The state of automation and AI

Introduction

For IT professionals, increasingly sophisticated AI tools dangle the promise of finally getting IT complexity under control for their organizations. According to our recent survey data, 75% of IT pros find their organizations’ tech stacks either moderately or quite complex, and any drive to streamline and automate must arm-wrestle with competing priorities, like modernizing core infrastructure and maintaining current operations. (IT Brew’s survey included 241 IT pros, and asked them about everything from ROI on automation to the complexity of their tech stack.)

As we’ll explore, IT pros who want to automate their infrastructure must scale AI and automation pilots, figure out what they want to optimize, and solve structural challenges. It’s a hard road, but a worthy one.

Given the potential expenses and risks, companies need a good reason to embrace automation and AI—and it can’t be the classic, “Everybody’s doing it!”

IT Brew’s survey respondents ranked two main reasons as the most important for adopting AI and automation in their organizations: “improving operational efficiency” and “reducing manual tasks.”

Let’s take a closer look at examples of those ideas in action:

• Improving operational efficiency. Russell Levy, chief strategy and AI officer at ZoomInfo, saw AI as a way of bringing disparate data sources—including structured Snowflake data, Salesforce data, and product data—into the hands of his 3,000-plus employees. “If you want to build reports, you should have all the tools you need to be able to build those reports,” he said.
• Reducing manual tasks. In the past, when a customer asked a sales-team member from cloud consultancy Innovative Solutions for a “one-pager” of company services, that could mean a busy afternoon; the sales pro would have needed to research details and present them in the appropriate company formatting, according to Travis Rehl, the company’s CTO and head of product. To speed things up, Rehl created an agentic system that detects the request as a to-do item and pulls the resource together.

There is a slight element of everybody’s doing it pressure. Eyal Bukchin, CTO and co-founder of software developer tool builder MetalBear, has encouraged his teams to use AI for tasks like script creation. He sees the technology as a way of delivering quality features more quickly: “Our competitors are picking these tools up, and we can’t get…behind.”

But 54% of our respondents said their biggest implementation challenge was a lack of clear strategy. When it comes to automation, where do you even begin?

For many organizations, the road to a strategy starts with experimentation and small tests, which can often reveal issues before a team makes a broader commitment.

Boris Kolev, global head of technology at entrepreneurial education nonprofit JA Worldwide, tried two AI ideas in the summer of 2025. One proof of concept (PoC) consolidated financial reporting from 100-plus participating countries, and another scanned the internet continuously for open grants.

The projects saved time for the organization’s early AI adopters, Kolev said, but he faces two main challenges as he expands automation:

• Costs are unpredictable. Sometimes employees use a lot of tokens, the data units that a large language model processes to create text, whereas other times, like a professor teaching a Friday afternoon class, the LLM doesn’t receive many questions. “The main problem is that I cannot give my CFO a projection of the cost,” Kolev told us.
• ROI is also volatile. Kolev currently tracks cost, token usage, query topics, and query quantities. He’s also considering monitoring for large file outputs (which suggest to Kolev that a team is using the platform for a valuable business task), but “it’s very hard to measure any return on investment.”

Muqsit Ashraf, group chief executive for strategy at Accenture, says small AI experiments need to connect to enterprise value, and orgs must be “more surgical” about their PoCs to find ones contributing to company growth, including the development of new products and services.

For many organizations that decide to automate significant chunks of their IT stack, a pilot is crucial for figuring out which functions can and should be automated. IT Brew’s survey showed that companies are deploying automation and AI across a variety of functions, including cybersecurity (24%), IT service management workflows like ticket routing (39%), and in IT infrastructure and endpoints like provisioning and patching (21%).

Pilots can set the foundation for future automation projects. Jerry Shu, co-founder and CTO of Daylit, a company that provides AI agents for accounts receivable processes, told IT Brew companies should start by defining what “value” they want to achieve through their automation initiatives, as what’s valuable to one company may differ from another.

“What is value in this modern world? In capitalism, that’s just a dollar amount…but in [a] real business operation perspective, how do people evaluate that value is very, I would say, abstract,” Shu said.

For example, he added, ​some organizations will typically equate value in automation projects with time saved on simple tasks, such as how long it takes to send an email. Value from AI projects, however, can also take on less quantifiable forms, like what employees gain in mental health benefits as automation reduces their stress levels.

Trying to find a win in the pilot stage can feel like finding a proverbial needle in a haystack, especially if there are significant up-front costs. Shu advises companies to establish their specific priorities during an early stage and remain open to experimentation and pivoting.

“As soon as you start trying [to automate tasks], you will immediately get feedback on, ‘Okay, this item is too big to automate,’ or, ‘This item is so easy,’” he said.

With those critiques in mind, he suggested companies tinker with their strategy as needed, and create a prototype that people can try. The more people engaged in the process, the more “robust” it becomes.

IT pros who have embarked on their automation journey may dream of the day when efficiency gains are finally reaped. However, it’s not always a straightforward path to getting there.

More than one-third (36%) of queried respondents said data quality, silos, or accessibility issues were a top challenge their team faced when introducing automation and AI to their companies. Sebastián Arriada, CIO at software product development services company Globant, told IT Brew that IT teams may find themselves inadvertently putting more work on their plate if they start automation initiatives without taking the proper steps to set their organizations up for success.

“If the systems you have are not well-prepared, the first issue you have is with data at the end,” Arriada said. “If you have AI, but if you do not have the right underlying data that support that AI, the results that you are going to get [are] not what you expect.”

IT pros can avoid complexity issues in their automation project the same way students approach studying for an intense science exam: with “chunking,” or taking things piece by piece. Pegasystems CIO David Vidoni said his organization, which he described as engaged in a late stage of automation implementation, has dodged complexity issues by breaking the project into smaller elements instead of trying to “boil the ocean” in one go.

“From a change management perspective, doing it that way gives people more comfort that whatever the automation is doing is correct,” Vidoni said. “They can actually evaluate things better if it’s done in smaller pieces.”

Like anybody who’s thrown a good party, developers have a way of pushing cleanup to the morning.

“Technical debt” traditionally refers to the not-quite-right code produced by teams favoring deployment speed over careful practices like adherence to standards and documentation. With its ability to generate enormous amounts of code in a very short time, AI may only accelerate organizations’ existing technical debt issues.

And technical debt is just one part of the complexity issue facing automation-minded IT pros. While AI and automation are widely viewed as essential tools for slicing through this clutter, 29% of respondents to IT Brew’s survey reported “slightly increased complexity” following AI deployments.

If executed incorrectly, automation may only tangle up infrastructure and workflows even more, particularly with regard to coding and data:

• Code. A code-generation agent—increasingly popular among engineering teams—leaves technical decision-making to AI and creates potential challenges for those human former decision-makers. “The AI isn’t writing the best-quality code. It’s not writing the things that are easiest to maintain,” Darren Meyer, security research advocate at appsec company Checkmarx, told us. “I think the gamble that people are taking is that AI is going to continue to be affordable and good enough at its job—and get better at its job—that it writing and making these poor technical decisions won’t matter, because it’ll fix itself in the future.”
• Data. AI introduces a shift in focus from predictable logic to a more volatile, data-dependent system—and sometimes IT pros and AI systems looking for that data hit a wall. When guiding customers in AI deployments, Brian Luckey, CIO at Integris, sees technical debt reveal itself in a company’s data silos. Banks, for example, may run on legacy systems containing data that’s hard to access via a simple API—for good reason! “You see those older systems that are rigid, and you want a fluidity when you use AI,” Luckey said.

Companies may also have lots of databases—each containing their own formats, redundancies, and abbreviations, CJ Combs, senior AI consultant business at Columbus, said, along with the number of large language models.

“What models are we using? What systems are we using? What databases are we using? It’s gluing all those together. It’s a nightmare to maintain,” Combs said.

When a company achieves an advanced stage of automation and AI deployment, the benefits become apparent. Vidoni recalled a couple of automation projects that made it to the finish line within his organization, resulting in boosted efficiencies and ROI.

One initiative within Pegasystems’s finance department aimed to automate the tedious, human-driven process of reviewing hundreds to thousands of transactions every quarter.

Another automation project involved automating its community-facing external ticketing system. Previously, a team would tend to all requests that came in from users, which often overlapped significantly.

“Now, we have AI agents that will analyze the requests coming in in the context of what’s being asked, and then serving them up with recommendations on how they could do it themselves,” Vidoni said. Automating this system has not only allowed Pegasystems to automatically address around 70% of community support tickets, but has also helped it scale its capabilities to address higher ticket volumes without adding additional members to its support team.

“The other benefit of what that’s done for us, for that same support team, they’ve been able to take on entirely new areas of support without adding a single person,” Vidoni said. “So, they’re becoming much more efficient,”

As more organizations embrace AI and automation in their workflows, there are several things to keep in mind. No matter what your stage of deployment, shadow AI is always, well…lurking in the shadows. Only 12% of surveyed IT professionals claim to be “very confident” that employees using AI and automation in their company are aware of the data security policies safeguarding the tech—a scary statistic when you consider the multitude of cybersecurity threats out there.

Some are already attempting to put a lid on unauthorized use of AI tools. Arriada said his company keeps a list of approved applications, and monitors whenever an unauthorized piece of software is installed on a user device. Once detected, users are contacted to review or remove those applications.

“We review them on a daily basis and every time we get an alarm, we check if there is any laptop, any asset with that application installed,” Arriada said.

But some professionals caution that patrolling shadow AI is a balancing act. Vidoni, who said securing shadow AI is a huge focus for him at the moment, advises organizations to promote some level of experimentation with AI tools: “I think it’s healthy…provided it doesn’t put the company’s IP at risk, provided it’s not going to unleash some really nasty AI agent that goes and deletes things.”

Cybersecurity isn’t the only aspect of AI and automation that demands a reality check; ROI remains a concern, even for mature initiatives.

Levy knows how valuable it is for a sales team member to contact a website visitor within a minute and a half.

“We could show, statistically, if you get them on the phone within 90 seconds, it’s going to lead to a much better conversion rate,” Levy said. He recently created an agent that, seconds after a potential customer fills out a site form requesting a tool trial, sends sales team members company profiles and talking-point suggestions.

“It usually took people six or seven minutes to do basic research,” he said, “and this is much better than basic research.” He added that the implementation has increased meeting and conversion rates for the ZoomInfo sales team.

That kind of time-saving, revenue-generating move can make a CIO’s day.

In its November 2025 report, Constellation Research concluded that today’s boards and CFOs expect proofs of value over PoCs when it comes to AI and automation, “with measurable improvements in the speed, accuracy, and effectiveness of the decisions that run the business.”

The intelligence firm defined this kind of “decision velocity” as increased speed and effectiveness in sensing signals and taking action in a way that drives outcomes.

When looking for the most decisive places in your business ops, Michael Ni, VP and principal analyst at Constellation Research, recommended searching for repetitive, high-volume decisions that feature a big payoff if there’s increased responsiveness (for example: getting a loan back in minutes, not hours.)

Ni also advised companies to start with a “small, atomic, operational” decision that has measurable impact, add automation to existing workflows, and deploy pilots that build internal expertise and buy-in.

Although 52% of IT Brew’s respondents said it was too early to tell if their AI and automation initiatives had delivered sufficient ROI, IT pros throughout the industry are already demonstrating that, with the right planning and resources, AI and automation can effectively handle crucial parts of the tech stack.