The debate is wrong
Most discussions about AI in programming are binary: either "AI will replace developers" or "AI is garbage and always fails." Both miss the point.
AI is a tool. Like a compiler, an IDE, or Stack Overflow. The question isn't whether to use it, but how to use it effectively.
What I actually did
Yesterday I built depx, a CLI tool in Rust that analyzes JavaScript/TypeScript projects to understand what's really in your node_modules:
- Find packages installed but never imported
- Explain why any transitive dependency exists
- Check vulnerabilities that actually affect your versions
- List deprecated packages
I shipped it to crates.io, posted on Reddit, got feedback from a user with 25k packages saying my audit command would make 25,000 API requests. Within hours, I implemented batch queries (v0.2.0) reducing it to ~25 requests.
How AI fit into this
I used Claude throughout the process. Here's what that actually looked like:
What I did:
- Identified the problem (node_modules chaos)
- Defined the architecture (analyzer, graph, lockfile parser, vulnerability checker)
- Made technical decisions (use oxc_parser, petgraph, OSV API)
- Evaluated the output and caught edge cases (@types packages, build tools)
- Tested on real projects
- Responded to user feedback and prioritized the fix
What Claude did:
- Wrote code faster than I could type
- Implemented the structure I defined
- Helped debug issues
- Generated boilerplate I would have written anyway
The key insight
People who say "AI writes bad code" are often asking it to replace them. They prompt "build me an app" and get garbage.
People who use AI effectively treat it as a pair programmer. You drive, AI accelerates. You still need to:
- Understand the problem deeply
- Know what good code looks like
- Evaluate if the solution is correct
- Take responsibility for the result
Transparency
If you look at depx's GitHub, you'll see Claude listed as a contributor. That's intentional. I'm not hiding that I used AI — I'm showing that using AI doesn't mean you didn't build something real.
The tool works. It solves a real problem. Users are giving feedback and I'm shipping improvements. That's what matters.
The real question
Stop asking "should developers use AI?"
Start asking "what can I build now that I have this leverage?"
Top comments (0)