Quality assurance · Production

Mercari's AI-Native Company transformation: ASDD, Knowledge Management, and AI Task Force

The problem

Mercari's initial AI Coding Assistant rollout did not deliver the expected productivity gains. Prompt quality varied widely between engineers due to no shared regulation, decision-making context was scattered across Slack, GitHub, and meeting notes, and generated code quality was inconsistent across teams.

First attempt

When AI Coding Assistants were first deployed company-wide with Cursor, in some cases repeated instructions made the process slower than working without AI. Context needed for AI agents was not collected correctly. When Claude Code and other tools subsequently emerged, engineers began using different tools, making it impossible to consolidate best practices.

Workflow diagram · grounded in source
1
Knowledge centralized in Notion
integration
“散在している情報を一つの文書管理基盤に集約し、必要なときにコンテクストを簡単に取り出せる状態にすること”
2
Agent generates implementation plan
ai_action
“メルカリのナレッジ基盤に最適化されたエージェントが自動的に一次情報にアクセスし、詳細な実装計画を生成します”
3
Validation agent checks plan
validation
“別のエージェントが、実装計画がサービスのコーディング規約に沿っているか、計画が所定のセキュリティ観点をクリアしているかなどのさまざまな調査を行います”
4
Agents iterate revisions
feedback_loop
“この2つのエージェントが交互に修正と評価を繰り返し”
5
Developer consulted on uncertainty
human_review
“最後に要件や仕様における不確実要素が残った場合には、開発者に追加の質問を行います”
Reported outcome

Mercari achieved 95% employee AI tool adoption, approximately 70% of code generation handled by AI, and a 64% year-over-year improvement in development speed.
The company launched ASDD as a standardized AI-driven development process and established Notion as a central knowledge base. A 100-person AI Task Force now spans 33 business domains with approximately 4,000 workflows catalogued for AI transformation.

Reported metrics
employee AI tool adoption95%
code generation handled by AI約70%
Development speed improvement year-over-year64%
engineers using AI Coding Assistantほぼ100%
Show all 8 reported metrics
employee AI tool adoption95%
code generation handled by AI約70%
development speed improvement year-over-year64%
engineers using AI Coding Assistantほぼ100%
code generation volume increase60%以上
AI Task Force members約100名
business domains in AI Task Force33
workflows catalogued約4,000
Reported stack
NotionCursorClaude Code
Source
https://engineering.mercari.com/blog/entry/20251225-mercari-ai-native-company/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Mercari achieved 95% employee AI tool adoption, approximately 70% of code generation handled by AI, and a 64% year-over-year improvement in development speed.

What tools did this team use?

Notion, Cursor, Claude Code.

What results were reported?

employee AI tool adoption: 95%; code generation handled by AI: 約70%; Development speed improvement year-over-year: 64%; engineers using AI Coding Assistant: ほぼ100% (source-reported, not independently verified).

What failed first in this deployment?

When AI Coding Assistants were first deployed company-wide with Cursor, in some cases repeated instructions made the process slower than working without AI.

How is this quality assurance AI workflow structured?

Knowledge centralized in Notion → Agent generates implementation plan → Validation agent checks plan → Agents iterate revisions → Developer consulted on uncertainty.