From Code Completion to Agentic Development: A Survey of AI Coding Agents, Ecosystems, and Evolving Software Engineering Practice

Large language models have extended software-development assistance from token-level completion to systems that can navigate repositories, edit files, execute tools, and iterate on test feedback. This survey studies the design, use, and evaluation of these AI coding agents, together with their commercial and societal context. Reproducible database queries support literature discovery, but the cited corpus is selected and coded for relevance; we therefore describe the work as a structured scoping survey rather than an exhaustive systematic review. We organize the field using five dimensions: interaction paradigm, computational architecture, deployment form, orchestration mode, and human-oversight mode. Treating oversight separately from architecture resolves a conceptual overlap in earlier classifications and gives each representative system one value per dimension. The technical synthesis covers repository context, planning, tool use, multi-agent coordination, SDLC applications, and benchmark limitations. We keep technical capability evidence separate from business and adoption reports. Vendor and news sources are used as dated context, not as proxies for maturity or user preference. The survey covers material available through 28 February 2026.