This week named two obstacles worth diagnosing, data that cannot leave and decisions that cannot wait. Part III supplies the discipline, through Richard Rumelt’s Good Strategy / Bad Strategy (Crown Business, 2011): most AI strategies are shopping lists with a growth target on top.
Rumelt’s kernel has three parts in order: a diagnosis that names the critical challenge and simplifies the mess; a guiding policy that sets the approach and rules things out; and coherent actions that carry it, not a shopping list. Most “AI strategies” skip the first, opening with tools, platforms or a growth percentage, goals and fluff with no obstacle named, which is bad strategy in Rumelt’s sense. In a plant the obstacle is rarely “we lack an LLM”, it’s latency, data that can’t leave, integration debt, or an unseen constraint. The quality-suite example shows it: write the diagnosis first and the tools become consequences; skip it and the same budget buys a chatbot and a dashboard. Four recurring bad-strategy patterns, goals mistaken for strategy, fluff, obstacle not faced, incoherent actions, each avoid the same thing: stating the obstacle.
The test costs two sentences. Your action this week: take your AI roadmap and write the obstacle in two sentences, the policy that addresses it, and the actions that follow, if they contradict each other, or you can’t write the first sentence, that’s where to start. Full kernel at renegrywnow.com.
Reflection questions
Can you state your AI strategy’s obstacle in two sentences, before the next vendor meeting?
Does your current roadmap’s actions cohere, or does a cloud agent sit next to an air-gapped PLC with no write-back owner?
Is your strategy deck leading with a diagnosis, or with vendors and model names?
Series: Energy Dominance · Week 45 · Part III
Previous: Part I — Data Sovereignty · Part II — Edge AI: Latency, Power, On-Premise.
