As AI agents take on real work, new rules for autonomy are emerging that favor reliability, clarity and human control.
were little more than a lab experiment. Today, however, they’re starting to move beyond prototypes, navigating workflows, answering support tickets, surfacing insights and even acting on behalf of enterprise users.
A recentputs AI adoption and experimentation across business functions at nearly 80%, up from roughly 55% a year earlier. But as the world inches closer to operational autonomy, the big question across the enterprise is about what actually makes these agents safe enough to trust., the answer is not in the size of the model but in a kind of intelligence rooted in semantics, control and contextual reasoning. “We want to build intelligent agents that are grounded in a customer’s data and that perform the kinds of tasks real users want, not just toy tasks,” he told me in an interview this month. That ambition is starting to reflect a broader shift in how companies are beginning to think about enterprise AI and is especially crystallizing across Snowflake’s platform, as the company bets on a future where enterprise AI agents aren’t just powerful but also accountable, precise and productively constrained., the company’s new suite of agentic capabilities, marks a pivot from passive AI to active execution. This includes tools that can automatically generate SQL queries, recommend actions across dashboards, or assist in data classification — features meant to support real work inside organizations, rather than experimental prototypes or showcase demos. Enterprise AI has largely remained passive, with most deployments today stopping at summarization or on-demand data retrieval. Snowflake is pushing beyond that boundary by building agents that can reason through multi-step processes and take limited action, all within user-defined rules and permissions“Productivity is a huge theme,” Ramaswamy said. “If a customer support agent can close five more tickets per hour because an AI assistant is helping them, that’s a real improvement. We believe AI can help make users 10x more productive, but only if it operates with the right constraints.” To get there, Snowflake is investing in what Ramaswamy called a “semantically aware platform” that understands how data maps to business processes and keeps the model grounded in the reality of each organization’s operations.today is hallucination — when an AI model confidently invents facts or outputs that don’t match reality. For most companies, the stakes are too high to tolerate that kind of error. Snowflake’s answer is infrastructure-first: Reduce hallucinations not by trying to teach the model everything, but by limiting what it’s allowed to see, touch, and say. “Every single action an agent takes is constrained by policies,” Ramaswamy explained. “If you’re in a heavily regulated industry, you need to be able to say, ‘This user is allowed to do this kind of thing with this data, and only in these situations.’” That control is built directly into the platform itself, across roles, permissions, and data mesh structures. The goal is to make AI outputs both reproducible and reviewable, so teams can audit what an agent did and why. “We want our agentic systems to be deterministic,” he added. “That means if you ask the same question twice, you should get the same answer, especially in a business context.”available in the market. Rather than asking customers to trust the model, the company is building systems that behave more like reliable tools than autonomous decision-makers, remaining narrow in scope and accountable.Ramaswamy pushed back on that idea. “This is not about replacing people. It’s about making their work more effective,” he said. “We believe users should stay in the loop and have control. The agent is there to assist, not override.” Snowflake’s interface design reflects that belief. In Snowflake Cortex Analyst, for example, users can prompt an agent to generate charts or explain data. But before acting on the output, the user always reviews it. It’s augmentation, not automation for its own sake. Some of the company’s earliest agentic deployments are already showing this in action. For instance, Cisco utilizes Snowflake Intelligence to enhance onboarding processes and promote self-service among its engineering and product teams. At TS Imagine, agents help reduce manual overhead in financial modeling. Other customers, like Fanatics and Toyota Motor Europe, are exploring similar patterns, embedding agents into workflows where speed and accuracy matter but decisions still require human judgment. For Snowflake, these use cases are not just proof points but real signals that trust and transparency are winning factors in the agentic arms race.None of this works without the right foundation. Behind every AI agent, there’s a stack of decisions: How the data is structured, how the model is grounded, how policies are enforced, and how results are monitored. This is where the conversation about AI shifts from features to infrastructure, and where many enterprise deployments still struggle. For many enterprises, the question is no longer whether to use AI, but how to integrate it safely into mission-critical systems. That’s why Snowflake’s vision goes beyond agents. It includes Snowflake Cortex , Streamlit , and Document AI . Together, these tools create a pathway from raw data to safe execution, without sacrificing visibility or human oversight on the way. “We think about this as a continuum,” Ramaswamy said. “From summarizing data to generating SQL to building full workflows — each step requires more context, more safeguards, and better alignment with how people actually work.” That perspective is influencing how Snowflake frames its role in the market, shifting from a data platform to AI infrastructure built for enterprise use. The idea is simple: Autonomy should be introduced carefully, with systems that prove they can be trusted before they are given more responsibility. Snowflake’s emphasis on trust and control reflects a broader enterprise reality today. As companies scale AI, most are still in the early stages of adoption and governance, and only a fraction have pushed agentic systems into widespread use. That means the next phase of enterprise AI is likely to be as much about culture, policy and data readiness as it is about model capability.
Agentic AI AI Agents Enterprise AI AI Displacement AI Hallucinations Agentic Tools AI Infrastructure Data Readiness
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Ford killing F-150 EV pickup, warns of whopping $19.5B writedown in dramatic electric shiftToday's Business Headlines: 12/12/25
Read more »
How the Ukraine war accelerated a defence shift and why Türkiye is leading the wayAs warfare undergoes seismic changes, countries are being forced to make radical changes to keep up with the evolving battlefield needs.
Read more »
Three Shifts Redefining Enterprise AI In 2026Just 5% of organizations have realized meaningful ROI from their AI investments.
Read more »
Suspect in Enterprise murder investigation now in custodyThe deadly shooting happened Monday night in the 400 block of West Watts Street.
Read more »
Enterprise man charged with attempted murder after Elba shooting SaturdayLee Display is a rarity in the Christmas decor industry: It makes artificial trees in the US and avoids tariffs. Americans are paying more for Christmas trees this year because they're nearly all made overseas and subject to tariffs.
Read more »
MCP Security: Why Enterprise AI Needs A New Trust ModelIn the MCP era, there is no 'expected behavior' to deviate from. Every workflow is unique. Every sequence of tool calls is generated on the fly.
Read more »