Snowflake Summit 2026: Summary of New Features

Snowflake Summit 2026 just wrapped up, and Snowflake emphasized that organizations are entering an era where AI agents will continuously operate across business processes, helping employees make decisions and execute work. A future where work increasingly involves collaborating with intelligent agents rather than simply using software tools.

Here is the summary of the main points and new features announced. You can view here too: https://www.youtube.com/watch?v=NFTdHYp9uJo

There have been many innovations in the four pillars: Interoperability (Iceberg, Polaris, OSI), Collaboration, Zero Copy, and Heterogeneous Access.

• Data is the foundation of AI: Enterprise AI success depends on trusted, governed, and accessible data. AI models are becoming commoditized, but enterprise data remains the differentiator.
• Enterprise trust and governance: Security, governance, lineage, and data quality remain critical as AI adoption scales.
• From experimentation to deployment: The focus is no longer proof-of-concepts. Customers are now deploying AI into production and seeking measurable business outcomes.
• Agentic AI: AI agents will not just answer questions but also perform tasks and workflows autonomously. Snowflake wants to be the platform where those agents securely access enterprise data.

Snowflake helps you operationalize AI by connecting models directly to governed enterprise data, workflows, and business systems. The differentiation is not just access to AI models. It is the ability to bring AI into the real enterprise with security, compliance, interoperability, and trusted context already built in. You will need a governed enterprise context, interoperability across systems, and flexibility as models continue to evolve. Snowflake gives you the ability to securely deploy AI where business users and workflows already exist, without needing to dedicate a team to build and maintain the stack.

You will need access to the latest frontier and open source models alongside governed business context in order to deliver trusted, actionable results. Snowflake’s posture is simple: use the best models, but run enterprise agents on Snowflake’s governed context.

Snowflake powers every data and AI workload

Snowflake gives you model and cloud flexibility with Day 0 access to leading models like OpenAI, Anthropic, and SpaceX AI models across major cloud providers. At the same time, Snowflake provides the governed enterprise data foundation that model providers do not. Governance, permissions, telemetry, workflows, policies, and spend controls are unified directly in Snowflake so AI can securely reason over enterprise data without introducing compliance or security tradeoffs.

Connecting a model to data through MCP is not good enough. That creates fragmented context, weaker governance, higher token usage, and lower quality outputs. Snowflake CoWork and Cortex Agents are grounded in an enterprise context, which drives higher accuracy, stronger governance, and better efficiency.

Prompt-Driven Data Management

Instead of traditional SQL-centric administration, users can increasingly interact with data platforms through natural language and AI-driven interfaces. This includes data discovery, governance tasks, data engineering workflows, and platform administration.

The Agentic AI Platform is not a single feature; it is a collection of capabilities that enable organizations to build, deploy, and govern AI agents on Snowflake. It consists of Snowflake CoWork, Cortex Agents, Cortex Search, Cortex Analyst, Cortex AISQL, Semantic Models, Openflow, and Horizon governance working together to create enterprise AI agents that can understand data, reason over it, and take actions securely.

Snowflake CoWork (formerly Snowflake Intelligence) is now a personal work agent for every knowledge worker, helping teams move from reactive analysis to proactive execution with AI-powered automation, reasoning, and enterprise context. It goes beyond answering questions by proactively surfacing insights, automating workflows, and generating polished deliverables like PDFs and PowerPoint presentations, all within the tools employees already use. It routes requests automatically across the right tools and workflows — no need to choose or manage individual agents. It learns your preferences over time with persistent memory, and personal skills let you capture workflows once and reuse them forever. MCP connectors now GA: Slack, Jira, Gmail, Salesforce, and more. It is also available in an iOS App for iPhone. Slack and Android are coming next.

Artifacts: Save and share live analyses, dashboards, and conversations, governed by role-based access controls. The future of BI is conversational.

Deep Research: Investigate complex questions across your enterprise data with agent swarm orchestration. Outperforms single-agent systems by over a third.

Snowflake CoCo (formerly Cortex Code) is the AI Coding Agent that allows builders to build faster. CoCo focused on one principle: your coding agent should work wherever you build. Builders can now use CoCo in the new native desktop app in addition to Snowsight, terminal, and IDEs. They can also embed CoCo’s expertise into the tools and agents they already use via MCP and SDK.

CLI developers: The Secured Local Sandbox (Private Preview) gives you VM-based isolation that keeps agent execution completely separate from your file system and OS. Full agentic power. Zero local exposure.

CoCo Automations: Set CoCo to run on a schedule — no manual trigger needed. Time-based workflows that work while you sleep.

Shareable Skills and a Skills Catalog are coming to CoCo and Snowflake CoWork. Browse, install, and publish reusable Skills across your org. Type /skill-catalog to explore. Press p to publish your own. Stop rebuilding from scratch.

CoCo Desktop: Take Snowflake’s coding agent anywhere, no browser required. CoCo is everywhere: Claude Code, Excel extension, VS Code extension, and mobile app. Your workflow, your choice.

CoCo Cloud Agent API: A single REST endpoint lets CI/CD, Slack, cron, and internal apps kick off CoCo without anyone sitting in a terminal. Basically, trigger CoCo from anywhere.

Cortex Sense: It is an enterprise memory that automatically learns how your organization defines its business — built from query history, metadata, and dashboards. No authoring. No manual setup. Result: 83% accuracy improvement vs. 24% without it. Agents get better, automatically. Context is the king.

Adaptive Compute is a workload-aware engine that automatically sizes, scales, and optimizes compute for every query. No more configuring warehouse size, multi-cluster settings, or auto-suspend. Customers set a performance goal, Snowflake handles the rest (with better performance than Gen 2 at similar cost)!

Enhanced Open Architecture and Interoperability

Snowflake reinforced its commitment to open ecosystems and interoperability. You can have open table formats like Iceberg and Delta, Cross-platform data access, Multi-model AI support, Open architectures for enterprise AI workloads.

Snowflake Lakehouse: Limited interoperability is an architectural tax compounding with every AI initiative. Teams move and copy data just to access it — spiking costs, slowing decisions, and starving AI agents of the governed context they need. Snowflake Interoperable Lakehouse replaces scattered copies with one governed data foundation, accelerating AI readiness and time to value. Built on Apache Iceberg™ and Apache Polaris™, every engine gets full read/write access with no duplication, migration, or split governance.

Snowflake now brings the broadest set of Apache Iceberg™ v3 capabilities to production: VARIANT for semi-structured data, Row lineage for cross-engine change tracking, Deletion vectors for faster row-level deletes, Nanosecond timestamps for financial & telemetry workloads, Geospatial types, and more.

Open Data Sharing: Any IRC-compatible external engine can now consume Snowflake data shares without a Snowflake account. Share the Delta and Iceberg data. Governance and lineage preserved. No data movement. No egress fees.

Delta Share support: You can also consume Delta Share in Snowflake.

Auto-gen Agents for Data Shares and Listings (Public Preview) instantly generates a Semantic View and Agent from any data listing or share. No manual development. Consumers get a governed, conversational experience, out of the box.

Horizon Context solves this with a governed semantic layer in Snowflake Horizon Catalog, giving every tool, team, and AI agent the same trusted business definitions — no bypassing, no reconciliation meetings, no more “depends which dashboard you look at.” Ask, “What drove last quarter’s customer churn?” and every team gives a different answer based on the dashboard they used. Definitions drift, agents improvise, and trust disappears. At a high level, it does three things:

• Collects context from schemas, query history, lineage, dbt, BI tools, and other connected sources.
• Enriches that into governed semantics like descriptions, popularity, lineage, tags, and Semantic Views.
• Activates that context automatically at query time so CoCo, CoWork/Intelligence, BI tools, and agents all work from the same definitions.

AI Security suite — 96% of orgs are stuck getting AI from pilot to production, and security is the #1 reason why. Snowflake’s new AI Security suite is built into the platform: agent identity controls, data exfiltration prevention, prompt injection defense, and multi-party approval. Not bolted on. Native, enterprise-grade, defense-in-depth — so customers can actually ship AI to production.

OpenFlow is expanding everywhere. The Data Connectivity Proxy now supports all three clouds — including GCP — and adds 5 new connectors: VeevaSystems, MongoDB, MariaDB, Shopify, and Google BigQuery. They are all guided setups in Snowsight, better observability, and lower costs. It now also provides wizard driven connector installation.

Snowflake Datastream is Snowflake’s native, Kafka-compatible streaming service built into Snowflake for real-time data pipelines. It lets you use the Kafka model/APIs you already know without running a separate Kafka infrastructure. One platform for streaming + analytics + AI, so data lands in Snowflake and is immediately available under Snowflake governance instead of moving through multiple systems. Basically, it collapses the boundary between data in motion and data at rest, with topic/table duality and governed tables, including Iceberg support.

Metadata Hub: A Metadata Hub aggregates and federates metadata across your entire data estate into a single, queryable interface. Rather than forcing data migration, it brings the understanding of that data together, giving every user a unified window into structure, semantics, lineage, quality and ownership no matter which system holds the data.

Snowflake Horizon Catalog is the connective layer, a single plane of visibility and control that sits above your existing catalogs and makes them more useful.

Postgres for Data Mirroring: It seamlessly mirrors data from Postgres into Snowflake. Replicate update with low latency (30 sec) across your OLTP and OLAP. Once you create a mirror, Snowflake maintains target tables that reflect the current state of their source tables, including schema changes and changes to the set of mirrored tables. Also, there is plan to Snowflake into Postgres.

Observe: With the acquisition of Observe, you can do Logs. APM. Infrastructure monitoring. One platform, without the cost or scale limits of legacy tools. Observe gives Snowflake a first-party, AI-powered observability platform for logs, metrics, and traces — with correlation, faster root-cause analysis, and large-scale telemetry retention — while still keeping the broader observability ecosystem open. Observe is a full standalone enterprise-wide observability platform, and comparable to Datadog, Splunk, or Elasticsearch, whereas Snowflake Trail is the built-in observability offering for Snowflake artifacts.

Natoma: Snowflake has **announced its intent to acquire Natoma, the enterprise Model Context Protocol (MCP) platform for AI agents. Natoma provides a centralized MCP gateway and verified library of MCP servers that securely connect AI agents to enterprise applications, databases, APIs, and collaboration tools. Natoma enables secure connectivity between Cortex AI Agents, Snowflake CoWork, CoCo, and other AI platforms. Natoma extends governance from data assets to AI-driven actions and workflows, and finally, Natoma helps customers operationalize AI agents securely at enterprise scale.**

Multi-Party Collaboration in Snowflake is a Data Clean Rooms capability that lets more than two parties collaborate in a fully symmetric, privacy-preserving environment instead of the older fixed provider/consumer two-party model. Multi-party collaboration holistically unlocks industry use cases by weaving signals together across ecosystem partners — while keeping their data private with full governance and control. For example, An advertiser, publisher, and identity partner can all join the same clean room to analyze campaign performance together. One party may own the collaboration, others contribute data, and only selected parties run the analysis.

Snowflake and Anthropic are expanding work across Snowflake Cortex AI — with Claude powering Snowflake CoCo and CoWork — as well as Claude Marketplace. As more organizations use Anthropic Claude models directly inside Snowflake, demand continues to grow for governed, production-ready AI that delivers the security, observability, and scale enterprises require. Together, Snowflake and Anthropic are helping accelerate enterprise AI adoption and build the foundation for the agentic enterprise.

The complet Snowflake AI Data Cloud

Summary

Snowflake is the trusted enterprise platform where governed data powers AI agents that can automate and augment business operations at scale. Go try out Snowflake AI, it is pretty easy and powerful!

Thanks for reading. Hope it is helpful!

Disclaimer: The opinions expressed in this post are my own and not necessarily those of my employer (Snowflake).