AI Coding Tools That Changed How I Work

1. Introduction

Full disclosure: I'm new to the AI/ML space. I started as a traditional software engineer building Java backends and React frontends. AI tools were something I read about, not something I used daily.

That changed in late 2025. I decided to actually learn this stuff—not just the hype, but the practical tools that could help me ship faster. This is my honest take on what actually works versus what's just marketing.

2. Cursor: The IDE That Changed Everything

Cursor was my gateway drug into AI-assisted coding. It's VS Code with AI baked in, but that description undersells it massively.

What Makes It Different

The killer feature is Ctrl+K—inline editing. You select code, describe what you want, and it rewrites in place. No copy-pasting between ChatGPT and your editor. No context switching.

The Composer feature (Ctrl+I) is next level—it can make changes across multiple files. I've had it refactor a component, update the types file, and modify the tests in one go.

What I've Learned

• Be specific with instructions—vague prompts give vague code
• Always review the diff before accepting
• It's great for boilerplate, bad for architecture decisions
• The context window matters—reference files explicitly

3. OpenHands: Autonomous Coding Agent

OpenHands (formerly OpenDevin) is different from Cursor. It's not an IDE—it's an agent that can actually do things. Run commands, create files, browse the web, execute tests.

When It Shines

I use OpenHands for tasks that would take hours of context switching:

• Setting up new projects with boilerplate
• Writing tests for existing functions
• Debugging with access to actual runtime
• Creating documentation from code

The Tradeoff

It's slower than Cursor. Sometimes it gets stuck in loops. But when it works, you come back to a completed task that would have taken you hours.

4. GitHub Copilot: The Reliable Standby

Copilot was my first AI coding tool, and I still use it daily. It's not flashy, but it's reliable.

Where Copilot Excels

• Auto-completing obvious code (the tedious stuff)
• Writing tests from function signatures
• Generating boilerplate implementations
• IDE integration is seamless

Where It Falls Short

It doesn't understand your codebase. It predicts based on the current file and maybe a few open tabs. Complex refactoring? Not happening. Cross-file changes? You're better off with Cursor or doing it manually.

5. Other Tools Worth Mentioning

6. My Current Workflow

After experimenting for months, here's what actually stuck:

• Day-to-day coding: Cursor with Claude Sonnet. Fast, inline, minimal friction.
• Complex tasks: OpenHands when I need something that can run commands and tests.
• Quick suggestions: Copilot autocomplete is still useful for obvious completions.
• Learning/exploring: Chat with Claude about architecture, then implement in Cursor.

The Honest Truth

I still write code manually. AI helps with the obvious stuff, but the interesting problems—system design, debugging weird edge cases, making tradeoffs—still need human judgment.

7. The Reality Check

Here's what the marketing doesn't tell you:

AI Tools Are Not Magic

• They hallucinate. Always review generated code.
• They don't understand your business logic.
• Context limits are real—large codebases confuse them.
• Sometimes the old way is faster for simple changes.

What Actually Improved

• Speed on boilerplate: 10x faster on CRUD, migrations, tests.
• Exploration: Can understand unfamiliar codebases faster.
• Consistency: AI follows patterns better than tired me at 2am.

What Didn't Improve

• Architecture decisions—still need human judgment
• Debugging complex race conditions
• Understanding business requirements

8. Conclusion

AI coding tools are genuinely useful. Not in the "replace developers" way, but in the "handle tedious stuff so developers can think" way.

If you're new to this like I was, start with Cursor. It's the lowest friction entry point. Then explore agents like OpenHands when you need more autonomous help.

The key is knowing what each tool is good at. Using the wrong tool for the job leads to frustration and the conclusion that "AI is overrated." It's not overrated—it's just not magic.