TL;DR

Vytautas Savickas, who leads Oxylabs, maintains that AI is transitioning from a focus on raw model capabilities toward dependable underlying systems. With AI shifting into an agent-focused phase, tools require up-to-date online content, automated browsing features and instant connectivity to function effectively outside training environments. Success in this space, he notes, will go to firms that deliver the most dependable platforms rather than those with the largest models alone.

Engineers, startup leaders and academics are converging in San Francisco for the AI Engineer World’s Fair, where discussions center on advanced models, self-directed agents and practical AI tools.

Yet Vytautas Savickas, CEO at Oxylabs, points out that the most significant industry evolution lies elsewhere.

“Over the last three years, AI progress has centered mainly on improving models,” he notes. “The coming phase involves the surrounding elements: foundational systems, supporting frameworks and current data streams that let AI engage with everyday environments.”

This evolution alters the requirements for AI platforms.

They no longer depend exclusively on pre-training knowledge. Instead, they increasingly draw on current details, active queries, site navigation and stable connections to the ever-updating internet.

“Information does not stay fixed,” Mr. Savickas explains. “Any model lacking ties to current details already falls short in understanding present conditions more than most appreciate. As AI moves toward practical choices, maintaining links to actual conditions grows essential.”

AI Transforms into an Infrastructure Challenge

Mr. Savickas observes that each major AI advance has reshaped the supporting systems required.

“Initial efforts targeted creation of base models, requiring vast volumes of varied public content.

Retrieval-augmented generation followed, making yesterday’s knowledge insufficient. AI now must track shifts from moments earlier.

We now approach the agent-driven stage, where AI begins to query, evaluate, confirm, buy, track and finish tasks for users.”

Each stage, he maintains, calls for distinct supporting structures.

“Many view AI chiefly as a modeling issue. It is turning more into a systems challenge.”

Public web content now serves runtime needs, not training alone.

“The internet was designed for human-scale use. We now require it to handle AI-scale activity, which alters the performance standards for all supporting layers.”

Preparing Systems Ahead of AI Adoption

Years before base models gained wide use, Oxylabs developed platforms allowing large organizations to access and apply public web information reliably at volume.

“It can seem as if AI sparked this demand suddenly,” Mr. Savickas states. “Actually, AI did not invent the issue. It simply revealed it at a much greater level.”

Enterprises have long relied on dynamic data for retail, security, tourism, finance and research. AI has broadened these needs across nearly all sectors.

Oxylabs now supports over 15,000 clients globally, maintains more than 160 patents and runs one of the largest public web data networks.

Raw Intelligence Will Not Decide AI Leaders

Current AI talks often highlight model benchmarks.

Mr. Savickas sees this emphasis moving.

“Leading models will keep advancing, yet for many practical uses, model strength by itself no longer sets firms apart. What counts more is the steadiness of AI links to external sources.”

This view also reframes issues like incorrect outputs.

“Models do not always err due to limited capability. Frequently they work from outdated, partial or unconfirmed details.”

Attendees at the AI Engineer World’s Fair may see Oxylabs signage in San Francisco reading: Models Hallucinate. Fresh Data Does Not.

“The message is straightforward on purpose,” Mr. Savickas adds. “Yet it highlights a core point. AI requires both logic and accurate context.”

He expects this to reshape industry competition.

“Firms that lead in AI will not always produce the largest models. They will create the platforms users rely on most.”

AI Agents Reshape Requirements

The growth of AI agents is altering how teams approach supporting systems.

Reasoning forms only part of the task.

AI tools must now browse sites, handle logins, check facts, weigh options and perform actions without failure.

“Everyone discusses creating AI agents today,” Mr. Savickas observes. “Soon the focus will turn to why those agents fall short.”

“In many cases, the cause will not trace to the model. It will trace to the systems linking that model to live conditions.”

This area, he says, now drives much AI development.

“Speed counts. Dependability counts. Site automation counts. These elements rarely dominate news, yet they decide if AI delivers results.”

Continued Hype Meets Familiar Engineering Needs

AI expansion has spurred new firms offering browser tools, agent platforms and web access solutions.

Mr. Savickas views the activity positively.

“It is encouraging to see developers grasp the importance of this technology tier.”

He also notes that AI has mainly updated terminology around established engineering tasks rather than changing the tasks.

“Each wave brings fresh terms. The underlying engineering issues often remain similar.”

“Labels matter less than whether systems perform steadily when many AI tools rely on them.”

Preserving Open Web Access

As AI grows more reliant on public content, discussions about web access intensify.

Mr. Savickas sees continued open access as vital for both AI progress and broader innovation.

“The open web stands as humanity’s largest shared knowledge base since content can be found, linked and extended.”

“When material is placed publicly, it should stay reachable. This supports researcher advances, business competition and emerging technologies.”

Beyond the Current AI Surge

After nearly ten years creating web data platforms for enterprises, Oxylabs has witnessed multiple shifts from travel platforms and online marketing to retail, security and now AI.

AI may prove the largest shift, yet Mr. Savickas expects more to follow.

“The opening AI phase showed machines can reason.”

“The following task is enabling those systems to function steadily in real settings.”

“AI progress will depend not only on stronger models. It will depend on stronger systems surrounding those models.”