Snowflake Summit 2026 — Strategic Report

We went in as a customer. We left as a reference.

Summit confirmed that RBNA's Wingman initiative is not just aligned with where Snowflake is heading — our work is actively shaping their product roadmap. Multiple Snowflake PMs described our team's approach as ahead of the curve. Our investment in semantic views, governed agent deployment, and rigorous validation puts us in a position that most of Snowflake's enterprise customers are still working toward.

"Red Bull's interest in charts has informed a lot of our roadmap on the charts — and in general, Snowflake's roadmap is informed from a lot of our conversations."

Effie Goenawan Principal Product Manager, Snowflake Intelligence & CoWork · Summit 2026

Summit at a glance

• SCALE~20,000 attendees, 350 organizations, 700 speakers across 4 days at Moscone Center, San Francisco.
• RB16 Red Bull attendees across 6 teams — IT, Finance, Sales Reporting, Data Science, Global Data, and Media House.
• PMs9 Snowflake PMs in one session — the largest PM engagement of the Summit, arranged by Thomas (Account Executive).
• MEMO9 of 12 features we requested were built by Snowflake and showcased in the keynote.
• PREVIEW9 private preview items unlocked directly from our PM sessions — from Agent Studio to XMLA endpoint.

Where Snowflake is going — and why our bets are right

Snowflake's direction is unambiguous: the entire company is pivoting around the Agentic Enterprise — governed AI agents that don't just answer questions, but take action. Our current Wingman architecture anticipates this shift.

• AISnowflake Intelligence → CoWork (now GA): Evolved from Q&A agents to a personal work agent with user memory, skills, and multi-agent orchestration. Features we've been requesting — skill catalogs, mobile access, curated publishing — are now shipping.
• DATACortex Sense Runtime: Auto-builds business context with zero manual setup, achieving 86% accuracy on hard enterprise questions vs. 24% baseline. Directly addresses the trust gap we currently manage with manual QA review.
• SAPSAP-Snowflake integration now GA: Bidirectional semantic view pipeline with automated lineage and CSN conversion. Given our SAP dependency, this unlocks future data unification without pipeline duplication.
• ENGdbt on Snowflake — CI/CD modernization: GitHub Actions + Snowflake CLI + zero-copy clone CI. Eliminates our single-developer deployment bottleneck.
• GOVHorizon governance expansion: New intent-driven governance, agent identity policies, and multi-party approvals to scale Wingman without increasing manual oversight.
• OBSAgent Studio + versioning (now GA): Git-backed YAML configuration, version aliases, traffic splitting. Directly solves our multi-version testing clutter and YAML management burden.
• EVALLLM-based automated monitoring (July PrPr): "LLM judges" auto-score traces on answer relevance, plan quality, logical consistency, and a new "user frustration" metric — replacing our manual daily review.

What Snowflake PMs said — directly

Our team secured direct briefings with Snowflake PMs across Intelligence/CoWork, Observability, Semantic Layer, and dbt Products. Click any session for detailed notes.

Effie Goenawan

Principal PM · Intelligence & CoWork

Snowflake Intelligence & CoWork roadmap

9 Snowflake PMs in one session. RBNA confirmed as top Artifacts PRP customer. Our conversations directly inform Snowflake's product direction. The memo that made the keynote.

View detailed notes →

Alex Issa

PM · AI Observability & Evaluations

Agent monitoring, observability & lifecycle

Agent versioning now GA. LLM-based automated monitoring (July). Coco Evolve: 4 devs, 4 months → 6 hours. Our manual review process has a clear replacement path.

View detailed notes →

Josh Klahr

Head of Product · AI-Powered BI

Semantic views, ontology & OSI alignment

"Ahead of the game." Our ontology approach mirrors Cortex Sense's own architecture. LOD, Power BI PBIT, and XMLA endpoint private previews unlocked.

View detailed notes →

Uche Nkadi

Product Lead · dbt Projects on Snowflake

dbt on Snowflake — CI/CD & orchestration

Ahead of most customers. GitHub Actions + Snowflake CLI + zero-copy clone CI recommended. VSCodium dbt IDE private preview June 15. Snowflake Tasks → Airflow parity by Dec 2026.

View detailed notes →

← Back to sessions

Effie Goenawan

Principal Product Manager · Snowflake Intelligence & CoWork

Snowflake profile →

The memo that made the keynote: Sameer, Avi, and Neil submitted 12 feature requests for the Intelligence platform. Snowflake built 9 of the 12 — all showcased in the Summit keynote on June 2, 2026.

Also in this session — 9 Snowflake PMs

Alan Jin

Sr. PM · SQL & Workspaces

Shrikant Shanbhag

Director of Engineering

Skills & Capabilities (new tab in CoWork)

New Capabilities tab includes Skills, MCP connected, and Agent sub-agent sections.

A Skill is a codified workflow — e.g., writing meeting notes, populating a book of business, engineering design docs. Created via chat or markdown file upload.

User Skills are personal. Agent Skills are predefined by the agent creator. A shared skills catalog lets teams upload and reuse skills.

Chart customization (e.g., Red Bull brand colors) managed at agent admin level — not user skill level — for consistency across all users.

New Dashboarding Capabilities

Dashboards built in Workspaces IDE — manually or fully via AI ("Coco"). Inspect and edit SQL for each chart. Advanced types: waterfall and box-whisker plots.

Publishing via RBAC: assign "Manage artifact publication" privilege to specific roles. Only curated dashboards go to the governed space; ad-hoc sharing restricted to small teams.

"Coco talk": users can ask follow-up questions directly on published dashboards. Agent has user context and memory.

User memory stored in "Cord," built from conversations. Users can instruct agent to remember preferences (e.g., "I like line charts"). Admin privacy controls for redacted vs. unredacted memory content.

Agent Routing & Multi-Agent Orchestration

Routing is description-based — agent descriptions determine which agent is triggered. User memory improves routing over time.

For conflicting data access (e.g., Trade Sales sees volume not dollars; Walmart sees dollars): use manual agent selector in plus menu, or a router agent with clear instructions.

Model routing rolling out: simpler models for simpler queries to reduce latency and cost without impacting answer quality.

PRP Timelines

Personal Work Agent PRP (user memory, multi-agent orchestration, user skills, skill catalog, automations): enroll leaders by June 8, 2026.

Dashboard PRP: can start June 8 but recommended to wait until June 15–17 for better experience (code freeze currently in place).

Ontology & Semantic Layer Recommendations

No dedicated ontology product yet; Business Glossary planned from Horizon team (no date).

Cortex Sense auto-reaches ~80% accuracy. To hit 90%+ (critical for business users): invest in semantic views.

Mobile & Notifications Roadmap

Current: email and selective Slack notifications. Teams notifications not yet on roadmap but under consideration.

Location awareness is implementable — pass location as request context. High value for field staff: personalized, time-of-day-aware information at a specific store.

← Back to sessions

Alex Issa

Product Manager · AI Observability & Evaluations · Snowflake

Background: distributed systems, ML, serverless architecture · CS BS & MS, UT Austin

Validated: RBNA's command center approach — tracking question volume, login frequency, and agent QA — is among the most structured feedback loops Alex sees from enterprise customers. Our methodology is being documented as a Snowflake reference pattern.

Current Pain Points — Addressed

Manual monitoring: our daily review of every question and all negative feedback is unsustainable at scale. Snowflake's roadmap directly targets this.

Multi-version clutter: deploying multiple agent versions for testing creates sprawl. Agent versioning (now GA) solves this with aliases and traffic splitting.

Agent Versioning & Agent Studio (now GA)

Tag versions with aliases (default, beta). Split traffic between versions. Agent and semantic configs defined as YAML files in Git.

Agent Studio and Semantic Studio are separate (different user personas) but expected to converge over time.

Feedback captured: UI edits don't currently sync back to YAML — Snowflake team acknowledged and is tracking.

Observability Roadmap

New workspace UI with built-in observability: session/token usage, cost, active users, latency (P50, P95, P99), error rates out of the box.

"Agent metric functions" enable alerting via email, PagerDuty, and other integrations for specific thresholds.

LLM-based monitoring (July PrPr): "LLM judges" auto-score traces — answer relevance, plan quality, logical consistency, and new "user frustration" metric. Replaces thumbs up/down dependency.

Agent Autopilot (Q3 2026): Coco reads monitoring signals, identifies problems, suggests fixes, runs evals, promotes a new version, and commits to Git — automated optimization across the full stack including semantic views.

Coco Evolve (Pre-research Preview)

Proof of concept: recreated an agent that took 4 developers 4 months — in 6 hours. Inspired by DeepMind's AlphaEvolve.

"Human-out-of-the-loop" framework. Provide a dataset of Q&A requirements. Coco Evolve scans the account for relevant data, generates semantic views and tools, and uses an evolutionary algorithm to produce a production-ready agent in hours.

Evals & Ground Truth (coming mid–end June)

Evals will support pointing to specific agent versions — enabling true A/B testing.

New eval metrics: Tool Execution Accuracy (TEA) and Tool Selection Accuracy (TSA) entering public preview week of June 9.

Datasets versioned as native Snowflake objects. Annotation feature to fix incorrect eval questions planned for June 2026.

Agent memory as ground truth: human-curated memory is a valid starting point for building the golden set, despite being slow and costly in prior tests.

← Back to sessions

Josh Klahr

Head of Product Management · AI-Powered BI · Snowflake

25+ years in data & analytics · Previously VP Product at Splunk, AtScale, Pivotal · Started in big data at Yahoo!

"Ahead of the game" — Josh's direct assessment of RBNA relative to most Snowflake customers. Our ontology-first Python approach mirrors how Cortex Sense internally represents its own knowledge graph.

Semantic Views & Cortex Sense

Our team built semantic views on critical datasets over the past year, exposed via "core intelligence," with thumbs up/down feedback and a command center tracking engagement.

Cortex Sense as "ambient context": mines query history to annotate objects, reaching 60–70% accuracy without curation. Semantic views + verified queries gets to 90%+.

Open spec for formatting metadata on semantic views: number formats, metric direction (good vs. bad), default coloring. Scope (global vs. per-view) still being determined.

Ontology & OSI

Our Python-based knowledge graph (nodes/edges tables, KPI construction, metadata provisioning) is the right approach — matches Snowflake's own internal Cortex Sense architecture.

Josh recommended "Ontology on Snowflake" from Snowflake Labs: models ontology first, deploys both the knowledge graph and semantic views from a single source. Search: "ontology on Snowflake."

Ontology as a native Snowflake object doesn't exist yet — but should. Potential announcement at the November 2026 Snowflake event.

OSI donated to Apache Software Foundation — Snowflake will submit PRs but expect some divergence over time (similar to Iceberg geospatial spec). Build to Snowflake-native semantic views as the durable layer.

LOD & Complex Aggregations

Our SAP BW nested aggregations (average of average, differing denominators) have historically been a gap. LOD (Level of Detail) plus facts across grains should enable these ratio aggregates.

LOD available in private preview — enabled for our team. Josh requested our concrete schema/query examples to share with engineering for evaluation.

Action: provide SAP BW nested aggregation examples to Snowflake for LOD evaluation.

Power BI Integrations (June Private Previews)

PBIT Export (June): Export Snowflake semantic view as a Power BI model. Once exported, disconnected — but gives teams a governed starting point.

XMLA Endpoint (end of June): Expose Snowflake semantic view as SSAS via XMLA. Any XMLA-compatible BI tool connects directly. Early build — we're signed up to file bugs and shape the feature.

Composable many-to-many ("holy grail") — also in private preview. Interest in testing bidirectional PBIX/PBIT ↔ semantic view conversion.

Skills & Vega Charting

Tested Vega templating for charting at semantic and agent levels. Josh favors a composable skill repository — reusable skills (e.g., "graphing skill") referenced across agents, semantic views, and entities. Not yet implemented.

← Back to sessions

Uche Nkadi

Product Lead · dbt Projects on Snowflake

6.5 years career (1.5 at Snowflake) · Developer Experience, Developer Tooling, Observability

Ahead of most customers in dbt maturity. Used by 1/3 of Snowflake companies including all top 20 by revenue. Our 1.5-year dbt investment is the right foundation — the recommended architecture validates what we've built.

Our Current Pain Points — Addressed

Single-developer bottleneck: critical pipelines depend on one person deploying from personal workspace. CI/CD migration to Snowflake CLI + GitHub Actions removes this entirely.

profiles.yml disappears on git pull: due to .gitignore. Fix: standardized secure backup and regeneration scripts.

Dagster dependency: Snowflake Tasks roadmap targets Airflow parity in ~6 months (GA ~December 2026). Can run deployed dbt project from any orchestrator today via SQL or CLI.

Recommended CI/CD Architecture

CI: GitHub Actions on every PR → snow dbt deploy deploys tester object → runs with service principal credentials → fails action if tests fail.

CD: On merge → separate CD job promotes to production account. CI/CD lives in GitHub; Snowflake Tasks handle production orchestration.

Zero-copy clone CI (June 15 preview): PR number deploys tester object + clones staging DB (clone_staging_PR_123). High-fidelity testing on production-equivalent data at zero storage cost.

Cross-account runs from workspace not on roadmap (security). Use CI to deploy tester to staging account; CD to production account — standard pattern for banking customers.

New IDE & Environment Variables

Cortex dbt IDE (VSCodium-based, June 15 PrPr): full dbt development experience with right-hand Cortex AI panel for debugging. Toggles between local and Snowflake-managed modes. Supports extensions and terminal access.

env.yaml (upcoming): pre-execution variable management. Developer-specific config via current_user(). Production orchestration: embed SQL to read task_history for dynamic date intervals.

Snowflake Tasks Roadmap

External API calls (inbound and outbound) — planned feature for Tasks.

Task Graph object in development: complex "wait on" dependencies similar to Airflow sensors. Milestone releases ~August and ~October 2026; GA ~December 2026.

DCM + Tasks already enables large task graph composition today.

First-party integration discussions ongoing with Dagster and Astronomer — nothing signed yet.

Event-Driven Pipelines (SAP Streams)

Use Snowflake Streams to detect new SAP data landing in bronze → trigger Snowflake Tasks immediately. Eliminates cron scheduling delays. Tasks always reference latest validated dbt project version.

Concrete actions & private preview access

Account expectations & strategic conversations

Beyond the sessions

The most valuable conversations often happen off the agenda — from the partner cruise to team dinners to a robot that serves Red Bull.

RBNA team arriving — Day 1 at Snowflake Summit

BI & Analytics — Sameer, Avi, Alex, Neil

~20,000 attendees — the scale of Summit 2026

Expo floor — networking across 350 organizations

Domo team — "We love Red Bull"

BI & Analytics team in San Francisco

Neil with Benoit Dageville, Thierry Cruanes & CTO S. Muralidhar Snowflake co-founders & CTO — SIGMOD Test-of-Time Award recipients. Neil attended "From First Principles" and published 5 lessons from the conversation.

Read the post →

Red Bull Basement — San Francisco

A separate Red Bull-organized event running in SF alongside Summit week. Red Bull Basement is Red Bull's global platform celebrating tech innovation and entrepreneurship.

Robot serves a Red Bull at Basement

A DYNA robotic arm at the Red Bull Basement picked up a Red Bull can — arguably the most on-brand moment of the week.

Partner cruise — past Alcatraz at sunset

Snowflake DACH arranged a private cruise for their European customers — Siemens, Palfinger, Red Bull, an Austrian bank, and Hepster. Key conversations around SAP, platform strategy, and cross-company benchmarking happened here.

The Commodore — SF Bay

Three hours on the bay — past the Golden Gate Bridge, around Alcatraz. Where the real networking happened.

BI & Analytics team on the bay

The BI & Analytics team on the upper deck of the Commodore.

Neil meets a humanoid robot

The expo floor gave a glimpse of where AI is heading physically. This one was already wearing a tech company shirt.

Team dinner — Day 1

The full extended team at dinner — 16 people across six functions, many meeting in person for the first time.

Wednesday night — Copra SF

RB team dinner at Copra, San Francisco. Cross-team debrief outside the conference walls.