Paper proposes Spec-Driven Agentic Development as a new paradigm for AI-native software delivery

A new arXiv preprint proposes Spec-Driven Agentic Development (SDAD), a process model that integrates machine-readable specifications with autonomous coding agents to restructure the Software Development Life Cycle (SDLC). The authors argue that large language model–backed coding agents with context windows ranging from hundreds of thousands to millions of tokens can now ingest substantial functional requirement documents and repository context in a single workflow, making specification quality the primary driver of autonomous delivery. The paper formalizes SDAD as a synthesis of disciplined up-front formalization and high-velocity implementation, emphasizing intent capture, machine-readable specification, agentic synthesis, and independent multi-agent verification under human sign-off.

The authors revisit the historical tension between Waterfall and Agile methodologies and introduce AI-code as a fourth production paradigm. They compare Human-Agile (circa 2020) with Agentic-SDAD (circa 2026) across artefacts, cadence, accountability, and security posture, arguing that agentic speed does not eliminate engineering discipline but relocates it upstream into specification precision, explicit gates, and auditable provenance.

Beyond process description, the paper extends SDAD to team role transformations across engineering, QA, platform, and product functions, and introduces quantitative governance metrics such as Ambiguity Tax, Spec Fidelity, SER, and TCI_agentic with a repair multiplier phi. It also proposes a pragmatic adoption blueprint via hybrid estimation and staged migration, integrating industrial and research evidence on AI-augmented testing and verification to motivate a separation between synthesis and release authority.