After steady progress in 2025, 2026 will be the year that agentic AI really takes hold in the enterprise.

As 2025 began, the industry expected an explosive, overnight arrival of agentic AI. The progress has been remarkable, and it’s still accelerating, but the real story is how the year reshaped our understanding of what’s possible. Organizations moved beyond simple chatbot use cases and began experimenting with systems that can plan, execute and iterate. Core agent capabilities grew stronger, enabling more complex, multistep tasks that were out of reach even a year ago. And with the market expanding quickly, investment and innovation continue to compound.

I joined more than a dozen Snowflake leaders to put together our annual Snowflake Data + AI Predictions report, where we share our perspectives on the year ahead, and the overall theme is that agents will make headway in the enterprise. Here’s a sampling of the predictions in this year’s report:

• Context windows and memory will be the keys to better AI agents: Key improvements to context windows and memory in the year ahead will allow agents to employ a bigger-picture understanding in order to handle complex challenges more autonomously. “It’s a more humanlike capability, to be able to remember the larger context of a situation to solve the problem at hand,” says Vivek Raghunathan, Snowflake’s SVP of Engineering and Support. 

• Workers will have to master human-AI collaboration and communication: Humans will remain in the loop, partly because not all the data that drives a decision is necessarily available to AI. Snowflake’s VP of Product, Chris Child, notes that AI can go deep on the data it has, but gut instinct still plays a part. “AI models will have a deep understanding of your data,” he says. “But you’ll still have to know when to doubt, when to ask deep follow-up questions before taking action.”

• Data strategy will determine AI readiness — and AI outcomes: “When AI delivers an accurate answer, you also have to be sure that private or proprietary data isn’t being exposed,” says Snowflake’s CIO, Mike Blandina. “Should the user have the permissions to see this answer? Is your marketing chatbot giving out employees’ Social Security numbers and customers’ credit card numbers? That’s not about the AI, that’s about how you govern and secure your data.”

By the end of 2026, the central question won’t be what AI can do; it will be how people and AI work together. In other words, how roles evolve, how decisions are shared and how leaders build trust and clarity in an environment where autonomy increases.

A decade ago a chief data and analytics officer’s (CDO) role was largely centered on data hygiene. But with the arrival of agentic AI, the role now expands into orchestrating how AI functions across the enterprise. CDOs are responsible for the quality and governance of the data that agents rely on, design the workflows agents integrate into and take responsibility for how those systems perform in the real world. This brings the CDO closer to a true AI COO, spanning engineering, governance, security, operations and product teams, ensuring that the AI operating model is stable, trusted and aligned to business objectives. 

In 2026, the challenge won’t simply be getting agents into production. Leaders will need to build the discipline around them. That means establishing verification frameworks, defining where human oversight begins and ends and maintaining observability so every agent action can be audited, explained and trusted. This will give rise to a formal AI quality control function, responsible for continuous monitoring and evaluation to keep agent behavior aligned with business intent. It’s the natural next step for enterprises that take reliability seriously.

This level of oversight depends on strong, centralized data foundations and governance. The federated models that worked during early experimentation created speed, but agentic systems require consistency: shared semantics, unified permissions and safeguards that hold firm even as agents scale across workflows.

As organizations re-architect processes and decision rights, enterprise-wide feedback loops become essential. They help teams refine guardrails, improve model behavior and ensure accountability is never ambiguous. In the near term, agentic systems will be best suited for structured, lower-risk workflows where boundaries are clear. As data maturity, governance and organizational readiness grow, agents will advance into more complex decision paths with greater autonomy and more strategic impact.

Agentic AI won’t eliminate work. It will rewrite it, opening new possibilities of opportunity and scale. For more on the year ahead, read Snowflake Data + AI Predictions 2026.