AR Advanced Technology (ARI) is redesigning workflows around AI, a shift it calls “AI-native transformation.” Executive Officer Takuya Yamaoka said AI makes development more efficient and reduces work hours, but traditional pricing based on labor hours also lowers project fees. The interview took place on October 7, 2026.

Founded in 2010, ARI has a workforce in which more than 80% are engineers or consultants. Yamaoka saw a turning point around November 2024, when he felt coding agents had reached a level of accuracy suitable for commercial use. The company formed a specialist team in 2025 to develop use cases, but faced challenges connecting technical interest to business reform and managing the organization. It later gave all employees usage guidelines and licenses and introduced a usage ranking. Use outside work also grew, and ARI exhausted its annual AI budget around the month after companywide rollout. After finding that usage volume did not match business results, it built a gateway and dashboard to track who used which model, for what purpose, and at what cost.

During tests connecting AI to internal systems, ARI found that information visible in a browser could differ from what was available through an API. Restrictions on actions a person can take through the screen alone do not necessarily control AI access as intended. ARI placed a gateway between AI agents and internal systems to handle authentication, permissions, operation records, audit logs, and connection-method controls. The company initially put detailed prohibitions and restrictions in Claude Code’s “CLAUDE.md” settings file, but Yamaoka said some constraints could obstruct work. Its current approach keeps necessary prohibitions while communicating goals and decision priorities instead of prescribing every step.

ARI aims to redesign work around AI, from individual tasks to companywide workflows, rather than stopping at individual use. In one sales-work example, AI checks Outlook appointments at 6 a.m. and prepares materials for the day using past meeting records and related information in SharePoint and other sources. After a meeting, when its recording is placed in a designated folder, AI drafts meeting minutes, a thank-you email, and materials assigned as follow-up tasks. A person reviews emails before sending them. Yamaoka said the company also needs to organize work knowledge at the individual, project, and company levels and define what counts as completion for tasks delegated to AI. In development, that includes preparing acceptance criteria and tests.

ARI’s assessment of human-AI collaboration emphasizes the work environment, how specifications are communicated, review of suggestions, and corrections when work goes off track, because it is difficult to measure precisely how token consumption relates to productivity. The company can make users, models, and token counts visible, but a way to score collaboration quality remains under development. The gateway does not automatically detect when work deviates; people currently check progress and make adjustments as needed.

ARI also uses AI to modernize older systems. It gathers existing system behavior through browser interactions, creates reproduction tests, then has AI build a system with upgraded operating system, middleware, and language versions while preserving the specifications, making corrections until the tests pass. People handle production infrastructure and reviews. Yamaoka said this can greatly reduce development hours and allow larger projects to be delivered in relatively shorter periods. But as efficiency reduces work time, project scope also shrinks, creating a problem for labor-hour pricing: efficiency gains are absorbed into lower fees.

ARI is considering payment tied to results and charges based on usage of its own AI platform and applications, but no outcome-based projects had begun at the time of the interview. In addition to system development, it has started offering an end-to-end service covering workflow assessment and redesign, safe connections to existing systems through a gateway, and agent development and evaluation. Because companywide transformation at once is difficult, ARI starts with test environments or a single department or workflow; it is supporting invoice-processing automation and expense claims through AI chat. Yamaoka cited overestimated security risks, inconsistent workflows, unclear responsibility between departments, and insufficient implementation capacity as reasons AI adoption stalls. He forecast that SaaS companies would move toward AI-native operations within six months, followed by large companies, with the shift spreading to other businesses over 2–3 years. He said ARI aims to become a recognized player “in these 2–3 years.”

AR Advanced Technology (ARI) is redesigning workflows around AI, a shift it calls “AI-native transformation.” Executive Officer Takuya Yamaoka said AI makes development more efficient and reduces work hours, but traditional pricing based on labor hours also lowers project fees. The interview took place on October 7, 2026.

Founded in 2010, ARI has a workforce in which more than 80% are engineers or consultants. Yamaoka saw a turning point around November 2024, when he felt coding agents had reached a level of accuracy suitable for commercial use. The company formed a specialist team in 2025 to develop use cases, but faced challenges connecting technical interest to business reform and managing the organization. It later gave all employees usage guidelines and licenses and introduced a usage ranking. Use outside work also grew, and ARI exhausted its annual AI budget around the month after companywide rollout. After finding that usage volume did not match business results, it built a gateway and dashboard to track who used which model, for what purpose, and at what cost.

During tests connecting AI to internal systems, ARI found that information visible in a browser could differ from what was available through an API. Restrictions on actions a person can take through the screen alone do not necessarily control AI access as intended. ARI placed a gateway between AI agents and internal systems to handle authentication, permissions, operation records, audit logs, and connection-method controls. The company initially put detailed prohibitions and restrictions in Claude Code’s “CLAUDE.md” settings file, but Yamaoka said some constraints could obstruct work. Its current approach keeps necessary prohibitions while communicating goals and decision priorities instead of prescribing every step.

ARI aims to redesign work around AI, from individual tasks to companywide workflows, rather than stopping at individual use. In one sales-work example, AI checks Outlook appointments at 6 a.m. and prepares materials for the day using past meeting records and related information in SharePoint and other sources. After a meeting, when its recording is placed in a designated folder, AI drafts meeting minutes, a thank-you email, and materials assigned as follow-up tasks. A person reviews emails before sending them. Yamaoka said the company also needs to organize work knowledge at the individual, project, and company levels and define what counts as completion for tasks delegated to AI. In development, that includes preparing acceptance criteria and tests.

ARI’s assessment of human-AI collaboration emphasizes the work environment, how specifications are communicated, review of suggestions, and corrections when work goes off track, because it is difficult to measure precisely how token consumption relates to productivity. The company can make users, models, and token counts visible, but a way to score collaboration quality remains under development. The gateway does not automatically detect when work deviates; people currently check progress and make adjustments as needed.

ARI also uses AI to modernize older systems. It gathers existing system behavior through browser interactions, creates reproduction tests, then has AI build a system with upgraded operating system, middleware, and language versions while preserving the specifications, making corrections until the tests pass. People handle production infrastructure and reviews. Yamaoka said this can greatly reduce development hours and allow larger projects to be delivered in relatively shorter periods. But as efficiency reduces work time, project scope also shrinks, creating a problem for labor-hour pricing: efficiency gains are absorbed into lower fees.

ARI is considering payment tied to results and charges based on usage of its own AI platform and applications, but no outcome-based projects had begun at the time of the interview. In addition to system development, it has started offering an end-to-end service covering workflow assessment and redesign, safe connections to existing systems through a gateway, and agent development and evaluation. Because companywide transformation at once is difficult, ARI starts with test environments or a single department or workflow; it is supporting invoice-processing automation and expense claims through AI chat. Yamaoka cited overestimated security risks, inconsistent workflows, unclear responsibility between departments, and insufficient implementation capacity as reasons AI adoption stalls. He forecast that SaaS companies would move toward AI-native operations within six months, followed by large companies, with the shift spreading to other businesses over 2–3 years. He said ARI aims to become a recognized player “in these 2–3 years.”