[Report] SNOWFLAKE WORLD TOUR 2026 - TOKYO: Day 2 KEYNOTE #SWTTokyo26

This is Kawabata.

"SNOWFLAKE WORLD TOUR 2026 - TOKYO" was held from September 10, 2026 to September 11, 2026.

https://www.snowflake.com/ja/world-tour/tokyo/

This article is a report blog for the session "Day2 KEYNOTE."

For "Day1 KEYNOTE," please check the link below.

https://dev.classmethod.jp/articles/snowflake-world-tour-2026-tokyo-day1-keynote-swttokyo26/

※Some features include those in preview or coming soon. Please check the latest official documentation for availability.
※Statements by speakers are summarized based on the author's notes. Conversations in English are paraphrased by the author.

【Addendum】
I was selected as a finalist in the "RISING COMMUNITY LEADER OF THE YEAR" category, APJ slot, of the Snowflake Community Awards.
Please check the details at the link below.

https://dev.classmethod.jp/articles/snowflake-community-awards-finalist-activities-review/

Speakers

• Sridhar Ramaswamy
  • Snowflake Chief Executive Officer (CEO)
• John Robertson
  • Snowflake APJ President and Chairman Executive Officer (VP, APJ Sales)
• Ryuji Ukita
  • Snowflake President and Executive Officer
• Munehisa Matsumoto
  • Executive Officer, Sumitomo Mitsui Trust Asset Management Co., Ltd.
• Chihiro Ueno
  • Executive Officer, Head of AI Operations Division, CyberAgent, Inc.
• Ken Takao
  • Head of Company-wide Data Technology Division, Group IT Promotion Headquarters, CyberAgent, Inc.
• Toshiyuki Takahashi
  • Snowflake Marketing Division, Product Marketing Manager and Evangelist

The event started with Kumataro's drum performance, just like Day1

It really got everyone pumped up with a rock vibe.

From Snowflake's Ukita

As mentioned on Day1 as well, there were references to the companies and partners presenting.

• Snowflake World Tour is held in more than 20 cities around the world. Tokyo is the only 2-day event in the world, making it the largest event in the World Tour

• Expanding even further than last year, with over 12,000 registered attendees, 144 sessions, more than 42 partners, and 202 speakers
• This event is centered around customers, not Snowflake. Day2 features a conversation with CEO Sridhar, plus introductions to data and AI initiatives at Sumitomo Mitsui Trust Asset Management and CyberAgent
• Words of gratitude to sponsoring partners and guidance on the Expo area. He hopes attendees find ideas useful for new technologies, services, businesses, and personal projects
• For those who couldn't attend Day1, Evangelist Takahashi recaps Day1

Day1 Recap by Snowflake's Takahashi

This was a segment that reorganized the Day1 KEYNOTE content as "The 4 Elements of the Agentic Enterprise." Please see the Day1 report linked above for details.

• We are entering an era where AI agents do the work, and the way we work and the nature of companies is about to change significantly. Snowflake calls this the Agentic Enterprise
• The reason behind the voice saying "I tried using agents but not everything works out" is one thing: data and systems are fragmented and siloed, leaving them in a state where AI cannot leverage them
• Snowflake provides the 4 elements to overcome this fragmentation on a single platform
  • Data: Govern with Horizon Catalog, assign meaning to data with Horizon Context. Control access for both humans and agents with Trust Center
  • AI Models: Choose models according to performance and cost. Cortex AI Gateway automatically selects the right model for each task and tracks agent activity
  • Applications: Connect to data from SAP and others zero-copy without copying, and integrate with other services via MCP
  • Control Plane: The command center that organically connects the three. Snowflake CoWork for business users and Snowflake CoCo for builders
• Memorable quotes from Day1 customer sessions
  • Nissan Motor: "AI is an enabler of corporate transformation." Having data accessible to AI changed the mindset of business divisions
  • Fujifilm: "Governance is not a brake, but a guardrail that allows everyone to safely press the accelerator"
  • Anthropic: "The biggest risk in AI is not using AI"
• Three closing points: Get your data assets in order, make yourself AI ready, and use AI to serve the business rather than making AI usage an end in itself

From Snowflake's John Robertson

• Head of APJ, based in Tokyo. This is his 35th year in Japan, having worked almost entirely in the IT industry for American companies
• American companies often don't understand Japan, and half of Snowflake Japan's work is communicating Japan to headquarters
• The most important thing is "whether headquarters and the CEO understand Japan." Sridhar has visited Japan almost every year since his Google days, and it's reassuring to have a CEO who loves Japan

Conversation between Snowflake's John Robertson and Sridhar Ramaswamy

Q. What do you love about Japan?

• The passion poured into everything. Not just the management of conferences, but even small things like hotel doorknobs carry craftsmanship and love
• Small coffee shop and restaurant owners take pride in their work. He tries to have every Snowflake employee bring that same passion to their products and contributions to customers

Q. What is the Agentic Enterprise, and what impact is it having on customers?

• Agentic AI is the "industrialization of intelligence." Just as writing was invented not for poetry but for tax records, software is a way to express processes. Now, creating software has become cheap and fast, and processes can be expressed as skills in natural language
• Especially important for Japan, which cannot scale by increasing population. Tremendous acceleration can be achieved with the people you already have
• Snowflake's internal practice: Distributed Snowflake CoCo and Snowflake CoWork to the sales team
  • Day-to-day tasks changed, such as "what use cases are open for this customer" and "how should I prepare for a meeting"
  • Sridhar took a photo of an invitation list for a lunch and put it into Snowflake CoWork on his smartphone to find out who he had met before. Previously, that would have taken someone a week to research
• Governance is necessary precisely because everyone uses the same agents. Not everyone should be able to see real-time sales figures for all customers
• In the near future, everyone will have a "text box to ask questions" filled with the entire company's wisdom and context. That is the beginning of the Agentic Enterprise era

QBR preparation done in 4 minutes

• This morning, when Sridhar asked "do you have a grasp on the top 30 customers' situations?", he got the answer faster than Sridhar using Snowflake CoWork
• Which customers are using which use cases and AI products, and where they are headed, was discussed in 4 minutes. Previously, that would have required a QBR and hours of preparation
• The entire workflow of asking a data engineer to create a chart, not understanding it, and setting up a Zoom meeting all disappears
• With RBAC, you can only access data permitted for your role as GM. That is governance itself, which is necessary when deploying AI

Q. How do you see the Japanese market?

• A progressive country with one of the world's largest economies that is eager to leverage technology. Many global brands, and deep relationships with key players have been established
• It is also a market that demands perfection. A CFO who says "the revenue numbers are roughly right" would not be tolerated. Snowflake lives up to those expectations
• Customers should expect more from Snowflake. There are customers completing legacy migrations in under 3 months that previously took over 3 years. The goal is to shorten this by a factor of 10 and complete complex migrations within a month
• Snowflake's own data team has also transformed from an organization with a 5-year backlog to one that leads transformation. Data teams can move from being a slow cost center to being the center of transforming a company with data
• What he tells every customer he meets: "If you fully adopt Snowflake CoCo and Snowflake CoWork, you'll be 5 to 10 times faster, not 5 to 10 percent"
• John: Japan still has a lot of on-premises infrastructure. It's a chance to migrate to the right cloud data platform and prepare for AI

Q. Flexibility and choice in AI and cloud. Any announcements for Japan?

• Adopted open data formats like Iceberg and Polaris, and available on all of AWS, Azure, and GCP. Data should not be locked in by anyone, including Snowflake
• Models can also be chosen from OpenAI, Anthropic, and open-source/open-weight models
• Snowflake will become available on Google Cloud in Japan (Tokyo region) in FY27 Q4

【Press Release】
https://www.snowflake.com/ja/news/press-releases/snowflake-google-cloud-tokyo-region-japan/

Q. What is your advice on ROI of AI investment and token economics?

• AI is technology from which real ROI can be demanded. There are only 3 reasons to spend in business: "earn more," "spend less," or "provide better service to customers"
• Inside Snowflake, they deployed AI to the sales team and stopped spending on dashboards, saving about $5 million. Spending on AI was considerably less than that
• Per-user quotas allow you to set "the maximum amount one person can spend in a month." This preserves the benefits of the consumption model while preventing runaway spending by individual users
• "Earn more" metrics: Proposed use cases per account executive up more than 40% year-over-year, deployed use cases up approximately 90%, new logo acquisition at one of the best quarters ever
• The cost governance skill in Snowflake CoCo (one of the top 10 skills) can reduce unnecessary usage. Cortex AI Gateway automatically routes to the lowest-cost model appropriate for the task
• Advice: Rigorously demand positive ROI. AI is a means to an end; there's no need to burn tokens just to say "we're using AI"

Q. What is your advice for making data a competitive advantage?

• In a world where software and intelligence are no longer expensive, your alpha is your data, context, and operational knowledge of your business. Complex UIs and algorithms can be expressed as natural language skills
• The presentational layer is generated on the fly and then disappears. Inside Snowflake, SKO reports are created on the spot, uploaded to Snowflake CoWork, and shared
• What matters is data quality and the logic that decides what to do with that data. Companies that are rapidly advancing are those that have organized their data assets and understand and optimize their most important processes
• A large company's growth rate leaping upward doesn't happen without real magic. That magic is the release of data assets and thorough process optimization, now drivable by the people who actually do the work

Presentation by Matsumoto of Sumitomo Mitsui Trust Asset Management

One of the core companies of Sumitomo Mitsui Trust Group, with approximately 111 trillion yen in assets under management and approximately 714 employees. The presentation covered their efforts to consolidate business data and logic centered on Snowflake, and what lies ahead.

Company-wide technology utilization framework

• Committed at a company-wide level to "building a lean company that can operate with half the members"
• Each department identifies all tasks, prioritizes them, and creates scenarios. About 650 tasks have been turned into Skills across the company
• Promotion members are selected from each department, and progress is driven by self-sufficiency and in-house development. Members who produce results are properly evaluated
• PDCA is automated rather than managed in Excel. It is run by combining three elements: the full task list, created Snowflake objects, and usage logs

Approach and skill creation

• ① Organize the data → ② Business staff use the data to execute tasks and put into words their perspectives, decision criteria, and outputs → ③ With just the phrase "turn what we just did into a skill," that knowledge becomes a skill
• In 2025, the stage was "LLMs write code for you," but with the arrival of agents, natural language can be run as-is, significantly lowering the barrier to entry for in-house development

• Example of a skill: a skill for creating stock commentary. "What to do," "when to use it," and "how to do it" are all defined in natural language, and with just the phrase "create a commentary introducing CyberAgent's investment stocks," the output comes out

• An episode illustrating the impact of skills
  • When Anthropic announced a business-specific plugin in February 2026, the stock prices of Thomson Reuters in the legal domain and IBM in COBOL analysis fell
  • What happened was "the release of a mechanism to load knowledge onto a general-purpose model in natural language," and that alone had an undeniable impact on the market

Activity results

• Out of 650 tasks, 120 cases have been initiated. In terms of number of cases, approximately 20%, and in terms of work-hours, approximately 35% business impact
• Approximately 4,000 onboarded tables and 200 created skills (more in development and testing)

• Snowflake CoCo usage rate went from 10% to 60% in April. Per-person usage volume is 7 times that of March

Governance: Rogue tasks disappear

• Often asked "are business controls okay?", but the answer is the opposite. Rather than rogue EUC increasing, rogue tasks disappear
• Traditional EUC centered on local Excel files and code that couldn't be controlled, but doing it on Snowflake makes everything observable. Since the substance of tasks is skills (natural language), it doubles as a manual as-is

• Introducing an auto-generated report from Snowflake CoCo
  • Evaluates a department's task coverage. Lists the trend of active members, each individual's usage volume, and created objects
  • Evaluates each task using criteria such as "established," "function exists but not established," "temporarily spreading," and "data only," showing where the problem lies — in function, adoption rate, or data
  • Also outputs team-by-team and individual-level status and improvement suggestions. Specifying "summarize for a presentation" produces a presentation-ready version instantly
• The reason EUC control was difficult was not because creating was free and easy, but because what was created could not be observed

Commitment to business outcomes

• The questions "what results will AI produce?" and "will it really increase revenue?" are fundamentally backwards
• The right questions are setting goals such as "I want to double sales touchpoints" and "I want to halve the speed of product launch," then thinking together about whether AI can accomplish that. An offensive stance of agreeing on KPIs with business impact with the field and using AI to achieve them is necessary
• Mid-term management plan, full task list, objects, and usage logs are connected to understand feasibility toward goal achievement and serve as a baseline for discussion

Declining cost of problem discovery and loop engineering

• As more business data is gathered, more tasks can be covered, and as regulations and manuals are gathered, requirements become fully defined. Natural language query logs become a wish list of "what we want to do next" as-is

• What comes next is loop engineering: a mechanism where systems autonomously and recursively continue to improve. What is needed is problem discovery capability and problem resolution capability
• Three technical elements for safe and reliable operation
  • Microservices architecture: Prevents auto-generated artifacts from affecting other components
  • Harness engineering: Supports safe and reliable operation
  • Business semantics engineering: Snowflake's comment function, declaration of data ownership, behavioral norms through skills, ontology via knowledge graphs and Cortex Sense

• His favorite quote is from Netflix's Reed Hastings: "Lead with context, not control." In the LLM era, if you can give an LLM the right information, the organization becomes smarter

Q. How do you run the problem discovery loop?

• Running two loops to improve efficiency per credit
  • For individuals: Automatically sends a daily email saying "here's how you can do even better" based on usage information, and traces the effect
  • For the overall system: Automatically proposes necessary hooks, controls, and context to give to objects, and implements them across the entire system
• Loops come in large, medium, and small, and it is best to start with small loops and work up to larger ones. All of this is realized in natural language

Q. How have development speed, cost, and the role of people changed through the use of skills?

• If asked to "build a search function," normally you'd build a search function, but with agents available today, that's no longer enough. You need to understand the motivation behind wanting to search and how the data will be used after searching, and then entrust the entire flow to an agent
• Introducing thin-sliced features risks becoming a bottleneck for overall throughput. This shift in mindset is the biggest change brought about by agent AI
• The human role becomes more business-oriented, and committing to business value becomes important

Session by CyberAgent's Ueno and Takao

History of AI utilization and the AI Operations Division

• With the vision of "creating a company that represents the 21st century," three pillars of business are internet advertising, media (ABEMA), and games

• AI Lab research organization was launched in 2016 (now with over 100 members), providing products such as advertising effectiveness prediction and creative generation
• When ChatGPT emerged in 2022, they recognized that "AI is not just for some engineers," and established the AI Operations Division in 2023. An AI-driven development promotion division for development organizations was also established in 2025
• Steps for company-wide rollout

• 2023: Held a contest to gather ideas for AI utilization in operations so that employees across each business could take ownership, along with reskilling on security and governance
• A one-day event where each business division brings its case studies to show off is held once a year

• What they are putting the most effort into recently is the AI Ranking. Modeled after sumo rankings, business leaders are ranked by their degree of AI utilization. It's an initiative for everyone to think about connecting AI to business value rather than just efficiency

• Characteristics of the AI Operations Division
  • A company-wide function not tied to any specific business. Approximately 20 dedicated members centered on developers and product managers
  • Business heads from each business division serve concurrently, enabling alignment of AI perspectives and messaging across the company
  • Rather than being a top-down decision-making company, they value freedom and self-responsibility in each business division
• Snowflake has been adopted for 5 years in internal systems for ad delivery data. Adding management information and other data is easy, and it is compatible with security reliability and machine learning features

• Collaborating with the AI-driven development promotion division, group IT promotion, and technology leadership to prepare an environment where everyone company-wide can use AI

Q. What have been the results of AI utilization?

• Not just engineers — nearly 4,000 employees company-wide are utilizing tools like ChatGPT and Claude
• In development organizations, development output has approximately doubled. Significant reduction effects also in areas like spec document creation, which are in the domain of product managers and planners
• In sales departments, there are also cases where individual employees can handle more customers, leading to increased revenue. Cases directly linked to business outcomes are growing internally

Data infrastructure and MCP integration

• In charge of the group's IT and data. There is an environment where everyone can use ChatGPT and Claude, and everyone is proficient in using AI
• Data from all corners of the company is in Snowflake, and MCP is used to enable secure access to data
• Examples of use
  • Corporate information search
  • Safety confirmation system: Instantly collect information during a disaster and check the status of your own department
  • Cosmetics business: Machine learning for store data, contract agent, sentiment analysis
• No matter how high-performance the model, if data is siloed, it cannot access the right data and AI will make mistakes too. The fundamental principle of not siloing data is important
• Infrastructure uses AWS and GCP in roughly equal proportions. They have high expectations for the GCP Tokyo region announced this time, and want to advance integration with data on Google Cloud

Q. What is your medium to long-term vision?

• Aiming to be a leading company in the AI era
• In development organizations, they set a goal of full automation of the company-wide development process by 2028
• In human resource development, new careers called "Business Lead Engineer" and "AIX Designer" have been announced, with job definitions, evaluation, and education frameworks being established
• Existing businesses will optimize operations and outputs through collaboration with AI agents, while providing new value through new businesses unique to AI

Closing

• We are in the midst of the largest structural change of this generation in enterprise technology
• With Snowflake AI Data Cloud, you can transform autonomous AI into the ultimate competitive advantage by leveraging a governed foundation, connected systems, agents, and a control plane
• AI utilization in business has entered the phase of "how to do it," not "whether it can be done"

Impressions

I attended the KEYNOTE of SNOWFLAKE WORLD TOUR 2026 - TOKYO Day2.

I was deeply struck by the point raised in the talk by Sumitomo Mitsui Trust Asset Management — rather than asking "what results will AI produce?" or "will it really increase revenue?", the right questions are setting goals such as "I want to double sales touchpoints" or "I want to halve the speed of product launch," and thinking together about whether AI can accomplish that.

The major event in Japan, SNOWFLAKE WORLD TOUR, has concluded, but our journey of leveraging data and AI continues uninterrupted. I hope that by taking the knowledge gained at this event back to our organizations and applying it, it will contribute to business development.
I hope we can also support others through sharing information and providing assistance.

I hope this article serves as a useful reference for someone!