Skip to content

pilot-discipline-ng.md

Latest commit

 

History

History
42 lines (31 loc) · 3.02 KB

pilot-discipline-ng.md

File metadata and controls

42 lines (31 loc) · 3.02 KB
Raw
Error in user YAML: (<unknown>): mapping values are not allowed in this context at line 2 column 25
---
title: Pilot Discipline — Ng + Stanford + PwC
source-id: (multi-source: landing-ai-playbook, stanford-playbook-2026, pwc-agentic-playbook-2026)
wiki-page: (02_References/enterprise-ai/wiki/skills/process/cap-pilots-90-days-co-evolve.md)
last-synced: 2026-06-07
---

Pilot Discipline

What this reference contains

The cross-source consensus on structuring a value-capturing AI pilot. Three sources converge: Andrew Ng's pilots-first methodology (Landing AI), Stanford's 51-deployment analysis, and PwC's ROI checklist. Key discipline: 90-day cap, pre-deployment metrics, narrow scope, stop conditions defined at kickoff.

Key principles

  • Pilots before strategy. "Most companies will not be able to develop a thoughtful AI strategy until it has had some basic experience with AI." The pilot produces the signal; the strategy names the moat. Source: Landing AI Playbook, p.3.
  • 90-day cap. Pilots that don't show measurable signal within 90 days rarely show it at 180. The 90-day cap is a forcing function for scope narrowness and metric discipline. Source: PwC + Stanford.
  • Pre-deployment metrics, not post. Name the three metrics that prove this is working BEFORE you deploy. If you can't name them at kickoff, the initiative isn't ready. Source: Stanford AI Index 2026 (measurement-gap failure mode).
  • Narrow scope + simple execution. NANDA Scope × Execution Grid: start upper-left (narrow + simple). Don't attempt broad scope and complex execution simultaneously. Source: MIT NANDA 2025.
  • Blast-radius assessment at design. Who sees it if this fails on day 1 — internal users, paying customers, regulators, public? The answer sets the reliability bar. Source: Stanford + IMDA.
  • Stop conditions defined at kickoff. If metric X isn't reached at day 30/60/90, the pilot pauses for diagnosis. Without stop conditions, pilots drift to perpetual "almost ready."
  • Co-evolve human workflow with AI deployment. Don't bolt AI onto the existing workflow; redesign the workflow as you pilot. Source: Stanford AI Index 2026, McKinsey 2025.

The pilot design checklist

At kickoff, answer:

  1. What is the scope? (Narrow; one function, one workflow)
  2. What problem is being solved? (Friction-first, not tech-first)
  3. What are the three metrics that prove this is working?
  4. Who is the blast radius if this fails?
  5. What is the stop condition at day 30, 60, and 90?
  6. How will the human workflow be redesigned alongside deployment?
  7. Who owns this pilot (named individual, not a team)?

When to consult

  • For skill process-pilot-design: this is the full reference for pilot structure — all seven checklist items are the output format.
  • For skill general-idea-diagnostic: Q4 (capability check) asks whether the org can execute a disciplined 90-day pilot on this concept.

Source

Multi-source synthesis: Landing AI Playbook (pilots-first principle), Stanford AI Index 2026 (pre-deployment metrics, workflow redesign), MIT NANDA 2025 (Scope × Execution Grid, 90-day timeline), PwC Agentic AI Playbook 2026 (ROI checklist items 3 and 4).