Unless you’ve been inconspicuously swapped into an alternate dimension where the letter “A” doesn’t exist, you’ve been hearing a lot about agentic AI this year. And if you were at Snowflake Summit 26, you definitely heard a lot about it. Here’s the Estuary takeaway from last week’s conference, and what we heard from our conversations with data practitioners.
An invitation to rethink what is possible
Snowflake Summit 26 drew more than 20,000 attendees to San Francisco with a rally cry: The era of the agentic enterprise is here. But how many businesses are prepared to meet the moment? Snowflake CEO Sridhar Ramaswamy opened the keynote by addressing the elephant in the room: Data is still siloed and fragmented across dozens of systems, cloud providers, and legacy platforms.
The cost of data fragmentation is higher than it's ever been. AI workflows depend on clean, current, and trustworthy data, which creates urgency for companies to re-evaluate their governance and ingestion layers. It’s a foundational imperative for any team that’s serious about their AI goals. And Snowflake has a vision for an agentic control plane to coordinate this kind of work across all your organization’s data, no matter where it lives.
Data hygiene work may be unglamorous, but much like a weird rash, it can’t be ignored. We heard it directly from folks at our booth and in sessions: AI models and their level of sophistication (or lack thereof) aren’t the blockers; the data underneath them is. If that data is stale, messy, and sketchy, your AI initiative never even clears the runway for takeoff.
The quality of your data is now your biggest advantage in the proverbial AI arms race.
That statement isn’t necessarily novel. We all understand why no one wants dashboards showcasing days-old metrics, or customer service agents making decisions based on outdated policies. An AI model running on bad data is at best inconvenient and at worst damaging, especially if it impacts customers.
Ramaswamy ended the platform keynote the next day with an invitation to “rethink what is possible.” What is the best possible version of your product? What is the best possible version of your user experience? What has your team always wanted to build, but never had the resources to try? These are the types of questions that AI is solving, and at a breakneck pace.
AI tools for builders: from months to days
If we had to sum up the message of Day One: If you’re building something, you better be building it with AI. But we also heard from many data engineers throughout the week who didn’t have a clear picture of what AI could do to improve their day-to-day role. Discourse tends to focus on what AI is doing for the end user, or how it can influence revenue and improve internal operations.
There are boring and cumbersome elements of most jobs that I’m sure we’d love to automate. But do we have a clear vision of what work can fill the vacancy? For example, LOVESPACE’s data team finally had time to build and implement an AI forecasting model for warehouse usage + a real-time transaction tracking system after Estuary automated their pipeline maintenance tasks. Think of all the fun and useful data projects that engineers in companies around the world have left on the back burner because they just haven’t had time to work on them yet.
Snowflake's announcements for builders centered on Snowflake CoCo (formerly Cortex Code), an AI-powered coding agent designed to turn natural language into working pipelines, migrations, and applications. Migrations that used to take six months now take six days. Pipelines that required a specialist can now be kicked off with an idea.
Snowflake also announced Datastream, a Kafka-compatible streaming service built to bring Kafka topics natively into Snowflake as governed tables, currently headed toward private preview. The signal is clear: Snowflake is moving to consolidate more of the data stack under one roof.
This all points to something bigger than productivity gains. When the low-level work gets completed faster, the ceiling for what a data engineering team can accomplish moves up. Each person has more time to spend on architecture decisions, data quality, and the problems that actually require domain knowledge.
Agent Skills, Estuary's new feature that lets teams build, monitor, and debug pipelines directly from within AI coding tools like Claude Code, Cursor, and Copilot, is designed for exactly this environment as well.
What do "AI-powered workflows" look like for data engineers?
Snowflake’s vision of the agentic enterprise—agents running continuously, coordinating across data, models, and applications—puts a lot of weight on the infrastructure layer. Agents are only as good as the data they reason over. If that data is stale, incomplete, or inconsistent, the agent's output reflects that.
This is where the data engineer's role becomes more strategic, not less relevant. Someone still needs to own the pipeline architecture, define what clean data looks like, and ensure the right data is flowing to the right destinations at the right time. The grunt work may be getting automated, but the judgment work is more valuable than ever.
Snowflake's direction reinforces what Estuary has been building toward. Reliable, real-time data movement is foundational to every agentic workflow a business wants to run. From our perspective, the engineers who were most energized were the ones asking how to get ahead of the shift and move from managing pipelines reactively to designing data systems that AI workflows can actually depend on.
And no dependable data system can exist without strong governance and quality standards. The technology may look different, but the strategy behind it isn’t revolutionary. Successful teams will understand this and prioritize accordingly.
About the author
Dominick Duda is the Growth Marketing Manager for Estuary. Previously, he was a Market Research Analyst at G2 for six years, where he covered buyer and market trends for B2B software across vertical industries like healthcare and life sciences, government, nonprofit, agriculture, and hospitality. He has over a decade of research and professional writing experience.