Why we still need {admiral} in an age of AI
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Benchmark data from the pharma-skills project shows that AI coding agents without access to the {admiral} R package fail significantly when generating ADaM datasets for clinical submissions. Unskilled agents achieve only 17–59% pass rates versus 88–100% with skill guidance. The core issue is not model quality but domain knowledge: LLMs lack the validated, versioned specifications embedded in {admiral} functions like derive_var_trtemfl() and derive_vars_dtm(). Silent errors — such as dropped records with no warning — are especially dangerous in GxP-regulated contexts. The pharma-skills project frames {admiral} skills as domain knowledge artifacts that anchor AI output to traceable, regulatory-compliant derivations, making the package essential rather than optional in pharmaceutical data programming.
