AI and Vibe Coding – Early Thoughts and Tools Worth Watching

Its a vibe, catch the wave, windsurf on copilot and let ai move the cursor

Author

Quintes van Aswegen
May 08, 2025

AI is changing how we build software. One of the most interesting emerging patterns is what is called vibe coding.

Think of Vibe Coding as letting an LLM (Large Language Model) sit in the driver’s seat—generating, editing, and shaping software artifacts on demand.

You don’t just get code snippets; you get conversations that result in working features. Sometimes even entire apps. It’s not just autocomplete — it’s AI acting as a creative pair programmer or solo dev, depending on your trust in it.

Vibe Coding: How It Works

With Vibe Coding, the LLM can:

  • Generate new files and functions

  • Modify existing logic based on natural language prompts

  • Handle full-stack requests (e.g. “Add a login page that uses Firebase”)

  • Respond to your context—your codebase, your folder structure, your style

Depending on the platform, the AI may read and write to your repo, track changes, and even commit directly. It’s fast. You ask. It builds.

Some people are using this to scaffold ideas, others are using it in production workflows. It’s part experiment, part shift in how we think about coding.

Vibe Coding: IDEs and Tools Powering Vibe Coding

Here are a few Vibe Coding environments worth exploring:

🤖 GitHub Copilot in VS Code

  • Still the most popular coding assistant

  • Good autocomplete, inline suggestions, and chat

  • Best for: embedded help while staying in your IDE

🏄‍♂️ Windsurf - which has just been bought - OpenAI agrees to buy Windsurf for about $3 billion, Bloomberg News reports see more

  • AI-native workspace built for chat-driven development

  • Structured prompt chains with full-code project creation

  • Great for: end-to-end app scaffolding and fast iterative changes

🧠 Cursor

  • AI-native editor forked from VS Code

  • Built-in memory, context awareness, and multi-step planning

  • Ideal for: serious coding with LLMs as a long-term assistant

🧠 Roo Code / Cline

  • Open-source pair programming

  • Roo is chat-based, Cline is CLI-first with agent-like features

  • Good for: devs who want local-first, transparent LLM interactions

🤝 Aider

  • Git-aware AI assistant for your local repo

  • Can read diffs, apply changes, commit intelligently

  • Excellent for: AI-led refactoring, CLI-first workflows, power users

❤️ Lovable.dev

  • Chat-based project assistant, especially frontend-friendly

  • Structured prompts and hosted workspace

  • Best for: quick MVP UI and beginner-friendly prototyping

🧪 Replit

  • Replit lets you create and deploy apps and sites with AI assistance

🔎 bolt.new

  • Online IDE with AI built in -Bolt is a minimal playground for AI agents that write code in real time

  • Useful for: quick experiments, learning, testing across languages

✨ V0

  • Prompt-to-UI generator by Vercel

  • Converts text prompts into working React/Tailwind UI code

  • Great for: frontend MVPs, idea validation, mockups

🔥 Firebase Studio

  • Not an AI IDE, but relevant—generates UIs, rules, and models based on Firebase schema

  • Can be paired with AI to accelerate backend setup

📋 The hard way - using ChatGPT, Claude or one of the others in the browser and copy-pasting the code generated into your IDE.

  • more control and less risk to your existing code base

  • not as flexible.

Benefits of Vibe Coding

🚀 Speed: Go from idea to working app in hours

🧪 Experimentation: Try out new stacks, frameworks, or patterns with less effort

✅ Validation: Build MVPs and PoCs to test ideas before committing fully

🧠 Learning: LLMs can teach as they code—explaining design choices and concepts

Risks and Considerations

🔐 Security: Generated code may include unsafe patterns or vulnerable dependencies

🛠 Code Quality: Varies by model and prompt; still requires review and testing

🧱 Maintainability: Can become messy without good structure, especially with repeated AI edits

🤖 LLM Drift: AI might rewrite good code unnecessarily or hallucinate solutions

Final Thoughts

Vibe Coding isn’t a buzzword—it’s a signal. Software development is becoming less about knowing what to type and more about knowing what to ask. The better your prompts, the better your results.

I’m still exploring these tools and ways of working having been doing most of my vibe coding using Claude and ChatGPT in browser rather than in the IDE. There’s more emerging the longer I wait to post this.

It’s early days. But the tools above are showing what’s possible—faster iteration, AI-native workflows, and entirely new ways of thinking about building.

Watch this space. Or better yet, vibe into it.

Making sense of it

Need some help figuring out how to make these emerging technologies work for you and your business?

Move your business forward with architectFWD™