In 2008, Masayoshi Son brought the iPhone to Japan through an exclusive SoftBank deal. That was the starting point for the Japanese IT industry’s mobile transition. In 2026, DeNA revealed the results of going “AI All-In” for one year. The scale of numbers and depth of organizational transformation suggest this is not just a tool adoption — it marks a new inflection point for the Japanese IT industry.
#The AI All-In Declaration: One Year of Results
In February 2025, DeNA declared it was going “AI All-In.” The plan was ambitious: maintain and grow its existing businesses — operated by roughly 3,000 people — with half the headcount, while reorganizing the other half into small 10-person squads to rapidly launch new ventures.
One year later, at DeNA × AI Day 2026 on March 6, 2026, Chairwoman Tomoko Namba revealed the results.
- 20x development productivity — Depending on the project, AI handles 95% of the work while humans contribute just 5%. Hands-on coding decreased significantly after the release of Claude Opus 4.5 in November 2025.
- 90% reduction in legal contract review effort
- 50% reduction in QA effort
- 60% reduction in Pococha live-stream moderation costs
The key insight is that DeNA did not simply layer AI onto existing workflows. Instead, it redesigned the workflows themselves with AI as the starting assumption.
DeNA calls this initiative “AI 100 Drills” — a term borrowed from baseball, where a fielder catches 100 ground balls in a row during practice. The idea is that skills are forged through relentless repetition. DeNA publicly shared over 100 AI use cases that emerged bottom-up from the front lines.
#Three Stages of AI Adoption
Namba outlined three stages in the evolution of AI adoption.
- Stage 1: Prompt Engineering — The stage of asking AI good questions. One-off requests like “Review this code” or “Summarize this contract” fall here. A year ago, this alone was enough. Most companies are still at this stage.
- Stage 2: Context Engineering — The stage of providing AI with relevant background information. Using techniques like RAG to feed project documents, codebases, and internal policies into the model, improving answer accuracy.
- Stage 3: Environment Engineering — As AI agents began retrieving information on their own, designing guardrails became the core challenge. Humans decide the boundaries: what the agent can access, what actions it can take, and what outputs it can produce. DeNA has reached this stage.
A concrete example of Stage 3 in action: DeNA’s IT department registered an AI agent called “Lemon-kun” as an employee, granting it access to internal wikis and calendars while requiring human approval for critical actions. The approach treats AI not as a tool but as a team member being onboarded into the organization. Namba herself reportedly interacts with this agent daily on Slack to carry out her work.
#The Efficiency Paradox — “Nobody Reports Having Spare Capacity”
On paper, this is a flawless success. But one part of the plan did not go as expected. While AI freed up working hours, employees filled the gap with tasks they had always wanted to do but never had time for. When time opens up, people naturally reach for the backlog they had been putting off. As a result, not a single person reported, “I have spare capacity.” Efficiency improved, but the original goal of reallocating talent to new ventures progressed more slowly than anticipated.
Namba’s solution was blunt: “Sometimes you need to lead with force.” Rather than waiting for the front lines to adjust organically, the approach flips the sequence — move people to new ventures first via top-down decisions, then backfill the gaps with AI. She also announced that manager evaluations would incorporate a new metric: how many people they successfully transitioned to new business initiatives.
#Building Unicorns with 10-Person Squads
The core of DeNA’s new-venture strategy is to focus on the application layer. Instead of developing foundation models (LLMs) in-house, the company competes through domain depth built on top of those models.
Namba warns that “mediocre expertise will be wiped out overnight.” In an environment where OpenAI and Google are rapidly raising general-purpose capabilities, survival requires combining proprietary data with deep domain knowledge.
To execute this strategy, DeNA established a dedicated subsidiary called DeNA AI Link, specializing in AI agent ventures. The company also became the first in Japan to deploy Cognition AI’s autonomous coding agent Devin company-wide.
Even after the iPhone arrived in Japan in 2008, the mobile transformation took years. Tools change overnight, but reshaping an organization to match takes time. DeNA’s numbers have proven AI’s potential. The remaining question is how fast the room created by efficiency gains gets redirected somewhere new.