Skip to content

hiten-skill-library.md

Latest commit

 

History

History
55 lines (40 loc) · 3.2 KB

hiten-skill-library.md

File metadata and controls

55 lines (40 loc) · 3.2 KB
Raw
title Hiten Shah — Every Company's First AI Strategy Should Be a Skill Library
source-id hiten-shah-skill-library-2026
wiki-page (01_Projects/20260616 GSB/References/Others/2026-06-07 Every Company's First AI Strategy Should Be a Skill Library.md)
last-synced 2026-06-07

Hiten Shah — Skill Library as First AI Strategy

What this reference contains

Hiten Shah's June 2026 thesis (X post): "Every company's first AI strategy should be a skill library." The cleanest single articulation of what makes AI strategy defensible and company-specific. A skill is a documented, versioned, forkable unit of how the work gets done — that an agent can load and execute.

Key claims

  • "Every company's first AI strategy should be a skill library." — Hiten Shah, June 2026.
  • A skill library captures what the company knows. It encodes the firm's processes (Process 20%), its people's expertise (People 70%), and is auditable for governance. It absorbs technology into something the company owns.
  • The model commoditizes. The skill library is yours. Foundation models are vendor-trained. The orchestration layer (MIT NANDA 2025) is where the moat lives — and a skill library IS the orchestration layer, expressed as institutionalized knowledge.
  • A skill library compounds as you operate. Each new skill added increases the system's capability. Each resolved edge case enriches the skill. The model doesn't improve with your company's operations — the skill library does.
  • Connects People (70%) and Process (20%) directly. A skill is: documented process (Process) + role-specific judgment encoded (People) + auditable trail (Governance) + executable by an agent (Technology). It is the 70/20/10 stack in a single artifact.

What a skill is

---
name: skill-name
description: Use when [trigger moment]. Runs [methodology] and outputs [deliverable shape].
---

[Role-based instructions that encode HOW the work gets done]

A skill is:

  • Documented — the knowledge is explicit, not locked in someone's head
  • Versioned — changes are tracked; the org can see how its knowledge evolves
  • Forkable — others can adapt it for their sector, role, or internal process
  • Agent-executable — Claude (or any capable LLM) can load and run it

Why this is the punchline

All five layers of the keynote point here:

  • Governance: skills are auditable, version-controlled, reviewable
  • Technology: skills absorb model capability without being dependent on any single model
  • Process: skills encode the redesigned workflows, not the legacy ones
  • People: skills capture the 70% — the expert judgment, the decision frameworks, the institutional knowledge
  • Strategy: a skill library is the moat — it grows with use, it's industry-specific, it's yours

When to consult

  • For skill general-maturity-assessment: a firm with a skill library is operating at Stage 3+ — it has architected for reuse, documented its processes, and is treating AI as infrastructure.
  • As the synthesis reference for all 8 skills: each skill in this collection is an instance of Hiten Shah's thesis.

Source

Hiten Shah, X post, June 2026. "Every company's first AI strategy should be a skill library."