Upgrade to Pro — share decks privately, control downloads, hide ads and more …

【mercari GEARS 2025】Keynote

Avatar for mercari mercari PRO
November 13, 2025

【mercari GEARS 2025】Keynote

Avatar for mercari

mercari PRO

November 13, 2025
Tweet

More Decks by mercari

Other Decks in Technology

Transcript

  1. Shunya Kimura ・2017/9 establishing R4D, Mercari’s research ・2022/7 Vice President

    of Platform Engineering ・2024/7 Vice President, Chief Technology Officer (Japan Region), General Manager (Japan Region Platform) ・2025/3 Vice President, Chief Technology Officer (Japan Region) ・2025/7 Vice President, CTO Japan Business Mercari / Vice President, CTO Japan Business
  2. Since 2018 to the Present of Mercari Group • 出来事

    ◦ 2017年11月:メルペイ設立 ◦ 2018年10月:英語化(グローバル新卒メンバー 44名入社) ◦ 2019年2月:メルペイリリース ◦ 2021年4月:メルコイン設立 ◦ 2021年10月:メルカリShopsリリース ◦ 2023年3月:ビットコイン取引サービス提供開始 ◦ 2024年3月:メルカリ ハロリリース ◦ 2024年3月:メルカリモバイルリリース ◦ 2025年1月:メルカリNFTリリース ◦ 2025年9月:グローバルアプリリリース
  3. “We will fundamentally transform our organization and products, with AI

    as the foundation for everything.” – Shintaro Yamada CEO of Mercari
  4. Current Status: Solid Achievements 95% Engineer AI Tool Adoption rate

    ≒70% Code Generation by AI tool 64% UP Development Speed (YoY)
  5. But We're "Just Getting Started" However, we don’t consider this

    state to be “AI-Native” This is where the real transformation begins.
  6. Paradigm Shift “Human-Centric” to “AI-Centric” Human-Centric Based on human limitations

    (time, focus, processing power) • 8-hour workday • 2-day weekend • 7-person teams • weekly meetings AI-Centric Based on vision and value propositio Rebuilding Optimal: • systems • teams • roles based on value
  7. AI as a Creative Partner AI is a creative partner

    and a team member, driving us toward our mission. この画像はこだわりな い
  8. Why is “Context” is so important? AI Agent ? Low

    quality output Rework (time lost) Context AI Agent High quality output Effective utilization Without Context With Context background, decision logs, etc.
  9. The Most Important Context: Decision Information Without information like "What

    is the project's goal?" or "What is permissible?", AI accuracy won't improve. Slack Decision Making Information Docs JIRA Confluence
  10. Our Approach Discussion Minutes Structure Database Actionable context for AI

    AI Agent Stronger alignment between top-down and bottom-up Accelerated company-wide decision-making “Structuring Tacit Knowledge”
  11. AI-Centric Development: Current Challenges Inconsistent Prompt Quality Varying degrees of

    AI utilization among individuals Gathering Context Inability to find or provide the right information Quality of Generated Code Increased review costs and bug risks Lack of regulations for AI usage leads to inconsistent quality across teams and individuals.
  12. Project Double Agent-Spec Driven Development (ASDD) Enables anyone to achieve

    high-quality, reproducible development through specs optimized for AI Agents.
  13. What is ASDD? To make the entire team's implementation a

    reproducible process. API Definitions Data Models Process Flows Test Scenarios Implementation Steps Service Spec Dependencies Architecture Prompts
  14. Scope of ASDD ASDD: Orchestrating End-to-End Development with AI Agents

    Devlopment • Backend • Web • Mobile • UI Design Quality Assurance • Auto QA • Test case • AI Review GRC • Risk • Compliance • Legal Review • Security Check • CS
  15. Overview of ASDD Assets PRD Figma UI Codebaase Double Planning

    Requirements Short Agent Spec Double QA Test Cases PJ Aurora UI Double Client Agent Spec (Client & FE) Implementation (Client & FE) Double Backend Agent Spec (Backend) Implementation (Backend) Double QA Test Implementation Double QA test Implementation Release PJ Double
  16. Project Aurora Automatically generate UI from Agent Specs ASDD Prompt

    Aurora Agent Auto-generated UI using our Design System
  17. The Value of ASDD Vibe Coding Agentic Development • Handle

    more tasks • Enable rapid trial and error • Focus on solving difficult problems • Focus on customer value
  18. The Ultimate Form of Customer-Centricity Being AI-Centric is the ultimate

    form of being Customer-Centric なにか画像が欲しい な。。。
  19. AI Task Force Drives Company-wide AI Transformation Challenge Speeding up

    the development process alone does not improve the overall time-to-release.
  20. AI Task Force We need to make all company-wide workflows

    agentic. Development [WAIT] Legal Review [WAIT] PR Review [WAIT] Security Review Release
  21. AI Task Force We need to make all company-wide workflows

    agentic. Development [AI Agent] Legal Review [AI Agent] PR Review [AI Agent] Security Review Release
  22. The Pioneer of Agentic Work: Project Socrates Request data retrieval

    and analysis using natural language Anyone can access data without complex SQL queries A company-wide realization that AI can perform difficult tasks, including decision-making.
  23. Launch of the AI Task Force Launched AI Task Force

    in July 2025 Accounting/Audit/Credit/Logistic/BI/IR/CorpDev/CoporateIT/CorporatePlanning/ CSTns/EngineeringOffice/Governance/HR/Kashima/Legal/Marketing/MS/PMO/P ublic Policy/R4D/Risk/Sales/Security/Treasury/ Corporate domain AI Product/Engineering/PdM/UIUX Service Development Domain
  24. AI Task Force: Responsibilities 3 Core Responsibilities Take inventory of

    all business processes Create AI transformation vision and roadmap Execute on AI transformation according to the roadmap 1 2 3
  25. Lessons from the AI Task Force Cross-Domain Synergy Engineers who

    moved into non-tech domains (e.g. Legal, Finance) came up with innovative ideas that led to improvements by challenging the existing assumptions. New Growth for Engineers Tackling unfamiliar domains led to new skill sets and boosted motivation.
  26. Case 1: Enhancing Fintech CS Quality with AI Role-Playing We

    used to conduct role-playing between people. Now, AI is our training partner. Practice Screen We simulate actual vocal conversation between the customer and the AI on the CRM. Evaluation Screen The AI provides feedback based on the conversation.
  27. Our Goal Beyond AI-Native Transformation Circulate all forms of value

    to unleash the potential in all people デザインむずい。アン リーシュ感出したい。
  28. The Challenge of Global Expansion Challenge Global expansion was limited

    by human resources for localization and operations Solution with AI Augmenting individual capabilities with agents to achieve global expansion with a realistic number of people.
  29. Moving away from a mindset of resource limitation to a

    world where we can achieve what we want to achieve
  30. Toward Ultimate Customer-Centricity By building on a foundation of AI,

    we can focus more deeply on customer value. This is the "AI-Native Company" we aim to be.
  31. Mercari MCP Application Layer Mercari MCP Mercari tools and apis

    - Authentication - Search - Listing - More coming soon End Users ChatGPT is a product of OpenAI, L.L.C. Claude Code is a product of Anthropic PBC. Claude Code
  32. Mercari MCP Roadmap Phases P1 P2 P3 Whitelist Only Full

    Release Expand Tools MCP accessible to applicants only. Validated applicants will be whitelisted Anyone can try with applications that support MCP Continue adding more tools and APIs (e.g. more search and listing capabilities, logistics, analytics, payments.)
  33. Signup Sign Up today, try Mercari MCP in ChatGPT and

    Claude Code. When it’s ready, we will inform you.