CyberAgent embeds Codex into design and mockup work, not just code generation

What CyberAgent actually reports

CyberAgent, a Japanese internet company spanning advertising, media and IP, and gaming, says ChatGPT Enterprise reached a 93% monthly active user rate across nearly all departments. The company frames AI as foundational rather than experimental, citing its AI Lab (established 2016) and AI Operations Office (launched 2023) as organizational commitments that predate the current tooling.

The security story is concrete. Before adopting ChatGPT Enterprise, employees hesitated because it was unclear what information could safely be entered, and usage varied unevenly across teams. Enterprise-grade access control, account management, and usage visibility, paired with internal guidelines on confidential information, gave staff a defined boundary to work inside.

Adoption spread without a mandate

CyberAgent explicitly does not enforce specific tools company-wide. Each team, department, and subsidiary evaluated ChatGPT Enterprise against alternatives on its own terms. That the tool still reached 93% monthly usage is the more interesting data point, because it was pulled in rather than pushed.

The mechanics behind that pull are worth naming. CyberAgent shares prompts and successful use cases, and built a private usage ranking so employees can see their own activity without it feeding evaluations. When someone goes idle, a Slack bot reaches out to ask whether they are using another AI tool or hitting obstacles. OpenAI ran more than ten training sessions of 100-plus participants each, from a 'ChatGPT Enterprise 101' introduction through custom GPT workshops, hands-on Codex sessions, and hackathons.

With enterprise features such as account management and visibility into usage, ChatGPT Enterprise made it possible to support business use of a wide range of information, excluding confidential data. As a result, the scope of AI use across the company has expanded, and many employees now use it in their daily work.Montana Labs

Codex moving upstream into design and frontend prototyping

The most specific claim in the announcement is that Codex is being used for far more than code generation. CyberAgent's Ken Takao lists three uses: pressure-testing design proposals from multiple perspectives, generating and selecting among options during code review, and maintaining knowledge documents like AGENTS.md so agents operate with richer context. The reported benefits are earlier alignment before implementation and clearer rationale behind proposals to speed decisions.

For frontend and product work, the interesting detail is that Codex is spreading to non-developer roles who use it for writing specifications, creating mockups, and structuring work adjacent to product and development. In the AI Business Division, Sou Yoshihara runs Codex through MCP in Cursor for the design and implementation planning of Kiwami Prediction AI, and describes its proposals as higher quality than other coding models. At GOODROID, Hidekazu Hora used Codex to develop the game WormEscape, which reached soft launch after roughly one month.

It felt like a reliable partner that supported the entire process from discussing implementation to carrying it out, helping increase development speed. With Codex, even in areas where I had no prior experience, I was able to resolve roadblocks caused by lack of knowledge more quickly, and I feel it helps achieve both quality and speed.Montana Labs

The implication: tooling that builds the scaffolding for its own adoption

The detail applied teams should sit with is that CyberAgent used Codex to build its internal usage ranking system, the same mechanism that makes ChatGPT adoption visible and drives further uptake. The tool was turned on the problem of its own spread.

That closes a useful loop. When a coding agent is capable enough to build the lightweight internal apps that support a rollout, and usable enough that non-developers reach for it to draft specs and mockups, the boundary between 'who ships software' and 'who uses AI' blurs. CyberAgent's frontend and product staff are producing prototypes without waiting on engineering queues, while the adoption program feeds itself with tooling that same capability produced. For teams weighing agentic coding tools, the lesson is less about raw speed and more about whether the tool can absorb the connective work — dashboards, follow-ups, mockups — that usually stalls a rollout.