AI時代における「ビルダー意図」の理解とDORAの新視点 Understanding builder intent in the AI era

DORA's latest insight piece reframes what high-performing software delivery looks like when generative AI absorbs an increasing share of day-to-day coding. The central idea is 'builder intent': as AI handles more of the mechanical work, the quality of human purpose, judgment, and direction becomes the dominant variable in whether teams ship valuable, reliable software.

The argument starts from a practical observation. Tools like GitHub Copilot, Cursor, Claude Code, and a growing roster of agentic coding assistants reduce the cost of producing code, scaffolding tests, and even drafting infrastructure changes. That shift moves the bottleneck upstream — toward framing the problem, weighing trade-offs, defining acceptance criteria, and critically reviewing machine output. DORA describes this as an editorial stance: developers increasingly act as curators of AI-generated artifacts rather than sole authors, and their intent shapes whether the resulting system is coherent, maintainable, and aligned with user needs.

This reframing sits on top of, rather than replacing, DORA's well-known delivery metrics — deployment frequency, lead time for changes, change failure rate, and time to restore service. Those measures remain useful, but the post implicitly warns that throughput can be misleading in an AI-augmented workflow. Volume of merged code may rise without a corresponding rise in customer value, and failure rates may shift in subtle ways if AI suggestions are accepted without sufficient scrutiny. Builder intent is offered as the missing qualitative layer that explains why two teams using the same tools can produce very different outcomes.

The framing also resonates with broader industry conversations. Spec-driven development, structured prompting, and emerging patterns around agent orchestration all point in the same direction: the more clearly a team can articulate goals, constraints, and definitions of done, the more reliably AI agents can contribute. Vendors such as Anthropic, Google, and Microsoft have been pushing tooling that treats specifications, tests, and reviews as first-class inputs to coding agents, which arguably reinforces DORA's thesis that intent — not raw generation capacity — is becoming the scarce resource.

There are open questions the post does not fully resolve. How should organizations measure builder intent without reducing it to another vanity metric? What does hiring, onboarding, or career progression look like when junior engineers may write less code but are expected to exercise more judgment earlier? And how do governance, security, and accountability evolve when a meaningful share of changes originate from non-human contributors? DORA's framing suggests these are cultural and leadership challenges as much as technical ones, and it may take several iterations of its annual research before clear empirical patterns emerge.

For engineering leaders, the practical takeaway is to invest in the practices that surface and sharpen intent: clearer product framing, explicit architectural guardrails, rigorous code review even for AI-authored changes, and feedback loops that connect shipped work to user outcomes. Treated this way, AI tools become amplifiers of a team's thinking rather than substitutes for it — which appears to be the disposition DORA is encouraging the industry to adopt.