No Priors Ep. 139 | With Snowflake CEO Sridhar Ramaswamy

Detailed Summary

Sridhar Ramaswamy Introduction (0:00 - 0:42)

• The episode introduces Sridhar Ramaswamy, CEO of Snowflake, highlighting his background as a former founder of Neva and SVP of Google Ads.
• The discussion will cover his first 18 months as CEO, Snowflake's shift to an AI-first company, enterprise AI ROI, and the impact of AI on cloud service providers and the ad model.

Snowflake’s Market Adaptation (0:42 - 3:14)

• Sridhar reflects on his 18 months as CEO, noting the significant changes and positive market reaction to Snowflake's execution.
• He explains that Snowflake, while always a strong product company, was initially slow to adapt to machine learning and AI trends.
• Former CEO Frank Slootman proactively sought a product-first leader to navigate the tumultuous product landscape of AI.
• The company embraced the wave of AI change, transforming its product, marketing, and go-to-market strategies.
• Despite initial doubters, Sridhar was motivated by strong customer loyalty to Snowflake's existing value, leading to the company's current position at the cutting edge of data and AI for enterprises.

Snowflake’s Evolution and AI Integration (3:14 - 5:44)

• Sridhar details the initial six months of his tenure, focusing on tactical changes around accountability and organizational structure.
• He streamlined the organization by reducing layers between engineers and customers, moving from a specialized structure to one emphasizing faster iteration.
• Product engineering was reorganized into areas like AI and core warehousing, with specialized teams created to bring new products to market quickly.
• The core philosophy adopted was "speed wins," prioritizing rapid iteration over meticulously planned strategies.
• Snowflake also refined its identity, evolving from a "data cloud" to an "AI data cloud," clarifying its unique value proposition in the AI space distinct from CSPs or foundation labs.

Introducing Snowflake Intelligence (5:44 - 9:01)

• Snowflake initially explored creating foundation models but pivoted due to the immense capital required to compete with leaders like OpenAI and Anthropic.
• The company shifted its focus to how AI can accelerate data utilization within Snowflake, aiming to attract more data to the platform.
• Snowflake's AI product strategy humbly focused on adding value to existing customer workflows, such as search and text-to-SQL capabilities.
• Snowflake Intelligence (SI) is introduced as an "opinionated agentic platform" specifically designed to create value from data faster, contrasting with general-purpose agentic platforms from CSPs.
• Internal use cases, like the sales data assistant "Raven," demonstrated SI's ability to move beyond inflexible dashboards by providing a more interactive and flexible interface for sales data.
• Early customer collaborations with Cisco, Fanatics, and the USA bobsled team helped refine SI's capabilities, aiming to free users from 2D thinking and enable faster data-driven insights.

Snowflake Intelligence User Experience (9:01 - 11:55)

• Snowflake Intelligence offers an interactive interface where users can ask questions, access canned queries, and inquire about available datasets and answerable questions.
• The aspiration for SI is for it to be used by every employee, not just those proficient in SQL, making it a daily use product.
• Internal tools like Raven provide comprehensive customer information, including contracts, consumption, recent conversations, and outstanding issues.
• SI is not intended to replace BI dashboards like Tableau or Sigma but offers unique capabilities for business users, focusing on trustworthiness and verified outputs.
• Snowflake emphasizes rigorous evaluation for all AI products, treating AI development with the same precision as software engineering to ensure reliability.
• This marks a new motion for Snowflake, moving beyond data teams to directly serve end-users, necessitating identity provider integration and a consumption-based pricing model to encourage broad adoption without runaway costs.

Drawing the Line Between Data, Agent System, and App (11:55 - 13:30)

• Sridhar acknowledges the blurring lines between agentic systems and traditional applications, but clarifies Snowflake's focus.
• Snowflake Intelligence is not designed to replace complex enterprise software like SAP or Salesforce for managing large supply chains.
• However, he foresees agentic systems handling tasks like updating Salesforce via APIs or filing vacation requests through an HR agent.
• Snowflake's approach is opportunistic, focusing on areas where it can provide value and leverage its strengths, rather than pursuing naked ambition to become a general-purpose application provider.
• The goal is to focus on value creation and user needs, which Sridhar believes leads to more durable solutions.

Leadership and Organizational Changes (13:30 - 16:19)

• Sridhar emphasizes that driving change in a large organization is difficult and requires acknowledging that change is hard.
• Initial changes focused on leadership, alignment, and clearer accountability, implemented within a few quarters due to fewer people involved.
• The "war room" or "pod model" was introduced, bringing product, engineering, and go-to-market functions together for faster collaboration.
• He advocates for combining bottoms-up and top-down approaches to change, citing the example of coding agents.
• Benoit Dageville, Snowflake's co-founder, championed coding agent adoption among engineers, proving more effective than top-down mandates.
• Sridhar believes in finding and elevating "forward-thinking, curious" individuals within the organization to drive change.
• Coding agents have significantly reduced the time to create customized demos for solution engineers, enabling more tailored customer experiences.

How Being an Investor, Entrepreneur Informed Sridhar’s Leadership (16:19 - 18:50)

• Sridhar believes past experiences are accretive, adding to one's leadership in unforeseen ways.
• His PhD taught him to focus on ideas and convey them crisply, even through concise abstracts.
• His entrepreneurial journey with Neva, though challenging, taught him about hustling, not taking success for granted, and understanding marketing and social aspects often overlooked at large companies like Google.
• He recalls his first Google launch being covered by the New York Times, highlighting the immediate scale and distribution available at Google, which is not typical for startups.
• These experiences have made him more grateful for his current role at Snowflake and more gracious about dealing with the complexities of leading a large company.
• He likens the energy and excitement around Snowflake to a "rock concert," appreciating the visible enthusiasm from customers and the broader community.

Importance of Product-Market Fit (18:50 - 22:46)

• Sridhar emphasizes that product-market fit is magical and the fundamental reason for Snowflake's existence, despite hyperscalers' desire to dominate the data space.
• He notes the unique value created by companies like Snowflake and Databricks, which hyperscalers haven't fully replicated.
• He finds model companies like OpenAI particularly interesting due to their rapid growth phase, where they believe they can achieve anything.
• It's crucial to anticipate the trajectory of these model companies and identify areas where they have a right to win versus areas that are sufficiently different.
• Coding agents are an example where OpenAI and Anthropic are laser-focused on having the best, indicating a competitive area.
• He advises against building solutions that are merely prompts on top of models, as this is a problematic space.
• Snowflake focuses on differentiating itself from model companies by emphasizing its data platform as a durable category, ensuring distance and unique value.
• He stresses that no software company can feel secure, and continuous innovation is necessary to avoid becoming obsolete, drawing a parallel to Intel's past dominance.

Snowflake’s Strategic Positioning (22:46 - 27:10)

• Sridhar emphasizes the need for Snowflake to continuously earn its position, acknowledging the infinite budgets and patience of CSPs.
• He highlights Snowflake's core strength in the data platform layer, aiming to support customers "from inception to insight"—from data conception to actionable insights.
• He argues that modern, great companies like Google and Meta are "data-first" rather than purely "product-first," contrasting with older models where data was an afterthought.
• Snowflake aims to be a companion for all types of data, providing customers with the data expertise typically found in companies like Google and Meta.
• AI is seen as a massive accelerant, elevating the perceived value of data for CEOs beyond mere modernization or running queries, impacting business operations and efficiency.
• Snowflake's integrated approach, emphasizing simplicity, governance, and cross-CSP functionality, offers a higher level of abstraction than raw compute and storage.
• His pre-AI thesis for joining Snowflake was that a data platform spanning CSPs could be as large as a CSP itself, with AI further accelerating this vision by enabling faster value and action from data.

Snowflake’s Partnership Strategy (27:10 - 30:20)

• Sridhar notes that Snowflake previously had a "Snowflake-centric view," but now recognizes the value of data residing in other enterprise systems like SAP, Workday, and Salesforce.
• He observes the blurring line between software, services, and data, prompting a shift towards a partnership mentality.
• Drawing from his experience at Google in payments, he emphasizes picking partners and collaboratively creating value.
• He cites the improved relationship with Microsoft, moving from a competitive stance to one of collaboration, understanding that Azure + Snowflake can be a "strict positive" for customers.
• Similar arrangements are being pursued with AWS and GCP.
• With software providers like SAP, the focus is on bidirectional data sharing and collaboration in analytics, AI, and agents to make it easier to create solutions on SAP data.
• These partnerships represent a maturing strategy for Snowflake, aiming to leverage partners' global presence (e.g., SAP) to expand its reach and generate more business.

How Sridhar Sees the ROI of AI (30:20 - 35:09)

• Sridhar identifies coding agents as having the easiest ROI, making new projects faster and demystifying technology for more people.
• Snowflake itself is a large user of coding agents and is integrating them into its platform to simplify its use.
• Customer support is another high-ROI area, leveraging AI to tap into human knowledge repositories, handle voice and typed questions, and provide easy backups when AI cannot resolve issues.
• Faster, easier, and more seamless access to data, especially without high per-user license fees, offers significant ROI, which is a motivation behind Snowflake Intelligence's consumption-based model.
• He cautions against obsessing over immediate, large ROI, instead advocating for taking many small "shots on goal" to prove value iteratively.
• He highlights that Snowflake's internal sales data assistant evolved through three versions, each building on the last, demonstrating the power of iteration.
• Sridhar advises customers to start with small investments (e.g., a thousand dollars at a time) in AI with Snowflake, scaling up once significant value is proven.
• He is bullish on applied AI companies that understand immediate usefulness and customer workflows, seeing vast opportunities for those with velocity and a "paranoia to keep expanding."

How AI Changes the Ad Model (35:09 - 38:15)

• Sridhar believes advertising is an incredibly powerful and enduring business model that will reinvent itself in the chat world.
• He expresses pride in his past work on Google's search ads, emphasizing that there are "good ways of doing advertising."
• His concern is that AI-driven ads might become more insidious, making it harder for users to distinguish between organic content and advertisements.
• He uses the analogy of a psychiatrist with an affinity for prescribing certain medications to illustrate potential ethical dilemmas.
• Despite these concerns, he asserts that the ad model is here to stay, albeit in different forms, and consumers must be smart about what they engage with.
• He is encouraged by the rise of citations and sourcing in AI models, which allows users to verify information and look at primary sources.
• He notes that tools like ChatGPT and Gemini make it easy to verify links and information in research articles, empowering users with more tools for critical evaluation.

Why LLMs Still Need Search (38:15 - 42:11)

• Sridhar addresses the misconception that traditional information retrieval and indexing (search) are becoming irrelevant with LLMs.
• He explains that Google's initial power came from PageRank, which harnessed the internet's popularity, but its long-term value came from incorporating click behavior and feedback loops.
• He applies this concept to AI systems, including Snowflake Intelligence, emphasizing the need for "eval loops" for products to improve over time.
• He draws a parallel to LLMs and math: while LLMs can attempt math, it's more reliable to use Python code for calculations, which can be directed and run by the LLM.
• Similarly, for trustworthiness in answering questions, well-known solutions from search should be used as external tools by AI systems rather than relying solely on the LLM's internal capabilities.
• He argues against a "maximalist mode" where AI is expected to solve everything, stating that practical people use the best available tools.
• The prevalence of search APIs demonstrates that LLMs benefit from external information that is not easily internalized into the model itself.

Conclusion (42:11 - 42:11)

• Sridhar thanks Sarah for the interview.
• The host encourages listeners to find the podcast on social media, subscribe, and visit the website for transcripts.

Notable Quotes

"The last 18 months have really been about embracing that wave of change." — Sridhar Ramaswamy

"Speed wins. Ability to iterate always trumps carefully laid out strategies." — Sridhar Ramaswamy

"Snowflake Intelligence is very focused on how do you create value from data whether it's structured or unstructured a whole lot faster." — Sridhar Ramaswamy

"Our aspiration was for this product to be used by every single employee in the company. So it's not for people who can write SQL." — Sridhar Ramaswamy

"We need to think of AI the same way we think about software engineering, which is there's a right and there's a wrong. It cannot be this mode of like yolo AI." — Sridhar Ramaswamy

"Defensibility is built, not strategized. That's correct. It's built every single day. You have to keep moving." — Sridhar Ramaswamy

"AI is a massive accelerant on how do you get value from data faster or how do you get better at acting on data quicker?" — Sridhar Ramaswamy

"Coding agents are probably the easiest ROI just in terms of making new projects faster, making tech demystifying technology so that more people can get at it." — Sridhar Ramaswamy

"I think obsessing about a lot of ROI too quickly is also a bad idea for many companies because you don't want your first step to be 100 ft. You want to do a lot of little things that sort of prove value." — Sridhar Ramaswamy

"The ad model is here to stay. It will just come in different forms." — Sridhar Ramaswamy

"A maximal intelligence will use reliable tools wherever it is available." — Sridhar Ramaswamy