This post, by a manager in the AI Driven Promotion Office at CyberAgent, explains the Engineer AI Banzuke, a version of the company’s AI Banzuke built specifically for development organizations. The company-wide AI Banzuke ranks organizations’ AI use on a sumo-style scale; the author argues that viewpoints such as revenue, cost reduction and contribution to business KPIs cannot capture how mature a development organization’s AI use is, because an organization where individuals use generative AI in daily coding and one that has built AI agents into its whole development process are both “using AI” while facing very different next steps.
The post describes how the assessment was designed — two axes, fourteen items and seven ranks — how the first round assessed 44 development organizations through self-reporting, evidence and deliberation, what the operators observed, and what they intend to do before the second round.
Key Points
- The engineer version was designed as a shared language for questions such as how far AI is built into the development process, whether AI is an assistive tool or acts autonomously, whether roles between humans and AI are clear, and whether AI goals connect to technical strategy, in support of CyberAgent’s stated aim of automating the development processes across the company by 2028.
- It scores organizations on “engineering quality and technical maturity” and “promotion structure and technical culture”, because the author holds that both technically advanced use and the ability to spread it reproducibly across an organization are needed.
- Results are expressed as seven ranks named after sumo ranks, from makushita (use limited to individuals) to yokozuna (an AI-first development organization), which the author says made positions easier to share intuitively than a plain score.
- Organizations reported their own status and supplied supporting evidence, which the operators checked against the evaluation items, with a deliberation council settling cases where needed; the author says the aim was for every organization to understand why it received its rank and what to improve next.
- The first round showed that many development organizations had already begun experimenting with AI well beyond implementation and review support, while how to turn individual or team practices into the organization’s standard development process was a theme common to many of them.
- The author argues that AI adoption must move from individual productivity gains to redesigning the development process — revisiting how specifications, reviews, tests and knowledge sharing are done on the assumption that AI is present, rather than adding AI to the existing flow.
- For spreading practices, the operators intend to go beyond sharing case studies to explaining why a practice worked, what conditions make it reproducible and where to start small; the second round was planned for announcement in October 2026.
Context
The post is the operators’ own account of an internal programme at a single company, and its observations are qualitative. It complements CyberAgent’s separate description of the company-wide AI Banzuke, which it points to for the overall background, and sits alongside the company’s internal AI maturity levels model as another way CyberAgent describes AI-adoption maturity; the post itself does not relate the two. The author acknowledges that a ranking can generate enthusiasm but also risks the evaluation taking on a life of its own, and frames the programme as aimed at helping each organization take its next step.