No PriorsNo Priors Ep. 139 | With Snowflake CEO Sridhar Ramaswamy
CHAPTERS
- 0:00 – 3:14
CEO transition and Snowflake’s pivot to an AI-first posture
Sarah frames Sridhar’s first 18 months as CEO and asks what changed after taking over from Frank Slootman. Sridhar describes why the company needed to react faster to ML/AI shifts and how Snowflake retooled product and go-to-market amid skepticism.
- •Snowflake’s early product advantage and IPO-era momentum
- •Why leadership changed: being “slow” to react to AI/ML waves
- •AI becoming a fast-moving, everyday reality forcing rapid adaptation
- •Transformation extended beyond product into marketing and GTM
- •Customer love for Snowflake as the core confidence signal
- 3:14 – 5:50
First-six-months operating model: accountability, re-org, and faster iteration
Sridhar explains the tactical changes he prioritized early—reducing organizational distance from engineers to customers and installing clearer ownership. He emphasizes speed and iteration as the winning strategy in an uncertain AI environment.
- •Over-specialization created 7–10 layers between feature and customer value
- •Shift to accountable product-area structures (AI vs core warehousing/analytics)
- •Tighter linkage from product/engineering to go-to-market execution
- •“Speed wins”: iteration beats overly rigid long-range planning
- •Finding Snowflake’s AI identity: from “Data Cloud” to “AI Data Cloud”
- 5:50 – 7:13
AI strategy reset: don’t be a foundation model lab—accelerate value from Snowflake data
They discuss Snowflake’s early attempt at building foundation models and the decision to pivot away from competing with OpenAI/Anthropic. The new strategy focuses on adding AI capabilities (search, text-to-SQL) that compound the value of data already in Snowflake and bring more data in over time.
- •Built an early credible MoE model, then recognized capital/competition realities
- •Strategic pivot: leverage AI to unlock value from customer data in Snowflake
- •Humble product strategy: augment existing workflows instead of “rethinking everything”
- •Investments in enabling components (search, text-to-SQL)
- •Using AI value as a pull to consolidate more enterprise data into Snowflake
- 7:13 – 9:00
What “Snowflake Intelligence” is: an opinionated agentic platform for data value
Sridhar introduces Snowflake Intelligence (SI) as an agentic platform with a narrow, pragmatic focus: help users get value from structured and unstructured data faster. He contrasts it with generic ‘one agent to rule them all’ platforms and anchors the product in concrete internal and customer use cases.
- •SI is agentic but intentionally opinionated and scoped
- •Focus: speed to insight/value from enterprise data (structured + unstructured)
- •Avoiding the paralysis of infinite workflows in generic agent frameworks
- •Internal catalyst: consolidating sales dashboards into a single assistant (“Raven”)
- •Early customer exploration across varied orgs to validate real workflows
- 9:00 – 10:29
Snowflake Intelligence UX: natural-language access for every employee (not just SQL users)
They walk through how users interact with SI: an interactive Q&A interface with prompts, dataset discovery, and guided questions. The aspiration is daily-use utility for all employees, exemplified by preparing for customer meetings with rich, up-to-date context.
- •Chat-like interface with canned questions to reduce “blank page” friction
- •Self-discovery: what datasets exist and what questions can be answered
- •Designed for broad business users—not only analysts/SQL writers
- •Sales assistant workflow: relationship, contract, consumption, and recent interactions
- •Not a BI replacement; complements dashboards by answering flexible follow-ups
- 10:29 – 11:58
Trust, evals, and cost model: making AI reliable and adoptable at enterprise scale
Sridhar stresses that enterprise AI must behave more like engineering—there is a right and wrong—so trust requires evaluations and safe iteration. He also highlights adoption blockers (identity/account setup, subscription fatigue) and explains SI’s consumption-based approach while preventing runaway spend.
- •Rejecting “YOLO AI”: reliability and correctness expectations
- •Evals for every launch and for model changes to prevent regressions
- •Shift from data-team-only tooling to end-user direct access
- •Enterprise enablement: identity provider integration to avoid per-user account friction
- •Consumption pricing + experimentation to balance adoption with cost controls
- 11:58 – 13:30
Where data ends and apps begin: opportunistic agents without pretending to be Salesforce
Sarah challenges whether Raven/SI is effectively an ‘app.’ Sridhar argues the boundaries between agents and traditional software will be messy, but Snowflake should expand only where it can deliver clear value—using integrations/APIs to automate adjacent workflows without overreaching.
- •No clean boundary: agentic systems and apps will overlap heavily
- •Avoiding category cosplay (not trying to be SAP/Salesforce)
- •Reasonable adjacent actions via APIs (e.g., updating CRM, filing HR requests)
- •Strategy: follow value creation and user needs, not “naked ambition”
- •Durability comes from leveraging Snowflake’s strengths in data and trust
- 13:30 – 16:18
Leading change at scale: war-room pods, champions, and bottom-up adoption of AI tools
Sridhar explains how he implemented organizational change without over-disrupting the company: start with leadership/accountability, then cross-functional pod models. For cultural/tooling shifts like coding agents, he combines top-down intent with bottom-up champions to drive real adoption.
- •Sequence changes: leadership alignment first, broader rollout later
- •Pod/war-room model: product, engineering, GTM, marketing working together
- •Tooling adoption is social: skepticism is normal and must be managed
- •Find and elevate internal champions (example: Benoît driving coding agents)
- •Coding agents improved SE demo customization speed and customer relevance
- 16:18 – 18:50
How investor and founder experiences shaped his CEO approach
Sridhar reflects on how research training, Google scale, and the hardship of building Neeva shaped his leadership. He emphasizes clearer thinking/communication, humility about distribution, and gratitude for operating at Snowflake’s scale and customer enthusiasm.
- •PhD training: crisp articulation and focus on core ideas
- •Google: distribution makes shipping feel easy and highly amplified
- •Neeva: taught hustling, marketing realities, and not taking success for granted
- •Greater appreciation for the privilege and responsibility of the CEO role
- •Perspective on operating large moments (e.g., Snowflake Summit’s scale)
- 18:50 – 23:17
Defensibility and product-market fit when building on “elephants” (CSPs and model providers)
They discuss how Snowflake can create durable value while built atop hyperscalers—and analogously, how companies build on foundation model providers. Sridhar argues PMF remains the magic, but companies must anticipate ‘empire’ expansion from OpenAI/Anthropic and continuously earn their position.
- •PMF explains why Snowflake/Databricks thrive despite hyperscaler incentives
- •Model providers as fast-expanding “empires” before they hit boundaries
- •Watch where platforms are clearly headed (e.g., coding agents as core battleground)
- •Avoid brittle businesses that are just thin prompt layers
- •Lesson from CSPs: infinite budgets/patience—defensibility is built daily
- 23:17 – 27:10
Three-to-five-year view: Snowflake as the integrated AI data platform from inception to insight
Sridhar outlines Snowflake’s strategic north star: serve customers across the full lifecycle from data creation to actionable insight, with AI accelerating time-to-value. He argues the value of data has risen dramatically and Snowflake’s integrated governance + sharing across clouds creates durable differentiation versus raw compute/storage approaches.
- •Vision: “inception to insight” companion for enterprise data workflows
- •Modern winners (Google/Meta) were data-first with rapid feedback loops
- •AI elevates data’s strategic value and operational impact for CEOs
- •Differentiation: simplicity, integration, governance, and cross-cloud data sharing
- •Higher-level abstraction than buying raw compute/storage and assembling systems
- 27:10 – 30:20
Partnership strategy: hyperscalers plus SAP as bidirectional data + AI/agent collaboration
Sridhar describes a shift from a Snowflake-centric worldview to a partnership mentality, learned in part from Google. He covers evolving hyperscaler relationships (especially Microsoft) and explains why partnerships with software vendors like SAP matter: shared customer data value and opportunities for joint analytics/AI/agent experiences.
- •Data from SAP/Workday/Salesforce is valuable and increasingly recognized as such
- •Partnership mindset: pick key partners and create ‘1+1=3’ value
- •Microsoft relationship maturation: compete in some areas, collaborate where Azure+Snowflake wins
- •Similar cooperative posture with AWS and work toward GCP alignment
- •SAP: bidirectional data sharing plus easier analytics/AI/agent building atop SAP data
- 30:20 – 35:08
Enterprise AI ROI: quick wins, democratized data access, and iterative ‘$1,000 at a time’ adoption
Sarah asks for the highest-ROI AI use cases in enterprises. Sridhar prioritizes coding agents and customer support, then broad data access without heavy per-seat licensing, while warning against demanding massive ROI upfront—iteration and many small experiments uncover the real wins.
- •Top ROI: coding agents (faster projects, wider technical leverage)
- •Strong ROI pattern: customer support with knowledge bases + human fallback
- •Democratized data access as ROI lever, especially vs expensive per-seat BI licenses
- •Anti-pattern: making the first AI step a huge bet; prefer many small trials
- •Iterate to value: multiple internal versions led to the sales assistant’s success
- 35:08 – 42:11
Ads in the chat era and why LLMs still need search and external tools
They close with two forward-looking topics from Sridhar’s Google/Neeva background. He argues advertising will persist but must remain clearly disclosed in conversational interfaces, and that search/IR remains crucial because trustworthy systems use the best external tools and feedback loops—like clicks and evals—to improve over time.
- •Ads will reinvent in chat UIs; key risk is more hidden/insidious ad disclosure
- •Consumers must preserve agency; citations/sourcing are encouraging trends
- •Cross-checking and verification across tools is becoming easier for users
- •Search value: ranking/trust and feedback loops (clicks, evals) drive quality over time
- •Principle: intelligent systems use reliable external tools (search, databases, code) rather than forcing LLMs to do everything
