This post, the final entry in CyberAgent’s 2025 developer Advent Calendar, is written by an engineer and creator in the AI Driven Promotion Office of CyberAgent who works on promoting AI use and developer productivity across the CyberAgent group. It is presented as a story about tool adoption, but the author says its real subject is how to “deliver” something so that an organization moves.
The argument is that in a large company, building or introducing a good tool is not enough for it to spread. The author reframes internal adoption as the go-to-market of an internal product and describes a strategy of creating attention outside the company so that it comes back inside. The centrepiece is an external event, AI Code Agents 祭り ～ 2025 Winter ～, and seven techniques combined around it.
Key Points
- The author’s starting point was being alone with Cursor in 2023, then creating a Slack channel dedicated to it on 11 September 2024 whose only member at first was the author; internal talks and demos drew little response.
- The post attributes the difficulty to scale and culture: more than 8,000 consolidated employees and 86 consolidated subsidiaries (as of the end of September 2025), separate Slack workspaces and independent technology choices per team, and a bottom-up culture in which a top-down “everyone uses this tool from tomorrow” rarely happens.
- In autumn 2024 the author decided to “blow wind in from outside” instead: information from outside — trends on X, YouTube, reputation in technical communities — is said to cross organizational walls and reach everyone at once. The author judged an external event to have the most leverage given the organization’s culture and the author’s own skills, while acknowledging guidelines, training and top-down policy as other valid means.
- Technique 1: send the first direct message. With no personal network, the author recruited influential speakers by cold direct messages that quote the recipient’s own work specifically, state the value offered in numbers, and make the request small and easy to answer; the post reports the speakers approached accepted and introductions followed.
- Technique 2: decide who it is for, broadly — a speaker mix of CyberAgent executives, technical influencers, outside practitioners, global perspectives, internal practitioners, non-engineers, young engineers and overseas CEOs, each chosen for how it spreads the message. The author singles out executives on stage as lifting the message inside the company.
- Technique 3: invest in design — a landing page built in Canva with AI-generated visuals and a three-metre poster — on the view that quality is a silent message that earns a hearing.
- Technique 4: remove barriers with AI, using an AI real-time interpretation service between English and Japanese for overseas speakers.
- Technique 5: design the experience so online viewers are the main audience, since over 1,300 of the pre-registrations were online; the post reports online satisfaction (4.5) slightly above on-site (4.3).
- Technique 6: recruit collaborators through a shared vision; speakers and organizers gave their time without payment or obligation.
- Technique 7: keep posting without a response. The Slack channel had over 500 members by the time of writing, CyberAgent had about 1,000 active Cursor users across job types, and Claude Code and Codex had been introduced on enterprise plans with Cursor under consideration.
- The post also lists what did not work: a first internal talk that leaned on technical explanation without a story, documentation alone, and the idea of one message for everyone — executives, veterans, young engineers and non-engineers are said to value different things.
- The author recommends re-evaluating AI tools about once a month, warns against writing a tool off on a first impression, and describes using several AI coding agents full-time as a multi-vendor, multi-agent approach.
Context
This is a first-person account by the person who ran the effort, published on the company’s own developers’ blog; the results it reports are its own figures, and the techniques are the author’s experience rather than a tested method. The post connects the effort to the company’s later stated aim of fully automating its development process by 2028 and to the AI Driven Promotion Office, which it describes as set up to help across the company with building an autonomous development structure in which engineers and AI agents work together. The user community that grew out of the event is AI Agent User Group (AIAU).