Hacker News
This article provides a practical guide for developers looking to get started with AI coding tools, breaking down different AI development modes and offering advice on which tools to try first.
AI 生成摘要
本文為開發者提供了入門 AI 編碼工具的實用指南，闡述了不同的 AI 開發模式，並建議了應優先嘗試的工具。
You took a few months away from coding -- maybe parental leave, a sabbatical, or just a busy period -- and suddenly everyone's talking about AI pair programming, autonomous agents, and tools that write entire features. Meanwhile, your current project forbids AI usage entirely.
Sound familiar?
This exact scenario is playing out for thousands of developers right now. The AI development landscape exploded over the past 18 months, and if you haven't been actively experimenting, it can feel overwhelming to catch up.
The good news? You don't need to learn everything at once. This guide breaks down exactly how to get started with AI coding tools, which tools to try first, and what types of projects work best for learning -- whether you're a .NET developer, JavaScript enthusiast, or working in any other stack.
Disclaimer: Yes! This was written by AI based on all of the content discussed in the video above, by yours truly, Nick Cosentino, while filming on Code Commute. I have ensured that it represents my thoughts and perspectives.
Before diving into specific tools, it's crucial to understand that AI can assist your development workflow in three distinct ways. Each serves a different purpose and requires different skills.
Think of this as having a senior developer available 24/7 to answer questions and explain concepts.
Best for:
Tools: ChatGPT, Claude, Google Gemini, GitHub Copilot Chat
Example workflow:
You're building a habit tracker app and need to add authentication. Instead of Googling through 15 blog posts, you ask ChatGPT:
"I'm building an ASP.NET Core habit tracker. What's the best way to add authentication? I want to keep it simple for now but scalable later."
The AI explains your options (Identity framework, Auth0, custom JWT), the tradeoffs, and can even follow up if you say "I still don't get it -- explain it differently."
Critical rule: Never blindly trust AI responses. Validate answers the same way you'd verify advice from a colleague. AI tools can be confidently wrong.
This is where AI writes code while you provide clear instructions and supervise the execution.
Best for:
Tools: Cursor, VS Code with Copilot, Claude Code, Visual Studio with Copilot
Example workflow:
You have a Blazer app with a basic structure. You tell Cursor:
"Create a new page called 'Habit List' that displays habits in a table. Don't worry about the database yet -- just stub out the data calls."
The tool creates a plan, shows you the steps, and generates the code. You watch, interrupt if it goes off track, and learn from seeing the implementation.
Why supervision matters: Without guidance, AI tools can generate working code that doesn't match your architecture, uses outdated patterns, or makes assumptions you didn't intend. Watching the process helps you catch these issues early.
This is the "hands-off" approach where you provide detailed requirements and let AI build entire features with minimal intervention.
Best for:
Tools: GitHub Copilot Workspace, Claude with detailed prompts, advanced Cursor workflows
When to wait: If you're still learning AI-assisted development, save this mode for later. It's frustrating to debug code you don't understand, and you'll learn less about both your codebase and AI capabilities.
The AI tool landscape is overwhelming, with new options launching weekly. Here's a pragmatic starter stack focused on tools that integrate into actual developer workflows.
Purpose: Conversational exploration and learning
Best for: Understanding concepts, exploring problem spaces, getting unstuck
Cost: Free tier available
Why it matters: Great for questions and explanations, but doesn't have access to your codebase context. Use it for conceptual learning, not code generation in active projects. You're able to pretty much pick your favorite LLM to use here.
Purpose: AI-native code editor (VS Code fork)
Best for: Developers who want cutting-edge AI integration out of the box
Cost: Free tier, paid plans for advanced features
Why it matters: Built from the ground up with AI in mind. Includes agentic workflows, codebase-aware chat, and excellent context management.
Purpose: Industry-standard editor with AI extensions
Best for: Developers already invested in the VS Code ecosystem
Cost: Free editor, extension costs vary
Why it matters: Similar capabilities to Cursor through extensions like GitHub Copilot and Claude. Choose based on your existing preferences.
Purpose: The complete .NET development environment
Best for: .NET developers who prefer full-featured IDEs
Cost: Free Community edition, paid Professional/Enterprise
Why it matters: If you're a .NET developer who's used Visual Studio for years, the AI integration through Copilot feels natural. No need to force yourself into VS Code if it doesn't fit your workflow.
Purpose: Terminal-based AI coding assistant
Best for: Developers comfortable with command-line workflows
Cost: Requires Claude API access
Why it matters: As of early 2026, the developer community is enthusiastic about Claude Code's capabilities, particularly for complex refactoring and architectural work. However, if you prefer GUI tools, Cursor or VS Code offer similar functionality with visual interfaces.
Start with what you already use. If you're already in VS Code, add GitHub Copilot or Claude extensions. If you use Visual Studio for .NET development, start there. If you want a fresh start with AI-first design, try Cursor.
Try both conversational and integrated tools. Use ChatGPT or Claude for learning, and Cursor or VS Code for actual coding. They serve different purposes.
Don't try everything at once. Pick 2-3 tools and use them consistently for a month. Tool-hopping prevents you from building real proficiency.
Here's the skill that separates developers who love AI tools from those who hate them: context management.
When you tell an AI tool to "refactor the logic that adds numbers together," what if your codebase has five different places that add numbers? The AI doesn't know which one you meant unless you provide context.
When your project is small (like a side project or learning app), you can point the AI to the entire codebase. The tool can usually figure out what you mean.
In production codebases with thousands of files, giving the AI everything creates problems:
Solution: Explicitly provide context by:
Example of good context:
"In HabitService.cs, refactor the GetUserHabits method to use the repository pattern. Keep the same return type and don't modify the database call yet."
Example of poor context:
"Refactor the habit code to use better patterns."
Forget building the next Facebook. The best learning projects are intentionally simple, allowing you to focus on practicing AI workflows rather than solving complex domain problems.
Habit Tracker
Calendar/Event Manager
Todo List with Extras
Simple Calculator (Multiple Implementations)
Looking for more beginner-friendly project ideas? Check out these 7 C# Project Ideas For Beginners that can help you escape tutorial hell while practicing AI-assisted development.
One of the biggest mistakes developers make with AI tools is asking for too much at once. The more you request in a single prompt, the more opportunity for AI to:
Let's say you want to build a "Habit List" page in your Blazor app. Instead of:
❌ Poor prompt: "Build me a complete habit tracker with list view, add/edit forms, and database integration"
✅ Better sequence:
Why this works:
If you're unsure how to break a feature into small tasks, use conversational mode to help:
"I want to build a habit list view feature in my Blazor app. What steps would you recommend to build this incrementally, keeping each step small and supervisable?"
The AI will propose a breakdown. You can refine it, ask for clarification, and then use that plan to guide your supervised co-pilot work.
If you're in the .NET ecosystem, you have multiple ways to build web apps, and AI tools can help you navigate the choices.
Best for: Staying in C# for both backend and frontend
AI advantage: AI tools understand Blazor component patterns well
Learning curve: Moderate - new concepts but familiar language
When to choose: You want to avoid JavaScript entirely, or you're building internal tools where Blazer's approach makes sense.
Best for: Modern web architecture with separate frontend
AI advantage: AI excels at both backend API and frontend framework code
Learning curve: Higher - two languages and frameworks
When to choose: You want to learn modern JavaScript frameworks, or you're building something that might need a mobile app later (shared API).
Getting started: If you're new to ASP.NET Core, this guide on setting up Visual Studio for ASP.NET Core walks through the proper workload configuration and project setup.
Use conversational mode to decide: Not sure which approach fits your goals? Ask Claude or ChatGPT to walk through the tradeoffs based on your specific scenario. Be explicit about your learning goals, timeline, and where you want to invest your effort.
Problem: Too much autonomy, not enough guidance
Fix: Break requests into smaller pieces, provide explicit constraints ("don't modify the database layer yet")
Problem: Operating in autonomous mode without supervision
Fix: Switch to supervised co-pilot mode, watch what it generates in real-time, interrupt when it strays
Problem: Missing context about your codebase conventions
Fix: Include example code or explicitly state the pattern ("use repository pattern like in UserService.cs")
Problem: Autonomous generation of complex features
Fix: Ask the AI to explain what it built and why. If you still don't get it, have it rebuild using simpler approaches or break it into smaller pieces you can understand.
Problem: Too much codebase context, AI is guessing
Fix: Explicitly select files or reference specific paths in your prompt
AI coding tools are powerful, but they're not infallible. Develop the habit of treating AI-generated code the same way you'd treat code from a junior developer in a code review.
Ask yourself:
When AI hallucinates: Sometimes AI tools confidently state incorrect information or generate code that doesn't work. This isn't a fatal flaw -- it's expected behavior. Your job is to catch these mistakes through review and testing, just like you would with human-written code.
Here's your action plan to get started with AI-assisted development this week:
Don't try to learn everything. Spend 2-4 weeks with this setup before adding more tools or techniques.
Track what works and what frustrates you. Over time, you'll develop intuition for when to use conversational mode vs. supervised mode, how much context to provide, and which tools fit your workflow best.
No. ChatGPT has a free tier, Claude offers free access, and many code editors have free AI extensions. You can learn the fundamentals without spending anything. Paid tiers offer faster responses and more advanced models, but aren't required for learning.
That depends on your company's policies. Some workplaces restrict AI on specific projects but allow it for learning and side projects. Others have blanket policies. Check with your manager or legal team. For learning purposes, work on personal projects outside of work hours using your own equipment.
Absolutely. The three modes (conversational, supervised co-pilot, autonomous), context management principles, and project recommendations apply regardless of your stack. The specific tool recommendations might vary (e.g., Python developers might prefer different IDEs), but the core concepts remain the same.
Start by asking yourself: "Do I understand what this code does?" If not, ask the AI to explain it. Then consider: Does it solve the problem? Does it follow patterns from the rest of my codebase? Does it handle errors appropriately? Over time, you'll develop intuition, just like learning to review human-written code.
This is a false dichotomy. AI tools are most effective when wielded by developers who understand fundamentals. Use AI to accelerate learning and reduce boilerplate, but don't skip understanding core concepts. If the AI generates code you don't understand, that's a signal to slow down and learn the underlying concept first.
Getting started with AI coding tools isn't about replacing your skills -- it's about augmenting them. Think of AI as an always-available coding partner who can explain concepts, generate boilerplate, suggest approaches, and help you explore unfamiliar territory faster.
The developers who succeed with AI tools are the ones who:
The AI development landscape will keep evolving. New tools will launch. Capabilities will improve. But the fundamental skills -- breaking down problems, providing clear requirements, reviewing code critically, and iterating based on feedback -- remain timeless.
Start today. Pick one tool. Build something simple. Learn by doing.
And remember: everyone feels behind when a new technology wave hits. The fact that you're reading this guide means you're already taking the right step.
Welcome to another issue of Dev Leader Weekly! In this issue, I discuss ways you can leverage AI as a developer and things to avoid too!
Welcome to another issue of Dev Leader Weekly! In this issue, I discuss my take on AI tools and getting started as a software developer.
Welcome to another issue of Dev Leader Weekly! In this issue, I discuss the usage of AI for junior developers just getting started.