How to Chain Prompts to Tackle Complex Tasks

PROMPT:
"Research the competitive landscape for project management software,
identify our three biggest differentiators, write a 500-word executive summary
for investors, and then create a 10-slide presentation outline."

RESULT:
The AI does all four things — but none well. The research is surface-level,
the differentiators are generic, the executive summary is padded, and the
presentation outline doesn't connect to the earlier analysis.

The model is trying to do four cognitively distinct jobs in sequence,
with no time to think carefully about any one of them.

[Prompt 1]  →  [Output 1]
                        ↓
               [Prompt 2 + Output 1]  →  [Output 2]
                                               ↓
                              [Prompt 3 + Output 2]  →  [Final Output]

                    [Prompt 1: Generate options]
                               ↓
              ┌────────────────┼────────────────┐
              ↓                ↓                ↓
         [Option A]       [Option B]       [Option C]
              └────────────────┼────────────────┘
                               ↓
                    [Prompt 2: Evaluate & select best]
                               ↓
                    [Prompt 3: Develop selected option]

[Prompt 1: Generate draft]
        ↓
   [Draft output]
        ↓
[Prompt 2: Critique the draft]  ←──────────────────┐
        ↓                                           │
   [Critique]                                       │
        ↓                                           │
[Prompt 3: Revise based on critique]               │
        ↓                                           │
   [Revised draft]                                  │
        ↓                                           │
[Check: Does it meet criteria?] ──── NO ───────────┘
        ↓ YES
   [Final output]

PROMPT:
"You are going to help me write a business proposal. Before writing anything,
interview me to gather the key information. Ask me these questions one at a time:

1. What problem does your product/service solve?
2. Who is the target customer?
3. What makes you different from competitors?
4. What is the ask (budget, timeline, desired outcome)?
5. Any context about the reader (who will receive this proposal)?

Ask question 1 first and wait for my answer before continuing."

---
[User answers all 5 questions in the conversation]
---
OUTPUT (saved as context):
- Problem: construction firms lose 15% revenue to invoice disputes
- Customer: GCs with 20–200 employees
- Differentiators: automated dispute detection, integrates with existing accounting tools
- Ask: $50k pilot contract, 6-month engagement, outcome = 50% reduction in disputes
- Reader: CFO at a mid-size construction firm, sceptical of new software

PROMPT:
"Using the context below, create a logical argument structure for a business proposal.
Do not write the proposal yet — just the skeleton.

Output format:
- Executive summary (key message in one sentence)
- Problem statement (3 bullet points)
- Solution overview (3 bullet points)
- Differentiation (3 bullet points vs status quo)
- Proposed engagement (scope, timeline, investment)
- Risk mitigation (top 2 objections + responses)

Context: [paste Step 1 output]"

---
OUTPUT: A bullet-point structure ready for Step 3

PROMPT:
"Write a 600-word business proposal based on this structure.
The reader is a CFO at a mid-size construction firm who is sceptical of new software —
lead with ROI and proof, not features. Use professional but direct language.
No jargon. No passive voice.

Structure to follow: [paste Step 2 output]"

---
OUTPUT: The full proposal draft

PROMPT:
"Review this proposal as the CFO who will receive it. You are sceptical — you have seen
many software sales pitches that overpromised.

Identify:
1. Any claim that feels unsubstantiated — flag it and suggest how to back it up
2. Any section that is too long or padded — cut it
3. The single weakest part of the argument — how would you strengthen it?

Then give me the revised version with all issues addressed.

Proposal to review: [paste Step 3 output]"

---
OUTPUT: Final polished proposal

PROMPT:
"Read this code carefully. Do not review or critique it yet.

1. In 3 sentences, describe what this code does
2. List every dependency or external call it makes
3. List every assumption it makes about its inputs

Code: [paste code]"

---
OUTPUT (feeds into Step 2):
- Summary: "This is an authentication middleware that validates JWT tokens..."
- Dependencies: jsonwebtoken, express, User model from database
- Assumptions: req.headers.authorization is always a string, tokens never expire in testing

PROMPT:
"Now review the code below for issues in each of these categories. Be specific — give line
numbers and exact problems, not vague observations.

Categories:
1. Security vulnerabilities (especially around auth since this is auth middleware)
2. Edge cases not handled
3. Error handling — are errors specific enough? What would a developer see in production?
4. Performance — anything that could be slow at scale?
5. Maintainability — anything that will confuse the next developer?

Code: [paste original code]
Context from Step 1: [paste Step 1 output]"

---
OUTPUT: Categorised issue list

PROMPT:
"For each Critical and Major issue below, give the exact fix as a code snippet.
Format each fix as:
- Issue: [restate the problem]
- Why it matters: [one sentence]
- Fix: [code block]

Do not include Minor issues in this step — I will handle those separately.

Issues from Step 2: [paste Step 2 output — Critical and Major only]"

---
OUTPUT: Actionable fixes ready to apply