Multi-Step Reasoning

The Problem

Complex problems cannot be solved in a single leap. When you ask AI to evaluate a system architecture, debug a race condition, or plan a migration strategy, a one-shot answer skips critical intermediate reasoning. The model needs explicit scaffolding to break the problem into sub-problems, solve each one, and synthesize the results.

The Prompt

Solve this problem using explicit multi-step reasoning. Do NOT jump to a conclusion.

PROBLEM:
[describe the complex problem]

Follow this reasoning framework:

STEP 1 — DECOMPOSE: Break the problem into 3-5 independent sub-problems. List them.
STEP 2 — ANALYZE: For each sub-problem, reason through it independently. Show your work.
STEP 3 — SYNTHESIZE: Identify how the sub-problem solutions interact. Look for conflicts or dependencies.
STEP 4 — VALIDATE: Check your synthesis against the original problem. Does it address all aspects?
STEP 5 — CONCLUDE: State your final answer with confidence level and remaining uncertainties.

If at any step you realize a previous step was wrong, explicitly correct it before continuing.

Example Output

STEP 1 — DECOMPOSE:
Sub-problems identified:
1. Can the database handle 10x read traffic with current indexes?
2. Does the caching layer invalidate correctly on write?
3. Will the API rate limiter throttle legitimate bulk operations?

STEP 2 — ANALYZE:
1. Current indexes cover 80% of query patterns. The remaining 20% (full-text search) will degrade.
2. Cache invalidation uses event-based purging — correct for single writes, but bulk imports bypass the event system.
3. Rate limiter uses a fixed window — bulk operations from a single client will be throttled after 100 requests.

STEP 3 — SYNTHESIZE:
The bulk import scenario is the weak link: it bypasses cache invalidation AND triggers rate limiting simultaneously.

STEP 4 — VALIDATE: All three original concerns are addressed. The bulk import gap was not obvious without decomposition.

STEP 5 — CONCLUDE: Scale reads via read replicas, add a bulk import queue that handles cache purging, and switch to sliding window rate limiting. Confidence: High.

When to Use

Use multi-step reasoning for problems with multiple interacting variables — capacity planning, security threat modeling, migration strategies, or debugging issues that span multiple system layers. It is particularly effective when the obvious answer feels too simple and you suspect hidden interactions.

Pro Tips

• Number your sub-problems — this forces the model to track each one explicitly and prevents it from quietly dropping difficult sub-problems.
• Require explicit corrections — the instruction to “correct previous steps” gives the model permission to backtrack instead of defending a wrong path.
• Limit decomposition to 3-5 items — more than 5 sub-problems causes the model to lose track of interactions between them.
• Combine with role-based prompting — assign a domain expert role before the reasoning framework to improve the quality of each step.