In my previous post titled "So, Is AI Commoditizing Coding?" I explored the idea that AI coding assistants might be turning coding into a commodity. Inspired by Oz Nova's recent thoughts on AI coding assistants, I decided to put this notion to the test using Bolt.new, an AI-powered platform developed by StackBlitz.
A Magical Experience with Bolt.new
I had been holding off on building five apps due to time constraints. With Bolt.new, I was able to bring these ideas to life in just a few minutes. It's like having a magic wand that instantly transforms ideas into prototypes! My first impression was one of amazement at how effortlessly it created near-functional apps and deployed them to Netlify—all without writing a single line of code.
Understanding Bolt.new and StackBlitz
To appreciate Bolt.new, it's essential to understand its creator, StackBlitz. StackBlitz is an instant full-stack web IDE for the JavaScript ecosystem, powered by WebContainers—the first WebAssembly-based operating system that boots the Node.js environment in milliseconds, securely within the browser.
The team at StackBlitz launched Bolt.new on October 3, 2024. According to its Twitter(X) profile:
"With bolt.new you can prompt full-stack web applications into existence, see them executed in real-time, debug errors as they occur, & deploy a fully functional app—all without ever leaving your browser or personally writing a single line of code!"
Putting Bolt.new to the Test
I decided to see if these claims held true:
- Prompting Full-Stack Applications into Existence
I successfully prompted five full-stack web applications using natural language descriptions. Bolt.new translated these prompts into working code, generating the front-end and back-end components seamlessly.
- Real-Time Execution
Bolt.new executed the apps in real-time, allowing me to see immediate results.
- Debugging Errors
When errors occurred, Bolt.new highlighted them, enabling me to prompt it by clicking "fix problems" button to debug on the spot without delving into the actual code right within the browser.
- Effortless Deployment
Bolt.new deployed all five apps to Netlify automatically, providing me with live URLs without any manual deployment steps.
This seamless experience was phenomenal—it essentially gave me the URL to the deployed app and showed it running in the browser.
My Take on Bolt.new
I believe Bolt.new is incredibly useful for experimenting and prototyping ideas, whether you're an experienced software engineer or new to coding. By using natural language prompts (English language), it lowers the barrier to entry and accelerates the development process.
How Bolt.new Differs from Other AI Coding Tools
Unlike AI coding assistants like Cursor that offer code suggestions or completions, Bolt.new builds the entire app based on your prompts. If you're directly coding or developing, tools like Cursor are fantastic, but Bolt.new stands out by accommodating a range of use cases—from small proofs of concept to production-ready apps.
The closest tool to Bolt.new I've come across is GitHub Spark, announced on October 29, 2024. However, Spark focuses on simple micro-apps without third-party dependencies or frameworks, whereas Bolt.new handles more complex applications.
The Upsides
- Speed and Efficiency: It works better than any other AI (ChatGPT, Gemini, Claude) I've tried for creating functional apps, delivering results in minutes rather than hours.
- Integrated Environment: The combination of IDE, code assistant, design preview, and deployment in one platform is remarkable.
The Downsides
Despite its impressive capabilities, I encountered some challenges:
- Token Consumption: After a few simple requests, I exhausted my token limit. Each action consumed a large number of tokens (sometimes 125k per request), leading to additional costs.
- Debugging Loops: The AI sometimes got stuck in debugging loops, consuming tokens without resolving issues.
- File Overwrites: Instead of updating specific lines, it often rewrote entire files, which was inefficient.
- Cost Concerns: The pricing plans can be steep, especially when tokens are consumed rapidly due to the issues mentioned.
Pricing Plans Overview
- Pro ($20/month): 10M tokens
- Pro 50 ($50/month): 25M tokens
- Pro 100 ($100/month): 50M tokens
- Pro 200 ($200/month): 100M tokens
Users need more clarity on token spending and limits, as the number of tokens doesn't convey practical usage effectively.
Recommendations
For non-technical users with zero coding knowledge, Bolt.new offers a way to build an MVP in minutes. However, the challenges I faced suggest that it's not yet fully ready for business-critical applications without technical oversight.
Tips to Save Tokens
Based on my experience:
- Start with Clear Requirements: Write detailed prompts or use tools like ChatGPT to refine them before inputting them into Bolt.new.
- Use External Help for Troubleshooting: If you encounter errors, consider using another AI assistant to generate prompts that fix issues efficiently.
-
Common Fix Prompts:
-
Image Not Loading: "Please fix the image paths so all images display correctly."
- Responsiveness Issues: "Update the code to ensure the website is fully responsive on all devices."
- Slow Load Times: "Optimize resources to improve website loading speed."
Reflecting on Oz Nova's Thoughts
Oz Nova, the founder of the Bradfield School of Computer Science and curator of TeachYourselfCS.com and CS Primer, shared his perspectives on AI coding assistants. He emphasizes the value of foundational skills and cautions against overreliance on AI tools:
"For almost all meaningful tasks, I don’t use AI. I use my brain, pen and paper, a minimally configured text editor, and something to run my code."
He questions whether large language models (LLMs) represent a significant abstraction comparable to historical innovations like compilers:
"I would ask LLM hyperenthusiasts to question the degree to which LLMs are good abstractions... If such abstractions don't eventuate, LLMs may end up only as tools, with an impact comparable to IDEs or Stack Overflow, but well short of the hype."
My Verdict
I agree with Oz Nova. While AI tools like Bolt.new are impressive, they haven't yet commoditized coding. They serve as accelerators and aids but don't replace the need for a deep understanding of programming concepts—those that impact user experience, performance optimization, and security.
Building something real requires more than just generating code; it demands comprehension, problem-solving, and the ability to adapt. The initial magic of AI-generated code fades when you need to make changes or add features.
Conclusion
AI is reshaping the way we approach coding, but it's not eliminating the need for skilled programmers. Tools like Bolt.new are valuable for rapid prototyping and can inspire innovation, but they also highlight the importance of foundational coding skills.
For those learning to code, focusing on building foundational skills and problem-solving techniques remains crucial. AI can augment our abilities, but the craftsmanship of coding is still very much a human endeavor.