---
type: "synthesis"
sources: ["agentic"]
tags: ["machine-readable", "knowledge-management", "codification"]
id: "cross-codification-imperative"
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-seg-agentic"
sourceUrl: "(unified vault: 9 sources)"
sourceTitle: "HBR — Firm Ⅱ-A · Agentic operating model / brand code / workstreams"
---
Across the corpus, the most concrete engineering prerequisite for agentic work is the same: **translate tacit, human-formatted knowledge into structured, machine-readable form.** Four articles independently reinvent this idea under four names.

- A002 calls it the **[[concept-brand-code]]** — a machine-readable base of taxonomies, prompt templates, decision trees, and tagged datasets forming the [[concept-foundation-layer]]; build it via [[action-codify-brand-code]] (see [[quote-brand-code-onboarding]]).
- A017 calls it **plain-text conversion**: [[action-convert-to-markdown]] is the *highest-leverage immediate change* ([[claim-markdown-highest-leverage]]) because [[concept-human-formatted-data]] like PDFs are [[quote-pdfs-are-outputs|"outputs for humans, not sources of truth for agents"]].
- A027 calls it **[[concept-judgment-infrastructure]]** built by [[concept-codifying-judgment]]; Nathan Mapp's canonical move is [[action-codify-into-markdown]], and [[action-use-transcripts-as-context]] mirrors the brand code exactly.
- A018 extends codification *outward* via [[concept-llms-txt]] and [[action-adopt-llms-txt]] — machine-readable data for third-party LLMs (see [[cross-machine-legibility]]).

The deep convergence: all four describe the same artifact — a dynamic, versioned, machine-consumable representation of "how we do things." A026 supplies the crucial caveat: codification captures only the [[concept-retrievable-layer]], never the [[concept-professional-discretion|discretion layer]]. And A027's [[contrarian-experts-cannot-document]] warns you *cannot* extract it by asking experts to write things down — use debate ([[framework-scenario-based-extraction]], [[quote-debate-externalizes-reasoning]]).

Every article flags the same unresolved problem — maintaining the artifact: compare [[question-brand-code-maintenance]] (A002) with [[question-maintaining-codified-judgment]] (A027). They are nearly identical open questions, which is itself evidence that codification maintenance is the corpus's shared blind spot.