NTTデータ先端技術, YCP and ABP recheck 7 SI hypotheses

The white paper is the first in a planned monthly series, and its starting point is the seven core hypotheses the three firms jointly presented in March 2026. Roughly four months later, they re-examined those hypotheses against subsequent market developments, rating each on a five-point scale and classifying the evidence into primary information, research, third-party surveys and reporting.

Among the key findings, generative AI has moved from single-answer generation toward agentic AI that autonomously carries out sequences of work across multiple applications and data. In June 2026, Anthropic disabled its top models for all customers under a US government export-control directive and resumed service in July, while OpenAI also began GPT-5.6 through a government-coordinated limited release. At the same time, mechanisms for safely operating agents, such as Microsoft's Agent 365, MCP and A2A, were productized in quick succession, shifting the competition from who has a high-performance model to who can operate agents safely and auditably.

The paper's argument is that generic layers will be handled by platformers while context-specific implementation remains with SI firms, because customer-specific authority design, approval paths, audit trails, legacy connections and accountability depend heavily on context. For Japanese SI firms this looks like a tailwind for now, but the paper says the shelf life of the person-month model is beginning to show, since under that model productivity gains feed directly into downward pricing pressure. The outcome likely hinges on how the four strategic showdowns—safety versus results, large enterprises versus startups, sovereign AI versus global, and training versus hiring freezes—play out, and on who is developed as the next five years' workforce.