The Strategy Kernel

The kernel — three parts, no substitutes

Every good strategy has the same logical skeleton (Rumelt, Good Strategy Bad Strategy). Missing any part means it isn't a strategy yet.

Part	What it is	One-line test
1 · Diagnosis	An honest, evidenced account of what is actually going on — the critical challenge, named. The hard part.	Would someone reading it wince a little because it's true?
2 · Guiding policy	The overall approach chosen to address the diagnosis — including what it rules out. A policy that forbids nothing is fluff.	Can you name a reasonable thing this policy forbids?
3 · Coherent action	Concrete, coordinated moves that carry out the policy and reinforce each other — not a grab-bag backlog.	Does each action trace back to the diagnosis?

The four tells of bad strategy (Rumelt)

Tell	What it looks like
Fluff	Buzzwords restating the obvious — "world-class", "best-in-class", "customer-centric excellence". Gibberish masquerading as insight.
Failure to face the challenge	No diagnosis. If the doc never names the real problem, there is no way to evaluate — or improve — the strategy.
Mistaking goals for strategy	Ambitions presented as approach: "reach 99.99% availability", "10x velocity". A goal is where you want to be; strategy is how you'll get there.
Bad objectives	Action lists that are impracticable, or a long grab-bag of unrelated wishes that fail to address the challenge — a to-do list, not coordinated action.

Larson's bottom-up recipe: write five, then synthesize

Write (or gather) five design docs for real problems in your area.
Synthesize the strategy: look across them for repeated dilemmas, controversial decisions, and contested trade-offs — write the doc that guides those trade-offs and explains the rationale.
Write five strategies (over time, across areas).
Synthesize the vision: extrapolate how their trade-offs interact over the next two to three years and reconcile them into one picture.

Good strategy is boring and specific — it often feels too obvious to bother writing down. That feeling is a sign you're doing it right, not wrong.

Strategy-doc review checklist — six questions

Is there a diagnosis at all — a named challenge, not a conclusion smuggled in ("we must modernize") or a list of ambitions?
Is the diagnosis evidenced and uncomfortable — at least one sentence backed by a number, incident, or trend that someone would rather not see written down?
Does the guiding policy exclude something — can you name a reasonable option it forbids? Would a skeptical peer dispute it? If nobody could disagree, it decided nothing.
Does every action trace to the diagnosis — no orphan initiatives riding along because they were already planned?
Do the actions reinforce each other — or do any two of them pull in opposite directions (funding the platform while blessing every bespoke pipeline)?
Is it one page — short enough that people will actually read it, remember it, and use it to make the next contested decision?

The whole page in one line Diagnosis names the real problem, guiding policy makes a contestable choice about it, coherent actions carry the choice out and reinforce each other — and a strategy nobody disagrees with is not a strategy.

Sources: Richard Rumelt, Good Strategy Bad Strategy (2011); Will Larson, "Writing an engineering strategy" (2023), "Write five, then synthesize" (2020), and Crafting Engineering Strategy (O'Reilly, 2025). Companion lesson: Lesson 10 — Writing Technical Strategy.