As thousands of engineers, founders and researchers gather in San Francisco for AI Engineer World’s Fair, much of the conversation is focused on increasingly capable models, autonomous agents and AI applications.

According to Vytautas Savickas, CEO of Oxylabs, however, the industry’s biggest shift is happening somewhere else.

“For the past three years, AI has largely been a story about better models,” he says. “The next chapter is about everything around those models: the infrastructure, the systems and the live information that allow AI to operate in the real world.”

That changes what AI systems need.

They no longer rely only on what they learned during training. Increasingly, they depend on fresh information, live search, browser interaction and reliable access to the constantly changing web.

“Knowledge isn’t static,” Mr. Savickas says. “A model that isn’t connected to fresh information already knows less about today’s world than most people realise. The closer AI gets to real-world decisions, the more important it becomes to stay connected to reality.”

AI is becoming an infrastructure problem

According to Mr. Savickas, every major breakthrough in AI has fundamentally changed the infrastructure beneath it.

“The first wave was about training foundation models. Everyone needed enormous amounts of diverse public data.

Then came retrieval-augmented generation (RAG). Suddenly it wasn’t enough to know what the world looked like yesterday. AI had to understand what changed five minutes ago.

Now we’re entering the agentic era. AI systems are beginning to search, compare, verify, purchase, monitor and complete tasks on behalf of users.”

Every wave, he argues, demands a different kind of infrastructure.

“People often think AI is primarily a model problem. Increasingly, it’s becoming an infrastructure problem.”

Public web data is no longer only used for training. It is becoming part of AI’s runtime.

“The web was built for billions of people. We’re now asking it to support billions of AI-driven interactions. That fundamentally changes what infrastructure has to deliver.”

Building for AI before AI

Long before foundation models became mainstream, Oxylabs was building infrastructure that enables enterprises to reliably access and operationalise information from the public web at scale.

“It may look like AI created this market overnight,” Mr. Savickas says. “In reality, AI didn’t create this challenge. It exposed it at an entirely new scale.”

For years, enterprises have depended on continuously changing information to power ecommerce, cybersecurity, travel, finance and market intelligence. AI has simply expanded those requirements to almost every industry.

Today, Oxylabs serves more than 15,000 clients worldwide, holds over 160 patents and operates one of the world’s largest public web data infrastructures.

Intelligence alone won’t determine who wins AI

Much of today’s AI conversation still revolves around model performance.

Mr. Savickas believes that discussion is beginning to shift.

“Frontier models will continue to improve, but for many real-world applications, model quality alone is no longer the differentiator. Increasingly, what matters is how reliably AI systems connect to the outside world.”

That shift also changes how he thinks about hallucinations.

“The model isn’t necessarily making things up because it’s unintelligent. Often it’s trying to reason using stale, incomplete or unverifiable information.”

Visitors arriving at AI Engineer World’s Fair may notice one of Oxylabs’ messages across San Francisco: Models Hallucinate. Fresh Data Doesn’t.

“It’s intentionally simple,” Mr. Savickas says. “But it captures something fundamental. AI doesn’t only need reasoning. It needs reality.”

He believes this will ultimately redefine competition across the AI industry.

“The companies that win AI won’t necessarily build the biggest models. They’ll build the systems users trust the most.”

AI agents change everything

The rise of AI agents is also changing how engineers think about infrastructure.

Reasoning is only one part of the problem.

AI systems increasingly need to navigate websites, authenticate, verify information, compare alternatives and execute actions reliably.

“Today everyone talks about building AI agents,” Mr. Savickas says. “Tomorrow everyone will ask why those agents fail.”

“In many cases, the answer won’t be the model. It’ll be the infrastructure connecting that model to the real world.”

According to him, this is where much of AI innovation is now happening.

“Latency matters. Reliability matters. Browser automation matters. None of those things make headlines, but all of them determine whether AI actually works.”

More hype, same engineering realities

The rapid growth of AI has introduced a new generation of companies building tools around browser automation, agent frameworks and web access.

Mr. Savickas welcomes the momentum.

“It’s exciting to see developers recognising how important this layer of technology has become.”

At the same time, he believes AI has changed the language around long-standing engineering challenges more than the challenges themselves.

“Every technology wave creates new terminology. The engineering problems are often more familiar than they appear.”

“What matters isn’t the label. What matters is whether your infrastructure works reliably when thousands of AI systems depend on it.”

Keeping the web open

As AI systems become increasingly dependent on publicly available information, debates around access to the web continue to grow.

For Mr. Savickas, maintaining an accessible open web is essential not only for AI companies, but for innovation itself.

“The open web remains humanity’s greatest shared knowledge resource because information can be discovered, connected and built upon.”

“If information is intentionally made public, it should remain accessible. That’s how researchers innovate, businesses compete and new technologies emerge.”

Looking beyond today’s AI boom

Having spent nearly a decade building infrastructure for enterprises working with the public web, Oxylabs has already seen several technology waves reshape the industry, from travel aggregators and digital marketing to ecommerce, cybersecurity and now AI.

AI may be the biggest wave yet, but Mr. Savickas believes it won’t be the last.

“The first generation of AI proved that machines could reason.”

“The next challenge is making those systems operate reliably in the real world.”

“AI won’t be powered only by better models. It’ll be powered by better infrastructure around those models.”