Skip to content

R0021/2026-03-25/Q006 — Assessment

BLUF

RFC 2119 requirement language has been applied to AI behavioral specifications in at least one published work — a practitioner blog post (deliberate.codes, Feb 2026) that uses SHALL/SHOULD/MAY for AI coding agent specifications. However, no formal standard, academic paper, or major vendor documentation applies RFC 2119 to prompt engineering specifically. The scarcity of examples is itself a significant finding: the engineering community has not adopted formal requirement language for prompt design.

Probability

Rating: Likely (55-80%) that RFC 2119 has been applied to AI-adjacent specifications but unlikely (20-45%) that it has been applied to prompt engineering specifically.

Confidence in assessment: Medium

Confidence rationale: Only one published example found. The absence of evidence is meaningful but not conclusive — more examples may exist in unpublished or non-English work.

Reasoning Chain

  1. One published blog post (Feb 2026) explicitly applies RFC 2119 to AI coding agent specs [SRC01-E01, Medium reliability, High relevance]
  2. RFC 2119 was designed for interoperability specifications, not behavioral constraints, though its scope includes "limiting behavior which has potential for causing harm" [SRC02-E01, High reliability, Medium relevance]
  3. No major AI vendor documentation uses RFC 2119 language [cross-reference Q003 findings]
  4. JUDGMENT: The gap between RFC 2119's formal requirement language and prompt engineering's informal guidance is evidence that prompt engineering has not adopted engineering specification practices.

Evidence Base Summary

Source Description Reliability Relevance Key Finding
SRC01 deliberate.codes blog Medium High Only published RFC 2119 + AI application
SRC02 RFC 2119 Original High Medium Original scope and usage constraints

Collection Synthesis

Dimension Assessment
Evidence quality Limited — one practitioner blog post is thin evidence
Source agreement N/A — only one relevant source found
Source independence N/A
Outliers None

Gaps

Missing Evidence Impact on Assessment
Academic papers on formal specification of AI behavior Moderate — may exist in AI safety literature
Non-English publications Minor
Internal corporate specification practices Moderate — companies may use RFC 2119 internally

Researcher Bias Check

Declared biases: Researcher advocates for applying engineering rigor to AI systems. Finding a gap validates this position.

Influence assessment: The absence of evidence is genuinely significant, but confirmation bias could lead to under-searching for examples.

Cross-References

Entity ID File
Hypotheses H1, H2, H3 hypotheses/
Sources SRC01, SRC02 sources/
ACH Matrix ach-matrix.md
Self-Audit self-audit.md