Software is moving from adding AI as a feature to operating inside AI-driven, agentic environments, and the development lifecycle (DPDL) must evolve in lockstep. It maps twin progressions—products (no AI → AI features → single agents → agentic systems) and processes (manual → GenAI-assisted → agentic IT systems)—into a 3 × 4 matrix that organizes practical guidance and guardrails. Across the SDLC, it catalogs quick-win GenAI use cases in requirements, UX, architecture, coding, QA, deployment, operations, and project management, stressing human governance, traceability, and the rising importance of data and security engineering. For autonomy, it clarifies when to prefer a single agent versus a multi-agent system, introduces Agent Experience (AX) design, and standardizes artifacts like system cards, golden sets, runbooks, and lightweight gates with clear RACI ownership. Looking ahead, it forecasts smaller, agent-orchestrated teams led by a Digital Product Creator and urges adopting standard agentic development systems, evaluating end-to-end outcomes, and scaling autonomy only with evidence.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Connecting Requirements Engineering to Agentic Development

  • Jens Kawelke,
  • Thomas Niebisch

摘要

Software is moving from adding AI as a feature to operating inside AI-driven, agentic environments, and the development lifecycle (DPDL) must evolve in lockstep. It maps twin progressions—products (no AI → AI features → single agents → agentic systems) and processes (manual → GenAI-assisted → agentic IT systems)—into a 3 × 4 matrix that organizes practical guidance and guardrails. Across the SDLC, it catalogs quick-win GenAI use cases in requirements, UX, architecture, coding, QA, deployment, operations, and project management, stressing human governance, traceability, and the rising importance of data and security engineering. For autonomy, it clarifies when to prefer a single agent versus a multi-agent system, introduces Agent Experience (AX) design, and standardizes artifacts like system cards, golden sets, runbooks, and lightweight gates with clear RACI ownership. Looking ahead, it forecasts smaller, agent-orchestrated teams led by a Digital Product Creator and urges adopting standard agentic development systems, evaluating end-to-end outcomes, and scaling autonomy only with evidence.