AI is only as smart as the context behind it

Jason Aloia has spent his career at the intersection of product and service operations, so he's watched the same fundamental pressures cycle through IT for years. Do more with less. Manage more complex technology. Improve employee experience without adding headcount. What's changed, he argues, is that AI is proving capable of delivering on promises that earlier generations of the technology couldn’t. 

As VP of product management at Freshworks, Aloia spends a lot of time on what service transformation requires at the operational level. The answer, he says, starts with a layer most organizations haven't built yet.

Read also: Srini Raghavan on why most AI projects stall before they scale

The pressures on IT have changed, but they haven’t diminished. What’s different now? 

In many ways, the fundamentals are the same. IT organizations are still managing increasingly complex technology ecosystems, still expected to do more with fewer resources, still on the hook for employee experience. What AI is doing is amplifying all of that simultaneously, which creates enormous pressure to figure out how to pull it off.

What is genuinely new is that AI is proving capable of things that earlier generations of machine learning and predictive analytics promised but couldn't deliver. With generative AI, improvements to employee experience that used to be aspirational are within reach.

But the most important shift is harder to quantify. This is an entirely new way of working, and organizations are still figuring out what that means operationally. Even at Freshworks, we're working through it. And I think that creates a real survival filter. Organizations that figure out how to operate in this world will pull ahead fast. Those that fall behind are at genuine risk of losing relevance quickly.

You mentioned that organizations need the right conditions for AI to actually work. What does that look like in practice?

The pattern I see most often in organizations that are struggling is that they don't have a clear picture of what they're trying to build, what success looks like, or the commitment to keep experimenting and optimizing. That's a challenge different from what most software implementations have asked of IT teams.

For a long time, the deal was: buy the software, configure it, get the outcome you were promised. Everything was deterministic. AI requires a different deal. Good inputs—accurate knowledge bases, clear policies, documented standard operating procedures. Many organizations try to deploy AI and wonder why it's not working, when the real problem is upstream. They don't have the right inputs in place yet. And until they do, no amount of AI capability will close that gap.

That brings up the build-versus-buy question. With AI more accessible than ever, how are service leaders thinking about that decision?

There's a real temptation to build your own. It's a pattern we've seen play out before. Think about when it became easy to build your own website. Organizations suddenly had a Wix account and things looked fine, until they watched their SEO rankings quietly disappear and had no idea what happened. What's coming with AI is that dynamic, but exponentially more consequential.

Now, some of our customers are genuinely sophisticated enough to navigate the build side. They understand what it involves, and for them, the conversation is more nuanced. We still see a real role for Freshworks in those ecosystems, particularly as the world gets more open. We're not under any illusion that we'll be the only AI in customers’ environments. But most organizations have a core job to do, and IT service management is a dependency, not their specialty. That's where a purpose-built platform still matters.

AI agents are moving fast. What are organizations getting wrong as they bring them in?

A lot of it comes down to governance—and it’s often the Wild West right now. CIOs are finding out after the fact that someone has an agent running against data it shouldn't have access to, or that the permissions model was never thought through.

The problem is subtle. An agent built with my credentials might technically have access to certain data. But if I let someone else query that agent, is it smart enough to know that they don't have the same access I do? A lot of agents aren't handling that correctly.

Read also: The striking gap between how much control IT leaders think they have over unsanctioned AI—and how much they actually do

What we're working toward is a canvas for rapid innovation that has guardrails built in from the start—full visibility into what the AI is doing, what people are doing, and the chain of access and permissions at every step. Organizations should be able to audit any of that at any time. That's not a constraint on innovation. It's what makes innovation sustainable at scale.

What does getting this right look like for the people working inside service organizations?

At its best, AI is proactively keeping the entire system healthy so employees can just do their jobs. The IT organization running behind the scenes becomes invisible, not because it doesn't exist, but because it's working.

The role of the people inside IT shifts significantly in that world, but it doesn't shrink. Technology isn't slowing down. The volume of novel situations, such as edge cases and other scenarios that AI hasn't encountered, is going to keep growing. The human role becomes handling that novelty: triaging what's genuinely new, and getting very good at converting those novel situations into patterns that AI can recognize and handle the next time around.

But everything we've talked about depends on one foundational thing. Not just data, but context. A genuine, 360-degree understanding of the people, technology, and processes involved in the work. That includes infrastructure monitoring signals, device data, workflow history, resolution history. That’s the layer that everything depends on. 

Don't miss Jason and more experts at Refresh Virtual Summit 2026. Register today.