What Snowflake Actually Announced
Snowflake’s vision of the “agentic enterprise” has four moving parts: the data, a choice of AI models, the business applications, and a coordination layer to connect them. The two agents most employees will actually touch are CoCo and CoWork.
CoWork is for asking. Business users pose questions to company data and act on the answers—drafting emails, updating Salesforce tickets, posting to Slack—without writing a line of code or routing requests through a data team. It connects to Google Drive, Salesforce, and Slack out of the box.
CoCo is for building. Describe a tool or a migration in plain language, and the agent produces it. For companies struggling to hire technical talent fast enough, that’s a meaningful unlock. Migrations that used to take months are now taking weeks.
Both agents are designed to reach the people who always needed answers from company data but had to wait on someone else to pull them.
The Speed Is Real—With One Condition
Marco Cheng-Perri, manager of digital transformation at King Township (a municipality of nearly 30,000 north of Toronto), described the shift bluntly: weeks of work compressed into 10 seconds. Frontline staff can now ask whether rising recycling app usage is changing actual recycling behavior—and follow up in real time.
That used to route through his team. Now it doesn’t.
But King Township got there by first consolidating its data out of dozens of separate systems into one governed platform. The speed came after the foundation was built, not instead of it.
Thomson Reuters told the same story from a larger stage. Head of data and analytics Caitlin Halferty credited Snowflake with providing the governed, curated data layer that underpins the company’s AI products—pulling together Reuters news, financial, marketing, and HR data, structured and unstructured, with access controls in place.
The pattern is consistent: agents accelerate work that a clean data foundation already makes possible. They don’t substitute for it.
Model Flexibility Is Genuine—Switching Costs Are Not Gone
Snowflake CEO Sridhar Ramaswamy made a pointed argument at Summit: customers aren’t locked to a single AI model. They can run Anthropic, OpenAI, Google, Meta, Mistral, or DeepSeek inside the platform and switch between them as costs and capabilities shift.
“We offer you cloud independence,” Ramaswamy said. “We approach the model providers in a similar way.”
The data layer reinforces this. Snowflake stores data in Apache Iceberg, an open format any tool can read. Your data isn’t trapped inside a proprietary system only Snowflake can open. If you want to leave, the files can come with you.
That’s a meaningful distinction—and a real one. But it’s only part of the story.
What Open Format Doesn’t Cover
The data is portable. The work built around the data is not.
Every company using a platform at this level builds rules, pipelines, and governance policies to make its data usable. Snowflake’s Horizon Catalog manages that layer—applying one set of access policies across every tool that touches the data. That work doesn’t export with the Iceberg files.
Switching platforms is never free, on Snowflake or anywhere else. What the open format changes is whether moving is possible at all. That’s a lower bar than “easy,” but it’s a higher bar than most proprietary platforms offer.
The Adoption Gap Is the Real Story
Only 12% of Canadian companies used AI to produce goods or services in the year to mid-2025. The federal government’s national AI strategy, released the same week as Snowflake Summit, targets 60% by 2034.
That’s an ambitious gap to close—and Snowflake’s announcements quietly illustrate why it’s hard.
Shannon Katschilo, Snowflake’s country manager for Canada, put it plainly: “There is no AI strategy without a data strategy.” She’s watched Canadian companies shift over the past year from talking about AI to actually using it. The ones doing it well had already done the unglamorous work of consolidating and governing their data.
The ones who hadn’t are still waiting for the magic to start.
What This Means If You’re Evaluating AI Tools Right Now
The Snowflake story is a useful lens for evaluating any enterprise AI platform in 2026. A few things worth keeping in mind:
- Data readiness is the real prerequisite. Agents, copilots, and AI-powered analytics all assume a clean, consolidated, governed data layer underneath. If that doesn’t exist yet, the tool category doesn’t matter much.
- Model flexibility is now table stakes. Any serious enterprise platform should let you swap models without rebuilding your stack. If a vendor’s pitch depends on their proprietary model being the only option, that’s a flag worth noting.
- Open formats reduce lock-in risk—but don’t eliminate switching costs. Apache Iceberg portability is genuinely valuable. The governance and pipeline work built on top of it is still yours to rebuild if you move.
- Speed claims are real, but conditional. “Weeks to seconds” is achievable. It just requires months of data work first.
The honest takeaway from Snowflake Summit isn’t that AI agents are overhyped. It’s that they’re correctly hyped—for the organizations that have already done the boring, essential work of getting their data in order.
For everyone else, the agents are waiting. The data foundation isn’t.
Comments (0) No comments yet
Want to join this discussion? Login or Register.
No comments yet. Be the first to share your thoughts!