From issue trees to subagents — porting strategy craft into Claude Code.

Draft in progress

What this essay will cover.

• The classical issue tree — root question, MECE branches, hypothesis at each leaf, evidence to test it. Why every McKinsey project really did start this way and why it works.
• The structural isomorphism — a Claude Code sub‑agent graph is the same artefact: root prompt, decomposed sub‑prompts, evidence (tool calls and outputs), recombination at the root. Same shape, different substrate.
• What transfers — MECE discipline, hypothesis‑first thinking, the rigour of the "so what?" test at each level, the partner habit of pruning a branch you've fallen in love with.
• What doesn't — the human apprenticeship around the tree (debating the branches with an EM, watching a partner reshape it in front of you) doesn't transfer; the model is not the EM, and pretending it is leads to lazy decomposition.
• The practical pattern I now use — a written root prompt, three to five sub‑agents, each with a specific evidence target, a synthesiser at the end. Looks remarkably like a McKinsey workplan.
• Why this is the most underrated career capital from a consulting background — the shape of how you decompose problems is the moat. AI fluency without decomposition discipline is just polished noise.

— Full essay coming. The thesis stands; the longer version walks through a concrete Frollie example.