Ask AI To Show Its Reasoning

Ask AI to Show Its Reasoning (AI Skill)

Overview

When an AI simply proclaims: "I recommend Option B," you have no way to evaluate whether the conclusion is based on deep technical trade-offs or an arbitrary bias in training data.

Forcing the AI to expose its reasoning path before stating its final recommendation achieves two critical outcomes:

1. Dramatically higher accuracy: Generating intermediate reasoning tokens allows the model to compute complex dependencies before finalizing the output.
2. Auditability: It allows humans to inspect the logical premises, identify flawed assumptions, and make fully informed decisions.

The 4-Step Transparent Reasoning Framework

┌─────────────────────────────────────────────────────────────┐
│                 Transparent Decision Flow                   │
│                                                             │
│  1. Evaluation Criteria  ──► Weight key factors (Cost, Ops) │
│            │                                                │
│  2. Analysis of Options  ──► Strengths & fatal flaws        │
│            │                                                │
│  3. Elimination Logic    ──► Why alternatives were rejected │
│            │                                                │
│  4. Final Recommendation ──► Clear, defended conclusion    │
└─────────────────────────────────────────────────────────────┘

Master Reasoning Prompts

Pattern 1: The First-Principles Decision Engine

Use this when choosing between technical architectures, business strategies, or tools:

I need to decide between [OPTION A], [OPTION B], and [OPTION C] for [SPECIFIC USE CASE & CONSTRAINTS].

Structure your analysis as follows:
1. **Decision Criteria**: Define the 3-4 most critical factors that determine success for this scenario.
2. **Side-by-Side Trade-off Table**: Score each option against those criteria (High / Medium / Low or 1-5).
3. **Elimination Analysis**: Explain why each rejected option fails to meet the threshold.
4. **Final Recommendation**: State your chosen option with a 2-sentence executive defense.

Pattern 2: The "Show Your Work" Logic Constraint

Use this for analytical, policy, or legal queries:

Evaluate [SCENARIO / QUESTION].

Do not jump straight to the verdict.
First, explain:
- What premises or rules apply here?
- What evidence supports the claim?
- What counter-arguments or edge cases exist?

Then, conclude with your verdict based strictly on the evidence above.

Real-World Examples

Example 1: Database Architecture Decision

Opaque Recommendation (Low Value)

User: "Should I use DynamoDB or PostgreSQL for an IoT sensor platform collecting 50,000 writes/sec?"

AI (Opaque): "You should use DynamoDB because it is highly scalable and handles NoSQL workloads well."

Transparent Reasoning Prompt & Output

User: "Evaluate DynamoDB vs. PostgreSQL for 50k sensor writes/sec. Show your reasoning, trade-offs, and elimination logic."

AI Response:

1. Evaluation Criteria:

• Write Throughput: Sustained 50,000 writes/sec with predictable sub-10ms latency.
• Query Patterns: Time-series rollups, range queries, and device alerts.
• Cost at Scale: Storage and I/O pricing under heavy ingestion.

2. Analysis & Trade-Offs:

• DynamoDB:
  • Pros: Effortlessly scales to 50k writes/sec with on-demand or provisioned WCU. Zero server maintenance.
  • Cons: Cost scales linearly with write volume ($$$); analytical range queries and time-series aggregation require exporting to S3/Athena.
• PostgreSQL (with TimescaleDB extension):
  • Pros: Native time-bucket aggregation, compression ratios up to 90%, cost-effective on large EBS volumes.
  • Cons: Requires cluster tuning (connection pooling with PgBouncer, auto-partitioning/hypertables, replica management).

3. Elimination Logic:

• Standard vanilla PostgreSQL is eliminated due to table bloat and write lock contention at 50k writes/sec.
• Pure DynamoDB is eliminated if complex time-series queries (e.g., 7-day moving averages across sensor fleets) are required in-database without secondary tooling.

4. Recommendation:

• Choose TimescaleDB (PostgreSQL) if you need rich SQL analytics and fixed infrastructure costs.
• Choose DynamoDB if zero-operational overhead and pure key-value ingestion speed outweigh cost and analytical flexibility.

Summary Matrix