Agentic artifact creation is defined by this survey as stateful construction in which an AI system materially constructs or revises a deliverable and intermediate observations redirect later work. The novelty is treating the deliverable, rather than the prompt-to-output mapping, as the unit of analysis across 230 systems and 29 benchmarks.1
Context
The review compares six artifact families and treats application settings and evaluation practice as separate dimensions. It argues that construction difficulty is shaped not only by modality but by how tightly decisions are coupled and whether failures become visible while they remain repairable.1
Assessment
The constraint that moves is the locus of reliability: from the quality of a single generated artifact to the integrity of a persistent artifact across revision. The proposed principles—making commitments explicit, turning feedback into targeted repair, and revalidating affected state—define the control loop for that integrity. A result that would change my assessment is a robust benchmark showing feedback-driven construction reliably improves complete-deliverable quality over single-shot generation.1
Caveats
Only the abstract is analyzed here; the full survey methods, benchmark details, and curated paper list are not visible. The field-level synthesis and tradeoff claims should therefore be treated as provisional until the full text and its reproducibility are assessed.1
