OpenWISP

There is still hope. Thank you CodeRabbit for this ❤️ United against #AISlop 💪 🦾

I've spent the last two months "using and abusing" every AI tool at my disposal to handle the avalanche of pull requests we're seeing in OpenWISP. Here is my current take: - Coderabbit: Excellent for uncovering a high volume of issues, though it occasionally gets fixated on trivialities. - GitHub Copilot: In my experience, it tends to focus on unimportant details and often misses the real severe issues. - Claude: Generates a fair amount of noise, but it catches major bugs, security flaws, and performance bottlenecks even after Coderabbit has given the "LGTM." - Opencode with Kimi 2.5: Performs similarly to Claude with significantly less noise but also hasn't found bugs as severe as the ones I found with Claude. - ChatGPT & Gemini: Useful in a pinch, but the signal-to-noise ratio makes the other tools preferable. My Verdict One tool alone is never enough. I've ended up using multiple tools at different stages, particularly for non-trivial changes. Even then, manual testing remains non-negotiable, especially for anything touching the UI. The sheer volume of code being churned out by AI agents right now is staggering. Shipping this messy, AI-generated code to production without an in-depth review is a ticking time bomb. Merge without in-depth review at your own risk.

Hi everyone, I’ve been getting a lot of DMs about GSoC lately, mostly the same questions around contributions, proposals, and selection. Instead of replying individually, I’ve written a blog sharing my full journey, what I learned along the way, and what actually mattered in getting selected. 📘 Read it here: 👉 Hashnode: https://lnkd.in/gj4AbarT 👉 Medium: https://lnkd.in/guZfwsUd