Agentic AI in the Software Development Lifecycle: Architecture, Empirical Evidence, and the Reshaping of Software Engineering

A comprehensive survey paper on arXiv, authored by Happy Bhati, synthesizes findings from Anthropic, OpenAI, Google DeepMind, Microsoft Research, Princeton, and Stanford to characterize the transition from code-completion tools to agentic AI systems in software engineering. The paper proposes a six-layer reference architecture for agentic software engineering systems and contrasts the traditional Software Development Lifecycle (SDLC) with an emerging Agentic SDLC (A-SDLC). Key agents analyzed include Claude Code, OpenAI Codex CLI, Google Jules, Devin, OpenHands, SWE-agent, MetaGPT, ChatDev, and DeepMind's AlphaEvolve, which now operate at the granularity of a repository, feature, or algorithm rather than a single line or function.

Empirical evidence shows a dramatic performance leap on SWE-bench Verified from 1.96% in October 2023 to 78.4% by April 2026. Controlled studies across multiple organizations report productivity gains of 13.6% to 55.8% in time savings. Labor-market impact is significant: Anthropic's 2026 survey found 49% of sampled jobs used AI for at least a quarter of their tasks. The paper argues the central focus has shifted from code generation to delegated execution under human supervision. It identifies five open problems—evaluation, governance, technical debt, skill redistribution, and the economics of attention—that will determine whether this agentic transition is net-positive for the discipline.