[Report] SNOWFLAKE WORLD TOUR 2026 - TOKYO: Day 1 KEYNOTE #SWTTokyo26

I'm Kawabata.

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

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

This article is a report blog on the session "Day1 KEYNOTE".

※Some features are in preview or scheduled to be available soon. Please check the latest official documentation for availability.
※The speakers' remarks are summarized based on the author's notes. The parts delivered in English are the author's free translation.

Speakers

• Christian Kleinerman
  • EVP of Product, Snowflake
• Ryuji Ukita
  • President and Executive Officer, Snowflake
• Shizuka Hirakawa
  • Executive Officer, Nissan Motor Co., Ltd.
    Chief Digital Information Officer (CDIO)
• Seigo Sugimoto
  • Fellow, FUJIFILM Corporation
• Natalie Mead
  • Head of Applied AI, APAC, Anthropic
• Sho Tanaka
  • Snowflake Product, Developer Experience
    Lead Developer Advocate

Starting with Kumataro's drum performance

I think this performance instantly raised the energy in the venue.

From Snowflake's Mr. Ukita

• Snowflake World Tour is held in more than 20 cities worldwide. Tokyo is the only city in the world with a two-day event
• This year the scale was expanded from last year, with more than 12,000 registrations, 144 sessions, and 202 speakers
• They want this to be a place where how customers and partners use Snowflake is delivered not by Snowflake, but by the customers themselves
• The partner sessions and the Expo area have many hints about new technologies, services, and business
• DATA DRIVERS AWARDS 2026 was already announced in June. The award-winning company FUJIFILM will speak later in the KEYNOTE

• The Partner Awards are announced today. The award ceremony will be held tomorrow
• With the evolution of generative AI and AI agents, we have entered an era in which executives and the finance, HR, sales, and marketing departments access data through AI
• A very large number of companies are focusing on data preparation in order to get accurate output from AI
• In Japan, about 70% of IT engineers belong to partner companies, so partners are indispensable for data preparation and AI utilization

• User community activity is also growing, and the membership has exceeded 2,500

• 84.2% of customers using Snowflake are using some kind of Snowflake AI service, and results are starting to appear

• Customers' interests are "building a data foundation to fully leverage AI," "security and governance," and "cost management." Today's KEYNOTE proceeds along these themes

From Snowflake's Christian Kleinerman: The four elements of the Agentic Enterprise

• We are now in the era of the Agentic Enterprise. This refers to an organization that uses AI in its core operations, where agents handle routine tasks and people spend their time on setting direction and making decisions

• "Introducing AI will solve data problems" is a misconception. It is AI that needs an excellent data strategy and data assets

• A chaotic data environment, with silos, a mix of streaming and batch, and multiple copies, limits the results of AI

• The four elements needed for the Agentic Enterprise

  • Element 1: AI models
  • Element 2: Data and context. Needed for AI to understand your company
  • Element 3: Software and applications. Integration with enterprise infrastructure such as email, messaging, and documents
  • Element 4: Control plane. The foundation that ties the above three together so that AI understands the organization's expectations and processes and delivers results

• Snowflake AI Data Cloud provides solutions for all four of these elements

• He introduced customers who are already getting results, such as Thomson Reuters (using AI on sensitive, highly regulated data), Samsung, and Under Armour (automating customer and partner handling processes)
• This time, they reversed the usual KEYNOTE flow, showing the demo first and then explaining each element

Demo: Meeting the demands of the CEO and CFO with Snowflake CoCo and Snowflake CoWork

This is a demo by Sho Tanaka of Snowflake G.K.

Scenario

• The subject is a fictional sports entertainment company, "Snow League"
  

• CEO John: VIP upgrade sales exceeded the forecast by 15%, so he wants to raise the targets for the rest of the fiscal year

• CFO Sarah: Before raising the targets, she wants to check costs and margins

• Sales representative Victoria: She is caught between the two. However, data is siloed by system and the definition of revenue differs by department, so the numbers don't match

• Data engineer Dan: After Victoria consulted him saying "the numbers on the dashboard are different every time," he began preparing the data foundation a week ago

The finished product: Streamlit dashboard

• With Streamlit on Snowflake, revenue, profit, number of VIP upgrades, and top cities are displayed on a single screen
• It analyzes video of the venue, scores congestion on a 10-point scale, and estimates the expected number of attendees
• It handles geospatial information natively and visualizes positive and negative posts on social media (X, TikTok, Facebook) around Tokyo on a map in real time

Data engineer (Dan) part: Snowflake CoCo

• Data is stored in Snowflake managed Iceberg Tables. Unstructured data such as images, video, and audio is also covered
• Data from Salesforce, SAP, and Workday is integrated through zero-copy connectors without writing ETL, and managed in Horizon Catalog
• He gives Snowflake CoCo a "skill" that defines governance and asks it to "protect my data with this skill"
  • It identifies personal information from the schema, sets masking policies and tags, and monitors logs
  • He confirmed that administrators get plain text, while roles without permission get masked values
• With the skill catalog, skills can be shared within the same account. Routine weekly and monthly tasks can be rolled out as company-wide knowledge
• Revenue definitions that differ by department are unified with a semantic view. Agents can refer to the definitions of revenue and LTV

Sales representative (Victoria) part: Snowflake CoWork

• Even without coding skills, you can analyze by conversing with data in Snowflake CoWork
• Agents are provided for different purposes, such as match briefing, VIP concierge, and fan support
• When asked "compare labor costs and revenue by city and tell me how the profitability is," it presents an insight along with a chart that "Yokohama is the highest"
• With MCP connectors, it integrates with email and Slack, and an automatic delivery at 6 a.m. every morning is set up with a single button
• The dashboard shared by Dan is referenced as is in Snowflake CoWork, and she digs deeper into the situation where VIP upgrades are increasing in Yokohama
• She has Deep Research propose "the next three actions to increase revenue further" and sends the results to Slack
• An iOS app is also available, so the same screen can be viewed on an iPhone (Android is also supported)

From Snowflake's Christian Kleinerman: Details of the four elements

Element 1: Enterprise data and context

• Since launching the product 12 years ago, Snowflake has worked on eliminating data silos and on governance. AI has made that need even greater

• Horizon Catalog
  • Records all information about systems and data and tracks it over time
  • With the Iceberg REST Catalog API, it can be used as a single catalog from other engines such as Spark and Trino
  • Settings for security, sensitivity, tags, column masking, and row filters are all within Horizon Catalog
  • Metadata from applications such as SAP and Workday, databases such as PostgreSQL, and Tableau, Airflow, dbt, and others is also covered

• Horizon Context
  • Accesses metadata inside and outside Snowflake to help understand, enrich, and utilize data
  • It can be used not only from Snowflake CoCo and Snowflake CoWork, but also from custom-built agents, Claude Code, Claude Cowork, and BI tools
  • It is the "glue" that connects AI models with your own data. Rather than a generic chatbot, it can behave like an AI built for your company
• Agent security

• Agent Identity: Restrictions such as read-only can be given to agents
• Data policies for agents: controls such as humans can output to a stage, but agents cannot move data outside the security boundary. Data leak detection is also possible
• Multi-party Approvals: Approval from two people can be required for sensitive changes
• Cost management
  • The question has shifted from "should we do AI?" to "it became expensive once we put it in production. How do we control it?"
  • What should be reduced is AI spending that doesn't generate value. Visibility by user and workload, budgets, per-user quotas, and notifications or stops upon overrun are provided
• Trust Center
  • Rather than telling you "you can set up MFA," it tells you "MFA is not being used on critical administrator accounts" — a check and verification of the operational status

Element 2: AI and model choice

• The answer to "the best model" differs between two years ago, one year ago, and this week

• Snowflake validates the most capable models and continues to provide them within Snowflake's security boundary. If data is sent to external endpoints, governance efforts are wasted

• You don't always need the best model. Snowflake helps with selecting the optimal model for the task, policies and controls for model use, and understanding costs

Element 3: Software and applications

• Zero-copy partnerships: Bidirectional integration with systems of record such as SAP, Workday, Salesforce, Veeva, and IBM. Mainframe data can also be used with AI

• Cortex AI Gateway will soon be in public preview
  • Captures all AI model calls at a single endpoint and provides policy, visibility, governance, and routing
  • Understands the content of requests and automatically routes them to the optimal model. There's no need to always use a top-tier model like Opus 4.8
  • Provides 100+ MCP connectors as Tools out of the box (Salesforce, SharePoint, Slack, Microsoft Teams, etc.)

• Cortex Sense: The AI itself observes data, queries, and user behavior in the background and generates metadata, semantics, and ontologies

Element 4: Control plane

• Snowflake CoWork: For non-technical users. It prioritizes accuracy and verifiability above all, aiming to be something you can put in front of a CFO or a head of marketing
• Snowflake CoCo: For data teams. It makes data engineers and data scientists 5 to 10 times more effective
• What Claude Code has done for software development, Snowflake CoCo will do for data

Talk by Mr. Hirakawa of Nissan Motor

• The Global Information Systems department has about 1,200 people. Centered in Japan and together with development members in the US, Europe, and India, it supports about 2,700 systems and 120,000 users
  

• The latest version of the IT department's mid-term strategy is "AI for Data / Data for AI"
  

• The journey with Snowflake

  • More than 10 years ago: struggled to ingest vehicle probe data on-premises
  • 2019: began full-scale use even before Snowflake's Japan entity was established
  • Afterwards, fully shut down on-premises and moved to the cloud. Recently began using generative AI

• Challenges at the time, as explained to management
  • Data managed by the IS department has high barriers in terms of usage requests and access rights
  • Business departments hoard data
  • Tacit knowledge at factories and development sites hasn't even been put into words. Excel and PowerPoint files proliferate in the office
  • At this rate, no matter what AI is introduced, it will be garbage in, garbage out

• What was gained from the shift to cloud native
  • Pay-as-you-go, zero initial cost, freedom from infrastructure operations. A common cloud environment across the three regions of Japan, the US, and Europe
  • Graduating from "box work" and moving to a phase of treating data as a company asset

• Structured, unstructured, factory, IoT, and vehicle data are consolidated into the "Nissan Data" platform

• The current challenge is to maximize the potential of data at minimal cost with data mesh and AI. Copies are minimized with Data Sharing and Iceberg
  
  
• Generative AI is the true democratization of data utilization
  • People who want to use data can build things themselves in natural language
  • As a side effect, business departments have started to notice things like "this data isn't included" and "the way it's being collected is wrong"
  • Management has also noticed the value of data and has begun demanding "faster and more detailed"

• This April, an AI Center was established directly under the CEO, led by Mr. Hirakawa. The IT department moves from service provider to driver of transformation
• For Nissan, AI and data are the "engine of transformation"

Q. What are the strengths of Snowflake that you notice precisely because you have been using it since 2019?

• Even with about 5,000 datasets, over 500 use cases, and more than 10,000 users using it daily, we hear no complaints of it being "slow" or "getting expensive"
• Queries became about 20 times faster without us noticing, and the business departments thanked us
• While some vendors "don't feed the fish they've caught," Snowflake takes good care of its users

Q. How has the way the accumulated data is used changed because of AI?

• Previously, data engineering to understand and process data and build marts cost a lot
• Now it can be accessed in natural language, and the business departments' level of understanding and speed are completely different
• The IT department has also greatly accelerated data engineering with Snowflake CoCo.

Conversation with Mr. Sugimoto of FUJIFILM

FUJIFILM uses Snowflake as its data foundation, and this session focused mainly on its approach to governance and rollout.

Q. How important is it to put "trust" first as a prerequisite for deploying AI?

• The theme is "Conquering the double-edged sword." Generative AI is powerful, but it carries risks of hallucination, intellectual property infringement, and high bills
• There are three kinds of trust: trust in oneself, trust in the environment, and trust as seen by others

• Governance is not a brake, but offensive guardrails that let all employees press the accelerator to the fullest

• Uncontrolled AI adoption is like everyone driving cars without brakes or fuel-inefficient cars on public roads. We want to prepare a safe circuit, machines that are fuel-efficient with working brakes, and a dedicated pit crew (AI experts)
• Precisely because we run diversified businesses globally, a foundation like Snowflake lets us use data safely across regions and businesses

Q. What is the structure for rolling out AI internally?

• Building an AI ecosystem that all employees can use with confidence, along three axes: environment, organization, and people
  • On top of the global security foundation, we provide a safe and easy-to-use generative AI environment and data utilization foundation
  • An AI CoE has been set up as the control tower for operations and hands-on support
  • Working as one team with the legal and ESG departments from the start, we design guidelines and approval logic
  • Level-based training and internal competitions make AI use personally relevant

Q. How do you balance global standardization and flexibility?

• We think in two layers. Common parts that are lifelines, such as zero-trust networks and endpoint security, are 100% standardized without exception

• For the variable parts, we emphasize the characteristics of each region and business, and speed. Snowflake's data foundation and AI environment are set up as self-service environments that each business and regional company can quickly try out

Q. What do you emphasize so that employees don't feel constrained?

• Governance is not for stopping the front lines, but proactive environment preparation so that the front lines can run at full speed
• We absolutely avoid a situation where the company's governance capability can't keep up with the speed of AI adoption
• Deploy attractive options first, decide review criteria together with the front lines, and automate the governance of applications, visualization, and risk prevention
• We want the trusted Snowflake environment to be a safe guardrail like Nissan's ProPILOT 2.0

Conversation between Snowflake's Christian Kleinerman and Anthropic's Natalie Mead

• Natalie is a former Snowflake employee, and this is her fourth time attending Snowflake World Tour Tokyo. She currently serves as Head of Applied AI, APAC at Anthropic

Q. What kind of role is Applied AI?

• It's a role close to a solutions architect, with three pillars
  • Making employees smarter with Claude Code and Claude Cowork
  • Speeding up the most difficult code migrations, such as mainframes
  • Applied AI engineers build agents and support everything from evaluation (evals) to production migration

Q. What has changed compared to a year ago?

• A year ago everyone was exploring how to use Claude Code. A little over a year since the arrival of code assistants, every month now feels like a year
• Claude Fable can work autonomously for about 12 hours. Give it a task and leave it, and hours later (sometimes minutes later) the deliverable is done
• Beyond coding, with Claude Cowork knowledge workers can ask it to "build a dashboard for our top clients"
• Internally, a "chief of staff" agent that runs every morning tells them about meetings, missed follow-ups, and travel arrangements (even visa guidance)

Q. How should we choose a model?

• Everyone wants Fable, but Fable is for the most difficult workloads. For non-technical users, Sonnet is often enough
• It's significant that Snowflake has released a mechanism that automatically performs model routing
• Models keep getting better and costs go down. Using nuance, like Opus 4.8's low effort / high effort, changes the game

Q. How do you approach responsible AI?

• Under Constitutional AI, Anthropic deals with child safety, biohazards, and cybersecurity every day
• It has no roadmap and moves week by week, but it also places emphasis on safety, governance, and trust (for example, in regulated industries where data can't leave the country)
• 90% of Anthropic's code is written by Claude. It is released after passing through human review and safety guardrails
• Recommendation for general enterprises: think about role-based access policies and data connections the same way you think about your own business. Don't leave everything to agents; put guardrails in place
• The biggest fear regarding AI is not using AI. Teams are moving from prompt engineers to harness engineers

Q. How do the two companies complement each other?

• The partnership with Anthropic dates back to the founding days. Both companies share a focus on accuracy, reliability, safety, and responsibility
• Anthropic is a Snowflake customer, and Snowflake is also a Claude customer. They bring out the best in each other
• Examples introduced
  • Combining Anthropic's models with Snowflake CoCo and Snowflake CoWork to build an AI-native operating model across the company
  • Decision-making, incident monitoring, and security monitoring are done with data on Snowflake, and skills and connectors carry out actions as well
  • One of the closest examples of an Agentic Enterprise

Q. A message to attendees

• Just start right now. Snowflake CoCo, Snowflake CoWork, Claude Code, Claude chat — anywhere is fine
• Try throwing your most difficult workload at it. It's not hard, and AI helps you decide the best solution
• AI is an assistant that makes your work faster

Final words from Christian Kleinerman: Three homework assignments

• Make your data secure: modernize, eliminate silos, and ensure appropriate security
• Make your data AI-ready: have the right semantics and ontologies. Cortex Sense does most of the work on your behalf
• Find use cases, build agents, and start autonomous activity

Impressions

I attended the KEYNOTE on Day1 of SNOWFLAKE WORLD TOUR TOKYO 2026.

What impressed me most was that the message "it is not that introducing AI solves data problems, but that AI is what needs an excellent data foundation" was shared by all the speakers from Snowflake, Nissan Motor, FUJIFILM, and Anthropic. New features such as Horizon Context, Cortex AI Gateway, and Cortex Sense are all meant to "give AI the right context and guardrails." I felt this was a KEYNOTE where the four elements of the Agentic Enterprise shown at Summit 2026 were made concrete together with Japanese customer case studies!

I'm looking forward to the Day2 sessions as well.

I hope this article is helpful to someone!