AI Agent Dev

🔍 Assignment: Build a Simple AI Agent

🧠 Objective:
To assess your curiosity, learning agility, and problem-solving approach, not just your technical
skills.

Even if you’re new to coding or AI, this is your chance to show how you think, explore, and work
with AI tools like ChatGPT to build something meaningful.

📦 What You’ll Build:

Design a simple AI Agent (like a chatbot or a task assistant) using the 4-step architecture model. You
don’t need to code everything perfectly. Instead, show how far you can go by:

●​ Asking the right questions

●​ Iterating on prompts
●​ Learning through ChatGPT​

🛠️ Your Task:
1.​ Choose a Use Case:​
Pick a simple problem that an AI Agent can help with.​
Example:​

○​ “AI Agent to suggest daily writing prompts for bloggers.”

○​ “AI Agent to help students revise for an exam.”​

2.​ Design the 4 Layers of Prompts:​

For your use case, create prompts for:​

○​ Input Understanding (What is the user asking?)

○​ State Tracker (What context/state is remembered?)
○​ Task Planner (What steps does the agent take?)
○​ Output Generator (How does it respond clearly?)​
3.​ 💡 Use the reference documents to guide you. You can modify the prompts or create your
own.​

4.​ Iterate & Build with ChatGPT:​

Use your prompts inside ChatGPT.
○​ Ask follow-up questions if the output is not as expected.
○​ Keep improving your prompts and logic step-by-step.​

5.​ Submit an Approach Document:​

This is the most important part.​
​
Your document should include:
○​ Your chosen use case.
○​ Prompts for each of the 4 steps.
○​ Snapshots or text logs of your ChatGPT interactions.
○​ What you learned, where you got stuck, and how you overcame it.
○​ Any code samples or system messages if generated.​

6.​ 📌 Think of this as a journal of your journey with ChatGPT.​

📤 AI Agent Assignment – Submission Format

​

Note: This document is both your journal and artifact of thought. We’re evaluating
how you think, build, and debug, not just what you build.
🧾 SECTION 1: BASIC DETAILS​
Name:​
AI Agent Title / Use Case: (e.g., AI Agent to help college students generate daily journaling prompts)

🧠 SECTION 2: PROBLEM FRAMING

1.1. What problem does your AI Agent solve?​
(Write 2-3 lines to describe the user problem and context)​
1.2. Why is this agent useful?​
(Describe the core value delivered to the user)​
1.3. Who is the target user?​
(Be specific — “college student revising for an exam”, “freelance writer”, etc.)​
1.4. What not to include?​
(Optional, but useful. Mention features you consciously avoided for scope discipline)

🧱 SECTION 3: 4-LAYER PROMPT DESIGN​

Create a subsection for each of the 4 components of the agent architecture:

🔹 3.1 INPUT UNDERSTANDING​

Prompt: (What did you write in ChatGPT?)​
What is this prompt responsible for? (1-2 lines)​
Example Input + Output: (Optional if tested)

🔹 3.2 STATE TRACKER​

Prompt:​
How does this help the agent “remember”?​
Did you simulate memory with variables / system messages? If yes, how?

🔹 3.3 TASK PLANNER​

Prompt:​
What steps does your agent take internally to solve the problem?​
Did you use chaining? Branching? How did you manage complexity?​
​
3.4 OUTPUT GENERATOR​
Prompt:​
What kind of output formatting or phrasing did you aim for?​
Any special behavior? (e.g., examples, markdown formatting, tone control, etc.)​

🔍 SECTION 4: CHATGPT EXPLORATION LOG

​

You may structure this as a table or bullet list.

 Attempt # Prompt What What You Why You
Variant Happened Changed Changed It
1 “Generate Output too Added user To make
journaling generic context prompts more
prompts” personal
2 … … … …

Minimum 3 meaningful iterations recommended.

🧪 SECTION 5: OUTPUT TESTS (Optional but Recommended)

Include 2–3 sample outputs from your agent. Try edge cases.

●​ Test 1: Normal input​

Input: "I want a journaling prompt about career confusion"​
Output: (Paste output)
●​ Test 2: Vague input​
Input: "Give me something to think about"​
Output: (Paste output)
●​ Test 3: Invalid input or challenge​
Input: ""​
Output: (Paste output)

🔄 SECTION 6: REFLECTION​
Respond to each question briefly (3–5 sentences each):

●​ 6.1. What was the hardest part of this assignment?

●​ 6.2. What part did you enjoy the most?
●​ 6.3. If given more time, what would you improve or add?
●​ 6.4. What did you learn about ChatGPT or prompt design?
●​ 6.5. Did you ever feel stuck? How did you handle it?

🧠 SECTION 7: HACK VALUE (Optional)​

Did you go beyond the brief in any way?

●​ Simulated multiple users?

●​ Added logic for memory or role play?
●​ Explored chaining multiple agents?

If yes, explain in 2–3 lines.

​
Rubric for Evaluation (Work Simulation Style)
Dimension Weight What Good Looks Like Red Flags
Problem Framing 20% Clear, original use-case. Chose a generic or unclear
Well-scoped agent behavior. use-case (e.g. “make a
Avoided vagueness. Shows chatbot”). Vague problem
user-centric thinking. statement.

Prompt 25% Each of the 4 prompts reflects Prompts overlap in function,

Architecture different system are too verbose, or fail to
Quality responsibilities. Prompts are create modular behavior.
concise, modular, and
testable. Shows
understanding of functional
decomposition.

Exploration Quality 20% Good trail of follow-up questions One-shot prompts. No

(AI and prompt debugging. iteration. No recovery
Collaboration) Attempts multiple iterations. from errors. No
Shows ability to collaborate exploration of
with ChatGPT like a alternatives.
teammate.

Output Clarity & 15% The generated outputs (agent The output is trivial,
Functional responses or flow) repetitive, or structurally
Coverage demonstrate useful behavior. broken.
Shows command of output
formatting, response
variation, or state
transitions.

Documentation of 15% Thoughtfully documents the Only dumps the final result.
Process journey. Annotates prompts No iterative record. No
(Approach Doc) with rationale. Reflects on reflection or analysis.
what worked and what
didn’t. Data trail is clean.

Initiative / Hack 5% Goes beyond base instructions Minimum viable effort with
Value — e.g. added context no signs of initiative.
memory, simulated user
inputs, nested workflows, or
tried to test edge cases.

🕒 Time Estimate:
8-10 hours, depending on how deep you explore.

💡 Tips for Applicants

​

These should be shared as part of the brief to guide execution:

🧠 Think Like an Agent Architect
●​ Your job isn’t just to “make a chatbot” — you are defining how an agent breaks down a
user’s request into logical steps.
●​ Try to separate “understanding”, “thinking”, and “responding” like different departments in
a company.

💬 Ask Better Questions

●​ If ChatGPT doesn’t give what you want, don’t settle. Ask:
○​ “How can I make this output more modular?”
○​ “What if the user gives a bad input?”
○​ “How do I retain memory across turns?”​

●​ Your ability to ask follow-up questions shows whether you’re teachable and resilient.

📐 Design with Clarity

●​ Good prompts are like good APIs — each does one job and can be reused.
●​ Don’t hardcode outcomes. Let the prompt guide structure, not content.

🧪 Debug Like a Researcher

●​ When things go wrong, don’t delete and restart. Tweak. Test. Compare.
●​ Log what changed and why. We want to see your diagnostic thought process.

📓 Document Like an Engineer

●​ Think of your approach doc as a hand-off artifact — someone else should be able to reuse or
build on your logic.
●​ Annotate your prompts: “This prompt ensures the agent understands ambiguous user
questions by…”
●​ Don’t be afraid to admit where you got stuck. We value truth over polish.
Beginner-Friendly WBS: AI Agent Assignment
🎯 Total Estimated Time: 8 to 10 hours, including orientation to AI terms​
🎯 Success = Curiosity + Clarity of Thinking + Documenting the Journey
🧭 PHASE 0: Orientation – “What am I even doing here?” (60–75 min)
Task Description Time
0.1 Read this brief (assignment, structure, and expected output) 10 min
0.2 Watch/Read 1 basic explainer on “what is a prompt” and “how AI responds” 15–20
(shared by DT mentor or link in doc) min
0.3 Read the 3 linked docs (architecture, prompts, and example) very slowly. Take 30–45
notes. min
0.4 Make your own summary: “What are the 4 steps in AI Agent design?” in your 10 min
words

🧠 PHASE 1: Choose a Use Case + Imagine the Agent (90–120 min)

Task Description Time
1.1 Think of 2–3 situations where an AI Agent could help YOU (in studies, writing, 20 min
planning, etc.)
1.2 Choose one. Write a short note: Who is the user? What problem does the agent 20 min
solve?
1.3 Write a few user examples — how would someone talk to this agent? (e.g., "Hey, 15 min
give me a writing idea today")
1.4 Think: What should the agent do internally before replying? (no need for code — 20 min
just logic)

1.5 Write down what the agent should NOT do (helps define the boundary of work) 10 min

🧱 PHASE 2: Design the Prompts One by One (2–2.5 hours)

You will now create a basic prompt for each layer. The goal is not perfection, but logic.

Task Description Time

2.1 INPUT UNDERSTANDING – Try a basic version of “Understand what the user is 20–30 min
asking”
2.2 STATE TRACKER – Can the AI remember past info or simulate memory? Try simple 20–30 min
instructions.
2.3 TASK PLANNER – How should the AI break the request into steps before 30–40 min
answering?
2.4 OUTPUT GENERATOR – What should the final reply look like? Try formatting, 30–40 min
tone, variation.

🔬 PHASE 3: ChatGPT Exploration and Debugging (2–2.5 hours)

​
 Task Description Time
3.1 Run each prompt in ChatGPT (one layer at a time), copy the output 20–30 min
3.2 Note what’s not working, and ask ChatGPT how to improve your prompt 30–45 min
3.3 Try 2–3 improved prompts for each layer 45–60 min
3.4 Save all your prompt versions + outputs in a table or doc 30 min

📄 PHASE 4: Write the Final Submission Document (2–2.5 hours)

Task Description Time
4.1 Section 1–2: Title, Problem, Use Case, What Agent Does 20–30 min
4.2 Section 3: Final version of all 4 prompts (with comments if possible) 30–40 min
4.3 Section 4: Exploration Log (attempts, changes, reasons) 30–40 min
4.4 Section 5: Output examples (Good/Bad inputs) 20 min
4.5 Section 6: Reflections – What you learned, what was tough 30 min

⏱️ Total Estimated Time: ~9 hours​

(Possibly more if done with deep thinking and meaningful iterations.)

💡 Final Tips for First-Time Explorers

Tip Why It Matters

❓ If confused, ask ChatGPT itself what a

prompt should do.
It can teach you how to work with it if you treat it
like a teammate.

🪜 Build prompt logic step by step. Don’t try to write perfect prompts from the start.
Work like a scientist.

📝 Keep a scratchpad next to you. Helps track how your thinking evolved. Makes
documentation easier later.

🚫 Don’t try to “complete the assignment”

fast.
This is a simulation of how DeepThought works.
Curiosity > speed.

🔍 Use analogies to understand prompts. Think of each prompt like a department in a

company — each doing its job.