Snowflake Summit in 60: Montreal data community event recap

Last week, we hosted a Summit in 60 Special Edition at the Snowflake Montreal User Group. Over 60 practitioners from across the Montreal data community joined us for an evening of open discussion over food and drinks.

Together, we distilled the biggest announcements from Snowflake Summit 2026 in San Francisco into a practical recap for people who missed the event.

Snowflake is framing 2026 as the start of the "agentic AI era”, the same way the company framed 2016 around the Data Cloud. Nearly every announcement ties back to one thesis: Snowflake wants to be the governed, trusted foundation that agents, BI tools, and applications all operate from.

Roundup of key announcements from Snowflake Summit 2026

Here’s what stood out from this year’s announcements at Snowflake Summit:

AI and agentic capabilities

The two headline products are CoWork and CoCo. Snowflake Intelligence has been rebranded to Snowflake CoWork, positioned as an "AI chief of staff" that connects enterprise data, email, conversations, and apps to produce dashboards, reports, and presentations. CoWork now includes a personal agent with memory and skills, autonomous agents with scheduled automations, live shareable dashboards (called Artifacts), and an iOS app.

CoCo (Cortex Code) is expanding its footprint across the development experience: CLI, Snowsight, VS Code, a Claude Code plugin, and an upcoming CoCo Desktop IDE. Cloud Agents now run CoCo entirely inside Snowflake with zero installation, per-user isolation, and task persistence.

The open question for practitioners is whether CoWork becomes a tool teams adopt day to day or stays a demo-stage product. CoCo is more immediately practical because it meets developers where they already work.

Platform and data engineering

Snowflake made its biggest push yet into owning more of the data stack. The pattern across announcements is clear: close the gaps that used to require a separate tool for each step.

• Openflow and Datastream take aim at ingestion and streaming, with a DCP Agent that handles on-prem sources without VPNs
• Zero-copy integrations with Salesforce and SAP are now GA, with Workday in private preview
• Snowflake Postgres adds fully managed Postgres with bidirectional mirroring
• The dbt integration is getting deeper through the Fusion engine and Cortex Code support across the dbt lifecycle

Underneath all of it sits the Horizon Catalog, built on Apache Polaris. Semantic Views are now GA with composable definitions and SCD Type-2 support, Semantic Studio gives semantic models a proper IDE with Git-native versioning, and Metadata Connectors (Select Star technology, now native) crawl external systems. The Horizon Catalog is where Snowflake's agentic AI thesis is tested, because every agent and BI tool operating on the platform will rely on this layer for context, governance, and trust.

On the compute side, Adaptive Compute moves toward a shared serverless pool that removes manual warehouse sizing, with purpose-built engines optimized for different workloads.

Governance and security

Three additions worth noting:

• Intent-Driven Governance lets teams describe what to protect in plain English and Snowflake deploys the policies
• Agent Identity provides tracking and governance over what AI agents do on behalf of users
• Horizon AI Guardrails add runtime prompt injection detection, on by default, with no configuration needed

Agent Identity is especially relevant as teams start deploying agentic workflows. Knowing which agent did what, on whose behalf, is a governance gap most organizations have only just started thinking about.

What stood out to me

The Horizon Catalog and Semantic Views felt like the most significant shift for practitioners, even though CoWork and CoCo got more stage time.

Semantic Views going GA with composable definitions and SCD Type-2 support means teams can start consolidating metric definitions inside Snowflake rather than managing them in a separate semantic layer. For the data community, where a lot of teams run lean and juggle multiple tools for the same job, having a native, Git-versioned semantic layer changes the day-to-day of how data gets modeled and governed.

I also think the Adaptive Compute direction is worth watching. If Snowflake delivers on removing manual warehouse sizing, it solves one of the most common sources of cost surprises and operational headaches for running production workloads.

FAQs from the community

Does Adaptive Compute mean teams should stop tuning warehouse sizes now?

The short answer from the room was: keep current configurations in place. The adaptive model is directional, and existing warehouse settings still apply for current workloads. The practical takeaway was to avoid building elaborate custom auto-scaling logic that Snowflake may soon handle natively.

Should teams adopt Semantic Views now that they are GA, or wait?

The consensus leaned toward starting with a focused pilot. Semantic Views with composable definitions are production-ready, but migrating an entire semantic layer takes planning. A good starting point is a single domain where metric definitions are currently duplicated across tools.

What does Agent Identity mean for existing service accounts and automations?

This was one of the more practical governance discussions. As teams move from service accounts running scheduled queries to AI agents acting autonomously, the audit trail changes. Agent Identity gives a way to track agent actions separately from the human who authorized them, which matters for compliance and debugging alike.

How real is Openflow as a replacement for current ingestion stacks?

The room was cautiously optimistic. Openflow simplifies ingestion significantly, especially the DCP Agent for on-prem sources. For teams already running Fivetran or Airbyte, the question is whether consolidating into Snowflake-native ingestion is worth the migration effort. For new projects, it lowers the barrier considerably.

Join us next time

If you’d like to attend more events like this one, join us for our upcoming Montreal data developer meetup on July 16th, featuring expert panelists as they discuss the future of agentic AI in data engineering. Register today!