AIエージェント時代の事業価値を4ドメイン×3区分で分解する枠組み This article proposes a framework for analyzing how business value will shift in the AI ag…

As AI agents move from demos into production deployments, companies need a structured way to assess how their business value will be reshaped. This article proposes a framework that decomposes the question along four domains and three categories, offering a lens for strategy and investment decisions in the agent era.

The author's underlying concern is that value distribution in an agent-driven world will likely diverge from the patterns established by SaaS and traditional web services. When agents autonomously execute parts or all of a workflow, several long-standing assumptions break down: UI-centric product value, defensibility through data accumulation, and cost structures built around human operations all need to be re-examined. Rather than collapsing this into a generic discussion of AI-driven efficiency, the framework insists on domain-level analysis.

The four domains appear to map onto the principal sites of value creation in a business — product, operations, data, and customer interface, or similar slices. Each is then evaluated against three categories, which can be read as enhancement, substitution, and dissolution (or value-up, neutral, value-down). The resulting matrix makes it easier to see where new opportunities emerge and where existing moats erode. A UI-heavy product, for example, is relatively easy to substitute once agents can call APIs directly, while domains anchored in proprietary data or deep operational knowledge tend to be reinforced.

This line of thinking connects to broader industry developments. Anthropic's Model Context Protocol (MCP), OpenAI's function calling and Responses API, and Google's agent frameworks are converging on standards that let agents reach external tools and data sources in uniform ways. As these protocols mature, the UI layer that classical SaaS monetized may be partially commoditized, with value migrating toward orchestration, trustworthy data, and domain expertise. Investors such as a16z and Sequoia have published similar theses, suggesting that value will move from the app layer to the agent layer, though the pace and extent remain contested.

It is worth tempering the narrative. The degree to which agents fully replace existing workflows will depend heavily on industry characteristics. Regulated sectors, safety-critical operations, and tasks that require human-in-the-loop judgment may preserve much of their current structure for years. Likewise, enterprises with strong process knowledge and integrated data assets may find that agents amplify rather than dismantle their advantage, at least in the near term.

Used appropriately, the four-by-three framework is less a forecast than a diagnostic instrument. It encourages teams to look at each part of their business individually, ask which forces apply, and avoid treating AI agents as a single monolithic disruption. For product leaders and strategists, that kind of granular mapping may prove more useful than headline-level predictions about which companies will win or lose in the agent economy.