Skip to main content

Getting Started with AI Coding Assistants

Your first prompts: what to know, what to try

Introduction​

AI coding assistants are powerful tools for learning and building software. But not all AI tools are equal — and the way you use them matters.

This guide helps you:

  • Understand the difference between chat-based and terminal-based AI tools
  • Write effective prompts for AI-assisted development
  • Leverage tools that can see your code, run commands, and view screenshots
  • Build confidence in AI-assisted learning

Types of AI Coding Tools​

Chat-Based Tools​

Traditional AI interfaces (ChatGPT web, Claude.ai) where you:

  • Copy-paste code into a text box
  • Manually describe your project
  • Copy-paste responses back into your editor

These work, but require constant context-switching and manual work.

Terminal/TUI-Based Tools​

Modern development tools (Claude Code, Codex CLI, OpenCode, Cursor, Aider) that:

  • Read your files directly — No copy-pasting needed
  • Run commands — Execute tests, builds, and scripts
  • View screenshots — See UI errors, design issues, terminal output
  • Make edits — Modify files with your approval
  • Understand your project — See package.json, directory structure, git history

DevFoundry recommends TUI-based tools because they match how developers actually work: in the terminal, with full project context.

Why Terminal-Based Tools Are Better for Learning​

No Context Loss​

Chat-based:

"Here's my function... [copy 50 lines]... and here's the file that imports it... [copy 30 more lines]..."

Terminal-based:

"Look at src/order.ts — why is calculateTotal returning undefined?"

The tool already sees your code. You describe the problem, not the context.

Immediate Feedback​

Chat-based:

1. AI suggests a fix
2. You copy it to your editor
3. You run the code
4. It fails
5. You copy the error back to chat
6. Repeat...

Terminal-based:

1. AI suggests a fix and applies it
2. AI runs the code
3. AI sees the error and iterates
4. You approve the working solution

Visual Debugging​

Terminal-based tools can view screenshots. This enables:

  • "Here's a screenshot of the UI — why is the button in the wrong place?"
  • "This error dialog appeared — what does it mean?"
  • "The layout looks broken on mobile — here's what I see"

Setting Up a TUI Tool​

Claude Code​

# Install
npm install -g @anthropic-ai/claude-code

# Run in your project directory
cd my-project
claude

Codex CLI​

# Install
npm install -g @openai/codex

# Run
codex

Basic Usage​

Once running, you interact conversationally:

You: What files are in this project?

Claude: I can see the following structure:
├── src/
│   ├── index.ts
│   ├── menu.ts
│   └── order.ts
├── package.json
└── tsconfig.json

Would you like me to explain any of these files?

Your First Prompt: Understanding Code​

With a TUI Tool​

I'm learning programming with the devfoundry curriculum.
I'm looking at the lemonade-cli example.

Can you read src/order.ts and explain:
1. What does calculateTotal do?
2. Why does it use reduce?
3. What would happen if items was empty?

Please explain in beginner-friendly terms.

The tool reads the file directly — no pasting needed.

With a Chat Tool (Fallback)​

If you're using a chat-based tool, you'll need to provide context:

I'm learning programming with the devfoundry curriculum.
Here's code from the lemonade-cli example (src/order.ts):

[paste the code]

Questions:
1. What does calculateTotal do?
...

Your Second Prompt: Modifying Code​

With a TUI Tool​

I want to modify the lemonade-cli to add a new menu item:
- "Frozen Lemonade" for $3.25

Please:
1. Show me where menu items are defined
2. Add the new item
3. Run the app to verify it works

The tool will:

  • Find the menu data
  • Show you the change before making it
  • Apply the edit (with your approval)
  • Run npm start to verify

With a Chat Tool (Fallback)​

Here's my menu.ts file:
[paste code]

I want to add "Frozen Lemonade" for $3.25.
Show me the updated code.

Then you manually copy the result back.

Debugging with Screenshots​

One of the most powerful features of TUI tools is screenshot input.

Example: UI Bug​

The order summary isn't displaying correctly.
[attach screenshot of broken UI]

Here's what it should look like:
[attach screenshot of expected UI or describe it]

Can you find and fix the CSS issue?

Example: Error Dialog​

I'm getting this error when I click the submit button:
[attach screenshot of error]

What's causing this and how do I fix it?

Example: Terminal Error​

The build is failing with this error:
[attach screenshot of terminal output]

What's wrong?

What AI Assistants Are Good At​

✓ Explaining concepts in beginner-friendly terms ✓ Generating boilerplate code following patterns in your project ✓ Debugging when they can see your code and errors ✓ Refactoring while maintaining your existing patterns ✓ Running tests and fixing failures iteratively ✓ Reading documentation and applying it to your code

What AI Assistants Struggle With​

✗ Guessing your intent when prompts are vague ✗ Choosing architecture without knowing your constraints ✗ Making decisions that depend on business context you haven't shared ✗ Knowing your preferences for style, naming, etc.

Key insight: AI assistants are collaborators, not oracles. They work with your project context, not in isolation.

Common Beginner Mistakes​

Mistake 1: Being Too Vague​

❌ "Fix my code"

Problem: Which code? What's broken? What should it do?

✅ "The calculateTotal function in src/order.ts is returning NaN. Can you find out why?"

Mistake 2: Providing No Context (Chat Tools)​

❌ "My function doesn't work"

Problem: Chat tools can't see your code.

✅ "Here's my function: [paste code]. Expected output: X. Actual output: Y."

With TUI tools, this isn't an issue — they can see your files.

Mistake 3: Asking for Magic​

❌ "Make this faster"

Problem: Faster how? What's the current performance? What's acceptable?

✅ "This function processes 1000 items and takes 5 seconds. Can you profile it and suggest optimizations?"

Effective Prompts for Learning​

Understanding a Concept​

I'm learning about [CONCEPT] in the devfoundry curriculum.

I understand that [WHAT I KNOW].

I don't understand [WHAT I DON'T KNOW].

Can you explain [SPECIFIC QUESTION] with a simple example?

Example:

I'm learning about React components in the devfoundry curriculum.

I understand that components are reusable UI pieces.

I don't understand how props work.

Can you show me in the lemonade-react example how data flows
from App.tsx to the MenuItem component?

Exploring a File​

Can you read [FILE] and explain:
1. What is this file's responsibility?
2. What are the main functions/components?
3. How does it connect to other files?

Example:

Can you read src/order.ts from the lemonade-cli-typescript example
and explain:
1. What is this file's responsibility?
2. What are the main functions?
3. How does it use the types from types.ts?

Experimenting Safely​

I want to experiment with [EXAMPLE NAME].

Current behavior: [DESCRIBE]

I want to try: [CHANGE]

Questions:
1. Will this break anything?
2. What should I expect to happen?
3. Can you make the change so I can test it?

Iteration and Conversation​

Good AI-assisted development is conversational:

You: Why is the discount not applying in calculateTotal?

AI: Looking at order.ts... The discount logic checks if subtotal >= 10,
    but I see subtotal is calculated after the discount check.
    Would you like me to fix the order of operations?

You: Yes, and add a test to make sure it works.

AI: I'll fix the function and create a test file. Here's what I'm changing...
    [shows diff]
    Should I apply this?

You: Yes.

AI: Done. Running the tests now... All passing.

This flow is natural with TUI tools — impossible with copy-paste chat.

Graduated Prompting Examples​

As you progress through devfoundry, your prompts evolve:

Week 1: Total Beginner​

"What does the console.log() function do?
Show me a simple example in a new file."

Week 4: Understanding Basics​

"In the lemonade-cli example, can you read src/order.ts?
I don't understand what reduce does.
Walk me through it step-by-step with the actual data from menu.ts."

Week 8: Building Features​

"I want to add input validation to the lemonade-cli:
- Quantity must be a positive number
- Item ID must exist in the menu

Where should this validation happen?
Can you implement it and show me how to test it?"

Week 12: Architecture-First​

"I'm planning to add user accounts to the lemonade fullstack example.

Before we write code, can you:
1. Read the current project structure
2. Propose where auth logic should live
3. Identify what new files we need
4. Consider security implications

Then we'll implement it together."

When to Ask for Help​

Good times to use AI:​

✓ Explaining unfamiliar syntax you encountered ✓ Debugging specific errors — especially with screenshots ✓ Understanding concepts from the curriculum ✓ Brainstorming approaches before implementing ✓ Reviewing your code for issues ✓ Running and testing changes iteratively

Times to struggle first:​

⚠ Exercises — Try yourself before asking (that's how you learn) ⚠ Design decisions — Think through tradeoffs first ⚠ Simple debugging — Check obvious things (typos, file paths) first

Balance: Use AI to accelerate learning, not replace thinking.

Best Practices​

1. Let the Tool See Your Project​

Start the TUI tool in your project root. Let it understand your structure.

2. Be Specific About Goals​

Not "fix this" but "make this function return X when given Y."

3. Ask for Explanations​

Don't just accept code changes. Ask:

  • "Why did you use const instead of let?"
  • "What does this line do?"
  • "What would happen if I removed this?"

4. Review Before Approving​

TUI tools show you changes before applying them. Read the diffs. Understand what's changing.

5. Use Screenshots for Visual Problems​

UI issues, error dialogs, terminal output — a picture is worth a thousand words of description.

Summary​

AspectChat ToolsTUI Tools
Code accessCopy-pasteDirect file access
Running commandsManualAutomatic
Visual inputDescribe in wordsScreenshots
Iteration speedSlow (copy-paste loop)Fast (approve and run)
ContextProvide each timePersistent project context

For DevFoundry learners: Start with a TUI tool if possible. The experience is dramatically better, especially when learning.

Quick Reference​

Starting a Session​

cd my-project
claude  # or codex, or your tool of choice

Understanding Code​

Read [file] and explain [specific question]

Making Changes​

I want to [change]. Can you implement it and run the tests?

Debugging​

[describe problem or attach screenshot]
What's causing this?

Learning​

I don't understand [concept].
Show me an example in [file] from this project.

The best prompts give AI context and ask specific questions. With TUI tools, the context is automatic — you just need to ask.