabling automated enforcement across systems. It emerged in the shadow of infrastructure as code (IaC), which allowed teams to programmatically manage servers, networks and cloud resources. However, as IaC scaled, manual policy checks became untenable.

Imagine deploying 100 cloud instances via IaC, only to realize half violate cost or security rules. PaC solved this by codifying policies (e.g., “no public S3 buckets”) directly into pipelines, blocking non-compliant resources before deployment. Tools like Kyverno and Open Policy Agent became the enforcement backbone for DevOps and platform teams.

Why is Policy-as-Code Having its Moment?

The transition to declarative systems such as Kubernetes, Terraform and serverless architectures introduced a powerful new paradigm — one where engineers specify the desired end state rather than the implementation details. While this abstraction accelerates development, it also introduces governance challenges. PaC has emerged to bridge these gaps by embedding policy enforcement directly into the software life cycle.

PaC offers key advantages: Flexibility to adapt policies across diverse environments — cloud, on-premises and hybrid — without rewriting workflows; transparency through human-readable code that fosters collaboration among developers, security teams and compliance officers; automation at scale, enabling real-time enforcement across complex infrastructures like Kubernetes clusters and continuous integration and continuous deployment (CI/CD) pipelines; and seamless integration with modern development workflows, allowing policies to provide immediate feedback and shift enforcement left in the development process.

Together, these capabilities make PaC an essential pillar in governing modern, declarative systems.

The Future of Policy as Code in the Age of AI

The Premji Invest blog, “AI Is Coming: Meet the Startups Building Cyber Defenses for the Age of AI,” highlights the unprecedented risks introduced by AI, ranging from data leakage in training pipelines and adversarial attacks on models to regulatory ambiguity. In response, startups are increasingly integrating PaC into AI workflows. However, PaC’s role is rapidly evolving in light of three seismic shifts.

First, the rise of Agentic AI — autonomous agents capable of managing infrastructure or making decisions — demands runtime policy enforcement. PaC provides a critical framework for governing these agents, defining clear boundaries around what data they can access, what actions they can perform and when human approvals are needed for high-impact operations.

Second, as AI models dynamically adapt to new data, a model context protocol (MCP) becomes essential. By integrating PaC with MCP, organizations can ensure AI models operate within contextual boundaries, preventing unauthorized access to sensitive data and enforcing compliance with regulatory constraints (e.g., blocking a medical AI from inferring patient identities).

Third, the emergence of agent-to-agent (A2A) ecosystems — where multiple AI systems collaborate across domains, such as supply chain management — necessitates cross-system policy harmonization. PaC can codify authentication standards, interaction limits and data sharing rules, maintaining organizational control even as agent interactions grow more complex.

In this rapidly shifting landscape, PaC is no longer just a static set of rules — it is becoming a set of adaptive guardrails for governing AI behavior, managing data flows and upholding ethical boundaries.

Will Policy as Code Still Matter if AI Generates Most Code? Yes, Here’s Why

AI is transforming software development, but its integration into workflows is not without risks. While AI-generated code can speed up development, it may also introduce vulnerabilities, like insecure application programming interface (API) endpoints, or overlook critical compliance requirements, such as the General Data Protection Regulation (GDPR) data residency rules. PaC acts as a crucial safety net, scanning AI-generated code for risks and automatically rejecting violations.

Beyond technical issues, policies also reflect human values such as privacy, fairness and legality. While AI may produce efficient solutions, it should not be left to decide which data to collect or how to weigh profit against ethical considerations. PaC encodes these human-centric judgments, aligning AI outputs with organizational principles.

As demands for transparency grow, driven by regulations like the European Commission’s EU AI Act, PaC offers explicit, auditable rules, providing clarity where AI-driven decision-making might resemble a black box. When it comes to accountability, “show me the policy” is more reliable than “trust the AI.” Additionally, as AI continues to evolve, so too will the threats against it will evolve as well. PaC provides a programmable, adaptable defense layer that can keep pace with emerging vulnerabilities, including those that originate in AI-generated code.

Ultimately, the future is not about choosing between PaC and AI — it is about leveraging their synergy. Imagine a workflow where AI drafts policy templates based on regulatory changes, humans refine them to include ethical and strategic nuances, PaC tools enforce them across systems and feedback loops help AI improve policy design. This cycle unites AI’s speed with human judgment and PaC’s rigor, creating a more secure and ethically aligned AI ecosystem.

Policy as Code is the Bedrock of AI Trust

As AI reshapes software development, PaC won’t just survive — it will thrive. It bridges the gap between innovation and governance, ensuring AI serves humans, not the reverse. Whether guarding Kubernetes clusters or GPT-4 workflows, PaC is the immutable layer that makes automation responsible.

In the age of AI, writing code is no longer a human monopoly. However, defining what that code should do— and what it must never do— will always be up to us. PaC is how we will keep it that way.

The call to action for organizations is to start embedding PaC into your AI workflows today. Tools like Kyverno (for Kubernetes) and emerging AI-focused platforms are your first line of defense in the automated future.

May 14, 2025
Survey: Software Quality Continues to Be Sacrificed in the Name of Speed

A global survey of 2,750 IT, DevOps leaders and software developers finds nearly half of respondents (45%) admit their organization prioritizes delivery speed over software quality, with 63% acknowledging their organization deploys code without fully testing it. Only 13% of respondents said their organization is focused on improving software quality.

Conducted by Censuswide on behalf of Tricentis, a provider of an application testing automation platform, more than one-third (34%) of respondents say ongoing maintenance and unresolved technical debt are the largest obstacles to delivering high-quality software. Other issues include poor communication between software development and quality assurance teams (33%) and a disconnect between leadership and development teams (29%).

The primary reasons cited for not testing code are the need to expedite release cycles (46%) and accidental slips (40%). Just under a quarter (24%) also admitted they are not confident in their test coverage.

More troubling still, 81% said software quality issues cost their business between $500,000 and $5 million every year, with nearly 40% estimating losses of more than $1 million. Two-thirds (66%) of respondents also conceded their organization is likely to suffer an outage or major disruption in the next year, with nearly a quarter (23%) deeming their organization to be at extreme risk.

Many respondents noted there is a lack of clear quality metrics provided by the organization (29%), a disconnect between leadership and development teams over what defines quality (28%), and budget constraints (25%). The top five benefits they seek from better software quality include improved team efficiency (17%), stronger customer satisfaction and retention (15%), reduced security and compliance risk (14%), increased revenue (14%), and faster delivery (14%).

Lee McClendon, chief digital and technology officer for Tricentis, said when it comes to software engineering there continues to be too much emphasis on speed over quality. The irony is that addressing quality issues in the latter stages of a DevOps workflow only serves to slow organizations down further when work needs to be redone, he added.

There is simply too much focus on metrics that track how fast software is developed versus how well, noted McClendon.

On the plus side, a total of 80% of respondents said they believe that advances in artificial intelligence (AI) will enable them to improve software quality without adversely impacting the rate at which software is being developed. Nearly 90% also said they can effectively measure the return on investment on generative AI tools and platforms. However, two-thirds (66%) noted it takes investments in software development and delivery at least three years to pay off, with over one-third (34%) anticipating AI technology investments will take even longer.

The majority (82%) also said they are excited about AI agents taking on repetitive tasks in the software development lifecycle, while 84% believe AI will help teams deliver under increasingly tight deadlines. A full 89% said they trust AI agents to make decisions that impact software delivery. At the same time, bias remains a key concern for 38% of respondents, and the same percentage are wary of over-reliance on AI tools.

Finally, nearly all respondents expect autonomous testing will be useful for quality assurance (QA) testing, with the biggest impact being improving software performance (28%), overall quality (28%), analyzing test results (25%) and test case maintenance (23%).

Hopefully, the quality of applications will improve in the age of AI, but the one thing that is certain is that when it comes to achieving that goal there is still a lot of room for improvement.

May 13, 2025