devopsy.kinsta.cloud

Tag: ai

Survey Surfaces Downstream DevOps Challenges Created by AI Code

A survey of 900 engineers, platform leaders and technical managers published today finds nearly two thirds (63%) report their organization is shipping code faster since adopting artificial intelligence (AI).

Conducted by Coleman Parkes on behalf of Harness, a provider of a DevOps platform, the survey also suggests, however, that same code is creating more downstream DevOps issues, with 72% of respondents noting their organization has already suffered at least one production incident caused by AI-generated code.

In total, 45% of respondents said that deployments involving AI-generated code have been problematic. More challenging still, only 6% of respondents said their continuous delivery process is fully automated. A full 83% say AI must extend across the entire software delivery lifecycle to unlock its full potential.

Additionally, 63% said vibe coding using AI agents is “a disaster waiting to happen” in the sense that DevOps engineers will be overwhelmed by downstream rework, including reviewing code that may often be fundamentally flawed because it was created using a general-purpose large language model (LLM). Nearly three quarters (73%) are concerned unmanaged AI assistants could significantly widen the blast radius of failed releases.

Nearly half (48%) of respondents also expect AI to increase software vulnerabilities and only 41% are confident their existing governance tools and platform will reliably catch issues before release. Nearly three quarters (73%) also said they believe organizations that fail to integrate AI safely and securely across their software development lifecycle (SDLC) in the next year will “go the same way as the dinosaurs.”

Finally, most survey respondents (70%) are also concerned inefficient AI-generated code will drive uncontrollable cost overruns once it is deployed in a production environment.

Trevor Stuart, senior vice president and a general manager for Harness, said the survey makes it clear that much of the code being created by AI tools is being sent back to the proverbial kitchen to make it palatable enough to deploy. At the same time, many DevOps teams now feel like AI has created a six-lane highway for application developers that leads to a two-lane bridge operated by a DevOps team, he added.

That bottleneck will inevitably require DevOps teams to modernize their pipelines and workflows to make them efficient enough to manage higher levels of throughput, noted Stuart. However, the goal is to not necessarily add more DevOps pipelines so much as it is to enable DevOps teams to leverage AI to more efficiently manage the same or even fewer number of reusable pipelines, he added.

Achieving that goal will require those DevOps teams to not only better manage their pipelines, but in many cases rationalize their existing tools, said Stuart. On average, the survey finds development and engineering teams are already using eight to 10 distinct AI tools, while 36% of respondents juggle even more. A full 71% say context switching between those AI tools actually drains productivity.

It’s not clear how long it might be before DevOps teams are able to take greater advantage of AI to manage the current onslaught of code moving through their pipeline, but the one thing that should be clear to everyone by now is that writing more code does not directly equate to deploying more applications faster.

September 30, 2025
Latest DORA Report from Google Surfaces Significant AI Adoption

Google this week published the results of its annual DevOps Research and Assessment (DORA) survey that finds 90% of the IT professionals surveyed are now using artificial intelligence (AI) tools, with 80% reporting they are more productive as a result.

The survey of nearly 5,000 IT professionals also finds that respondents now have a median average of 16.22 months of experience using AI tools, with respondents spending a median of two hours of their most recent workday interacting with AI, representing about one-quarter of an eight-hour workday.

Among AI users, a full 60% of AI users said they employ it about half the time or more when encountering a problem or completing a task. Only 7% report “always” using AI when faced with a problem to solve or a task to complete, while 39% said only sometimes do they seek AI help.

Overall, 65% report relying on AI a moderate amount (37%), a lot (20%), or a great deal (8%). The number one use case for AI is writing new code (71%), followed by analyzing requirements (49%), internal communications (48%) and calendar management (25%). A large number of respondents also use AI for literature reviews (68%), modifying existing code (66%), proofreading (66%), and creating or editing images (66%).

The survey also, however, uncovered a significant amount of AI skepticism, with 30% of the technology professionals surveyed reporting little or no trust in the code generated by AI.

Mitch Ashley, vice president and practice lead for software lifecycle engineering at the Futurum Group, said the results make it clear that while AI is an accelerator it is not a cure-all for software development. Teams with strong internal platforms, disciplined engineering practices, and clear workflows are seeing AI multiply their speed and impact more than others, he added.

For example, the survey finds that 90% of organizations have adopted at least one platform and there is a direct correlation between a high quality internal platform and an organization’s ability to unlock the value of AI, making it an essential foundation for success.

In this latest edition of the DORA report, Google is also now identifying seven common team profiles or archetypes, based on a cluster analysis of the survey data it collected, with the lowest ranking having foundational challenges (10%) while at the top end are harmonious high achievers (20%). The more mature the DevOps team is in terms of its ability the greater the impact AI is having, according to the DORA report. Assessments of where teams fall on that spectrum is based on lead times for changes, change failure rates, deployment frequency, rework rates and recovery times for failed deployments.

The takeaway for technology leaders is that they should invest first in the quality of the system of work across platform engineering, automated testing, and DevOps feedback loops, said Ashley. AI can then deliver sustainable gains in software delivery and product performance, he noted.

Overall, the DORA report finds that the top two clusters of the archetypes represent nearly 40% of the total sample, which the report claims provides proof there doesn’t need to tradeoff between speed and stability because top performers are excelling at both.

It’s not exactly clear how much AI is accelerating the pace at which applications are deployed but there is evidence to suggest DevOps teams are becoming more efficient. However, reliance on AI can vary widely depending on the nature of the task. Writing a script to automate a DevOps workflow doesn’t typically require the same level of depth and expertise as business logic. At this point, however, it’s no longer a question of whether development teams will use AI, but to what degree.

September 27, 2025
MLOps at Scale: How Community Is Driving AI Into Production

At swampUP 2025, Alan caught up with Demetrios Brinkmann, founder of the MLOps Community, to discuss the growing gap between AI research and real-world production deployments. Brinkmann leads a global network of more than 100,000 developers dedicated to bridging that divide, helping teams move beyond flashy demos and academic models to systems that deliver tangible business value.

The MLOps Community provides multiple ways for practitioners to engage: a Slack workspace, curated one-on-one matches, in-person meetups, workshops, conferences, virtual events, a newsletter, and a popular podcast. The goal, Brinkmann explained, is to raise the overall level of education and awareness across the field while connecting practitioners who are tackling similar challenges in bringing AI and ML into production.

One of the recurring themes is the operational complexity of scaling AI workloads. Many organizations can train models, but deploying them reliably, monitoring performance, and ensuring governance at scale requires different skill sets and tools. The community’s role is to share hard-won lessons from practitioners and highlight emerging best practices in areas like observability, CI/CD for ML models, and managing technical debt created by fast-moving experimentation.

Brinkmann noted that with the rise of generative and agentic AI, the stakes are even higher. Teams are under pressure to move fast, but deploying AI without proper guardrails risks security lapses, compliance failures, and business disruptions. The MLOps Community’s mission is to give practitioners a space to navigate these issues openly and collaboratively, rather than reinventing the wheel in isolation.

As Brinkmann put it, AI won’t transform industries unless it can be reliably operationalized — and that’s where communities like his make the difference.

September 26, 2025
Securing the Software Supply Chain with Full Visibility and Compliance

At swampUP 2025 in Napa Valley, JFrog’s Yossi Shaul, senior vice president of DevOps, joined Alan to reflect on both his personal journey and JFrog’s evolving role in shaping modern software delivery. Shaul has been with JFrog since its early days, starting with the creation of Artifactory nearly 16 years ago — before “DevOps” was even a recognized term. He recalled leading the Artifactory team, driving its adoption, and later stepping into architectural leadership before returning to oversee JFrog’s entire DevOps organization across product and engineering.

Shaul highlighted how much has changed since those early years. What began as a simple repository manager has grown into a platform that now touches nearly every aspect of software supply chain management. With the rise of AI, security concerns, and increasingly complex pipelines, the pressure on DevOps teams has only intensified. JFrog’s answer has been to expand beyond artifact management into governance, automation, and AI-driven workflows — all areas Shaul now helps drive.

One theme that stood out in the discussion was continuity. Shaul’s long tenure gives him a unique perspective on how JFrog has evolved with the industry — from open source beginnings to enterprise adoption, from managing binaries to securing supply chains, and now from human-driven workflows to AI-assisted delivery.

Shaul also emphasized the importance of listening to customer feedback. swampUP, he noted, is not just about unveiling what’s next but about validating that the technologies JFrog delivers address real-world challenges DevOps teams face daily.

JFrog’s DNA remains rooted in DevOps, even as it pushes into the AI era. For Shaul, the mission is clear: build platforms that help teams deliver software faster, more securely, and with confidence at scale.

September 26, 2025
Zencoder Adds CLI Edition of AI Agent That Generates Code

Zencoder this week added a command line interface (CLI) option to its artificial intelligence (AI) agent for creating code.

Company CEO Andrew Filev said application development teams can use Zen CLI as an alternative to previous support for integrated development environments. The overall goal is to provide developers with a range of options to take advantage of AI coding from within whatever development environment they prefer, he added.

Zencoder is making a case for an AI agent that simultaneously indexes code patterns and dependencies across repositories. That capability then makes it easier to generate and update code across multiple files in a way that also validates code against syntax rules, project conventions, and integration requirements.

Designed to be integrated with continuous integration/continuous deployment (CI/CD) platforms such as Jenkins and others, the Zencoder AI agent can resolve issues, implement fixes, improve code quality, generate and run tests, and create documentation. As such, the goal is not just to write more code faster, but rather enable DevOps teams to take advantage of AI agents running in the background to re-engineer workflows in ways that result in more applications being deployed faster, said Filev.

Less clear is to what degree AI tools are benefitting senior developers more than junior ones. In theory, AI tools should automate tasks that were once handed off to junior developers. However, it’s also arguable that as AI agents evolve they will be able to take advantage of more advanced reasoning capabilities to build more sophisticated applications without the aid of a senior developer. One way or another, AI will make it easier for generalists to build, deploy and update applications, noted Filev.

Regardless of who is using these tools, the volume of code being generated is already substantially increasing. However, the quality of the code being generated can vary widely. In addition to creating more vulnerabilities because many of the examples used to train large language models (LLMs) are from a cybersecurity perspective significantly flawed. More challenging still, much of that code is verbose which many developers are finding challenging to debug simply because they didn’t write the code themselves.

Additionally, application development teams need to carefully consider which LLMs are being used to build applications. The cost of using one model versus another can vary substantially, so some care needs to be taken to use the right tool based on actual business value that application represents, noted Filev.

At this point, it’s only a matter of time before code being created using AI tools will overwhelm existing DevOps pipelines and workflows so the need for AI agents to automate more tasks is rapidly becoming self-evident. In fact, there is already no shortage of AI tools and platforms for writing and testing code. Unfortunately, many organizations still lack a cohesive strategy, so many individual developers for better or worse are naturally experimenting with different tools and approaches on their own. The challenge now is deciding which one of these tools and platforms best suits the needs of the organization versus being a tool that one individual developer happens to prefer over another.

September 19, 2025
Empowering Secure, Agentic Software Delivery

At swampUP 2025 in Napa Valley, JFrog co-founder and CTO Yoav Landman joined Alan to reflect on the company’s latest announcements and the broader direction of software delivery in the AI era. Landman noted that this year’s event may have been JFrog’s most ambitious yet, with a wave of product releases designed to tie together themes of trust, automation, and AI-driven DevOps.

One highlight is the company’s push to unify governance through AppTrust, which provides end-to-end traceability and compliance across the software supply chain. The aim is to ensure that every artifact — whether human- or AI-generated — can be validated and trusted before moving through the pipeline.

Another centerpiece is JFrog Fly, billed as an “agentic repository” that enables AI agents to securely interact with APIs and development workflows via the Model Context Protocol (MCP). As AI coding tools become commonplace, Fly is intended to help DevOps teams manage the complexity of agent-driven development while maintaining control and visibility.

Landman emphasized that these aren’t abstract roadmaps — they’re production-ready features shaped directly by customer feedback. The feedback loop from swampUP, he noted, often influences products within months rather than years, especially as AI accelerates iteration cycles.

For Landman, the common thread across these initiatives is addressing the realities of modern DevOps: accelerating release velocity while ensuring governance and security don’t lag behind. With software supply chains under constant pressure and AI poised to reshape workflows, he argued that platforms like JFrog must become the system of record for DevOps in order to keep pace.

The conversation underscored the urgency: in a world where AI is changing how code is created and deployed, trust and automation aren’t optional — they’re foundational.

September 19, 2025
Building End-to-End Trust in the Software Supply Chain

At swampUP 2025, JFrog’s Guy Levi, head of the architects team, highlights some of the technologies shaping the company’s next chapter. Levi’s role centers on exploring advanced technologies “beyond the horizon,” but he emphasized that swampUP isn’t about future promises — it’s about showing what’s available today.

One of the highlights Levi pointed to was AppTrust, JFrog’s initiative to establish end-to-end trust across the software supply chain. By unifying governance, risk, and compliance capabilities into a single framework, AppTrust is designed to give enterprises more confidence that applications are secure and reliable from development through deployment. The goal is to tie disparate security and verification processes into one cohesive approach that simplifies how organizations enforce trust at scale.

Another major focus was JFrog Fly, which Levi described as a breakthrough in supporting AI-era DevOps. Fly provides an “agentic repository” that integrates with Anthropic’s Model Context Protocol (MCP) to enable AI agents to securely interact with APIs and development pipelines. With AI-generated code and autonomous agents becoming part of daily workflows, Fly aims to ensure those components can be governed, tracked, and optimized just like traditional software artifacts.

Together, these initiatives underscore JFrog’s push to position itself as the system of record for modern DevOps and AI-driven development. AppTrust tackles the governance side, while Fly addresses how AI agents and automation will reshape pipelines.

For Levi, the key message was that the industry can’t afford to wait for “next year” to solve these challenges. AI and security pressures are already here, and JFrog is working to deliver the tools teams need now to navigate them.

September 19, 2025
Honeycomb Adds Ability to Orchestrate Multiple AI Agents to Observability Platform

Honeycomb has added a Canvas framework for managing and orchestrating multiple artificial intelligence (AI) agents as they observe and query massive amounts of telemetry data.

Morgante Pell, technical lead manager for engineering at Honeycomb, said Canvas is the latest extension to a series of Honeycomb Intelligence initiatives that, for example, provide access to an AI agent that can convert natural language into queries based on the syntax Honeycomb created to query data. There is also now a Model Context Protocol (MCP) server that provides third-party agents and coding tools with access to observability data.

These capabilities will enable DevOps teams to both more easily interrogate terabytes of telemetry data and be alerted any time there is anomalous behavior that might adversely impact application performance and availability.

Honeycomb also plans to add additional AI agents that are capable of completing tasks that run in the background later this year, he said. That will make it possible for an AI agent to, for example, autonomously observe application behavior and surface remediation suggestions that might be required, said Pell.

Additionally, Honeycomb is working toward integrating the codebase analysis tools it gained earlier this year via the acquisition of Grit into the Honeycomb Intelligence framework.

The overall goal is to make observability data more accessible to not just DevOps engineers but also additional members of the IT teams that no longer need to know how to craft a Honeycomb query, noted Pell.

Arguably, one of the reasons that observability platforms are not more widely employed is that many IT professionals simply lack the expertise required to create the query needed to analyze data in the first place. In effect, AI is democratizing access to observability platforms.

Less clear is to what degree AI may encourage more IT teams to rationalize legacy IT monitoring platforms that are designed to track a set of pre-defined metrics. While observability platforms are today used primarily to troubleshoot application environments, they can also be used to track events. As AI tools become more deeply embedded within the observability platform, it should become easier to use a single platform at a lower total cost to both monitor metrics and analyze logs and traces.

Ultimately, each IT organization will need to decide to what degree it may empower AI agents to autonomously act on guidance surfaced by an observability platform, but in most instances a DevOps engineer is going to want to review them before allowing them to be implemented in a production environment. Eventually, however, platforms will automatically handle issues with relatively low risk of impacting an application, while routing issues that could impact customers being sent to an on-call system to ensure they get immediate attention from a DevOps professional.

Hopefully, in the age of AI there will soon be a lot less firefighting as more issues are resolved before they can have a catastrophic impact on the application environment. The challenge now will be organizing the small army of AI agents that might be needed to achieve that goal.

September 17, 2025
Pulumi Previews AI Agents Trained to Automate Infrastructure Management

Pulumi today revealed it is now making available in preview a set of artificial intelligence (AI) agents to its namesake infrastructure-as-code (IaC) platform that are capable of diagnosing issues and autonomously completing tasks when permitted.

Company CEO Joe Duffy said Pulumi Neo adds a set of AI agents specifically trained to automate the management of IT infrastructure using data collected by Pulumi. In effect, Pulumi Neo adds a member to the DevOps team that understands dependencies, executes changes, monitors outcomes, and maintains compliance throughout the entire infrastructure lifecycle, he added.

Neo is designed to enable DevOps teams to launch short- or long-running tasks with approvals and interactive guidance provided in a way that adheres to the policies enforced by the Pulumi platform, said Duffy. It also provides a complete history of which tasks were automated to enable organizations to meet compliance requirements, he noted. In effect, Neo becomes a digital member of the platform engineering team, added Duffy.

Each DevOps team will need to decide for themselves to what degree to rely on AI agents to automate tasks. In the case of Neo, for example, DevOps teams can use it to simply generate optimization suggestions or alternatively autonomously perform tasks based on permissions that have been assigned.

Additionally, Neo will also analyze the underlying IT platforms being used to run certain classes of workloads, which may lead to it recommending, for example, that certain cloud-native application be migrated from Kubernetes to a simpler environment such as the Elastic Container Service (ECS) from Amazon Web Services (AWS).

Finally, Neo has also been included in the Model Context Protocol (MCP) server developed by Pulumi, which means AI coding tools can invoke it as well to create agentic AI workflows that conform to a set of best DevOps teams.

It’s not clear how rapidly DevOps teams are embracing AI agents to automate tasks, but as they do the overall level of toil encountered should continue to be reduced. Early adopters of Neo such as Werner Enterprises are, for example, reporting they reduced infrastructure provisioning time from three days to four hours while both maintaining SOC 2 compliance and enabling development teams to ship features 75 percent faster.

Hopefully, as AI agents become more pervasively employed the number of misconfigurations leading to cloud security incidents will decline. While application developers routinely use IaC tools to provision cloud platforms, they don’t always have the greatest appreciation for best DevSecOps practices, which can lead to ports through which data can be exfiltrated being left wide open.

Of course, it might be a while yet before DevOps engineers are completely comfortable with AI agents but given the volume of workloads that will need to be deployed in the AI era the need for those agents is already apparent. The issue, as always, is making sure that whatever winds up being automated can also be undone should, for whatever reason, the AI output being provided winds up being worse than the proverbial disease it was meant to cure.

September 16, 2025
MongoDB Taps AI Agent to Launch Application Modernization Platform

MongoDB today unveiled an application modernization platform that makes use of artificial intelligence (AI) agents to analyze and convert legacy applications into code that can be used to deploy a modern application on its document database.

Shilpa Kolhar, senior vice president of product and engineering for MongoDB, said the MongoDB Application Modernization Platform (AMP) will make it possible to refactor code in a way that is compatible with the open source Java Spring framework, which in turn could then be deployed on MongoDB.

The overall goal with the help of a MongoDB consulting team is to accelerate the pace at which an application modernization project could be completed by a factor of two to three times, she added. Early adopters of MongoDB AMP include Bendigo Bank, Lombard Odier and IntellectAI.

MongoDB has gained significant traction in recent years as a platform for deploying modern greenfield application, but with the rise of AI, the time and effort required to refactor legacy applications is dropping. The challenge, of course, is that refactoring legacy applications has typically required months, sometimes even years, of effort usually involving expensive professional services organizations.

However, AI agents are now making it simpler to analyze and convert code in ways that promise to sharply reduce the amount of time that historically has been required to complete an application modernization initiative. DevOps teams should still review that code before it is deployed in a production environment, but the overall amount of effort required should be considerably reduced.

It’s not clear to what degree organizations are looking to modernize legacy applications but each IT platform they need to support increases total costs. The challenge has been that migrating an application from one database to another has been especially difficult. While AI agents are not going to automate that entire process, they should reduce a lot of the tedious effort currently required.

Of course, MongoDB will not be the only provider of an IT platform that is employing AI agents to help organizations migrate legacy applications to a modern platform that is less expensive to deploy and maintain. In fact, in the age of agentic AI it is now only a matter of time before the cost of switching from one IT platform to another declines to the point where the ability to reverse engineering code advances to a level where it becomes difficult to lock any organization into a particular platform. As a result, DevOps teams might want to start identifying applications that they now want to modernize that previously would have simply been too complex an undertaking to attempt.

In fact, the impact of AI is likely to be way beyond simply making it possible to write new application code faster, but also address a lot of the technical debt created by legacy applications that no one today knows for sure how they were actually built in the first place.

September 16, 2025