Part V — The Frontier

The volatile layer. LLM agents now draft derivations, generate modules, and write QC comparisons faster than any human — and the failure mode is uniform: confident, clean code over unsourced decisions. These chapters treat every capability claim as dated evidence, not direction, and each carries a last-verified date for exactly that reason.

Chapter 12 reads the 2024–2026 production cases as ledger entries — task, model role, human gate, failure mode — and names the exact wall each one hit. Chapter 13 builds an agent workflow over clinical data with audit-friendly guardrails. Chapter 14 judges how close natural-language-to-submission really is, and which bottleneck is regulatory rather than technical. Chapter 15 compresses the whole series into one argument about the next five years.

The rule every frontier chapter ends with: if an agent drafted it, a human cites it — SAP reference, standard, or spec — before it ships. The bottleneck that remains is not generation; it is verification.