The pace of innovation in artificial intelligence (AI) has been relentless. With new generative and agentic AI models launching almost monthly, big tech is placing bold bets on AI as the next industrial revolution. But in 2025, the industry’s most meaningful transformations aren’t coming from sci-fi visions or grandiose promises; they’re being driven by something more grounded.
At NTT’s Upgrade 2025 event in San Francisco earlier this month, one message stood out: the most meaningful advances in AI this year aren’t about pursuing artificial general intelligence or replacing humans—they’re about streamlining enterprise workflows and boosting productivity.
“The real value of AI right now isn’t in replacing humans—it’s in supporting them,” said Naveen Rao, vice president of AI at Databricks. “That’s what we’re (Databricks) building for: workflows that amplify productivity and accuracy, especially in fields like software development. Not automation for automation’s sake.”
While the AI landscape continues to buzz with excitement around agentic models and multimodal systems, Naveen offered a measured outlook: AI agents are not autonomous masterminds. “They’re mere intelligent retrieval systems,” he said. “They pull from various formats, can generate code, like SQL, and streamline repetitive tasks. That’s the frontier—not sentient machines.” Specificity, he added, is what unlocks real power. “You can’t ask a model vague questions and expect genius. The sharper your query, the sharper the result. The open web has scale, but context and nuance still matter.”
Naveen believes in what he calls the “two-year rule”: it’s important for developers to only build and for enterprises to implement what’s technically feasible and commercially valuable within 24 months. “If you miss that window, you’re designing for a future that may never come,” he said.
From Moonshots to Measured Execution
Sridhar Ramaswamy, CEO of Snowflake, echoed Naveen’s pragmatic tone. Rather than focusing on artificial general intelligence, his team is prioritizing scalable systems that deliver real business value now.
“It’s about picking the right problems and helping solve them through innovations in technology. At Snowflake, we’re focused on being world-class in inference because we believe it’s a key business differentiator,” said Ramaswamy.
Story continues below this ad
He highlighted Snowflake’s investment in AI tools that automate data migration from legacy systems and support integration. “We’re also investing heavily in AI tools that help migrate data from on-prem systems to Snowflake, and we’re using AI to automate testing and integration. In many ways, AI is redrawing the line between product and service—and that shift is just getting started,” added Ramaswamy.
For Ramaswamy, usability is mission-critical. “We operate in a fast-moving environment,” he said. “Simplicity is our strategy. If users can’t get started in a couple of hours and scale across their organization, it won’t work.” He emphasized that most of a company’s valuable data lives inside its firewall—meetings, documents, product feedback.
“As data grows, you need applications that deeply understand the context. Whether it’s a chatbot to answer questions or a tool to search internal documents, AI is becoming essential to analyzing both quantitative and unstructured data within enterprises,” he said. “That’s where AI innovation is moving fast—toward domain-specific tools that answer tough, internal questions. What was our daily revenue? How did our product line perform? AI innovations and implementations should be focused more towards getting those answers in an instant.”
The multi-billion-dollar UI opportunity
While the spotlight often falls on chips and model parameters, Naveen sees the user interface layer as the true goldmine. “If agents are going to be useful, we need to reinvent how people interact with them,” he said. “The UI is where billions will be made.”
This is especially true in highly contextual industries like law, finance, and logistics, where generic AI tools often fall short. Snowflake, for instance, is embedding agents directly into enterprise workflows—quietly performing high-value tasks in the background.
Story continues below this ad
“Think about accessing a manual on a noisy factory floor,” Ramaswamy said. “You’re not pulling up a laptop. You need fast, relevant answers, surfaced at the right time. That’s the vision for agents—low-friction, high-impact tools that disappear into the flow of work.”
Still, fundamental challenges remain. “Hardware is still catching up,” Rao said. “Inference costs, edge computing, and specialised chips will determine who wins the next phase.” Companies like Japanese tech giant NTT are already investing in data center alternatives like photonics, which could dramatically accelerate AI performance.
In 2025, the future belongs to those who can make intelligence practical. The companies that win won’t be the ones chasing sci-fi fantasies—they’ll be the ones solving real problems, at scale, with precision.
