Tree-of-Thoughts for Planning
When linear reasoning fails — enumerate branches, score independently, surface trade-offs, recommend a hybrid path.
“CoT picks one path and rides it — even when 3 better paths existed. ToT enumerates the paths, scores them on independent dimensions, and recommends the hybrid that no single path could produce. This is how you avoid the 'we should have considered X' regret 2 quarters later.”— SHE · YOUR AI GUIDE
Yao et al.'s 2023 Tree-of-Thoughts paper exposed a fundamental limitation of Chain-of-Thought: linear reasoning is fast and elegant, but it commits to one branch early and rationalizes from there. On tasks where multiple strategies are viable (market entry, build-vs-buy, pricing structure, hiring path), CoT picks the most familiar option from the model's training distribution and produces a confident-sounding analysis of that single option — while better options remain unexplored.
The production fix is structural branch enumeration. Force the model to name 4 distinct strategies BEFORE evaluating any. Score each on 4 independent dimensions — feasibility, speed, risk, long-term-value — so a strategy that's high-feasibility-low-value doesn't get falsely promoted by an averaged single-dimension score. Identify the top 2 and surface their COMBINATION potential: real-world strategy is rarely 'pick one,' it's 'do A first, pivot to B if signal X appears.'
The deeper insight: ToT is overkill on most tasks. Use ToT when (a) the decision is irreversible or very expensive to reverse, (b) multiple strategies look viable, (c) you'll be asked 'why not the alternative?' by a stakeholder. Skip ToT when there's one right answer (extraction tasks, well-specified outputs), when the decision is cheap to reverse, or when the model has clear training-data superiority on one approach. The meta-skill is knowing which problem class you're in.