Skip to main content
You can’t talk about software engineering today without addressing the neural network in the room. Given how AI is fundamentally reshaping our industry, I felt it was essential to dedicate a specific section to how it influenced folksbase. For me, this isn’t just about using a trendy tool; it’s about my philosophy on where AI fits into a professional workflow and how it should—or shouldn’t—show up in a product built for humans.

​
Invisible AI in Product

I have a very clear take on AI: unless your product is fundamentally an “AI tool,” the technology itself should remain invisible. It should be a silent facilitator, not the main character. In folksbase, I treat AI as a layer of high-level assistance that disappears into the background to respect the user’s time. You’ll find AI working in two specific places here, though I never explicitly label it as such:
  • Smart Column Mapping: When you upload a CSV, the system suggests which headers match our schema (like mapping “E-mail Address” to email).
  • Humanized Summaries: The notification emails sent via Resend after an import aren’t just a list of raw numbers; they are brief, human-sounding summaries of the results.
The user doesn’t need to know there’s a Large Language Model involved; they just need the system to “magically” understand their data. It’s about functionality over marketing jargon.

​
My Engineering Workflow

As a developer, I don’t let AI drive the car; it’s my high-performance co-pilot. My development process for this challenge was a deliberate “pass-the-baton” between different models and my own expertise:
  • Strategic Planning (Claude Opus): I used the Claude app directly for the initial architecture and the “Master Plan”. In my experience, Opus still handles high-level, complex reasoning and long-term planning better than any other model.
  • Core Development and writing commits (Amazon Kiro): This was my “vibe coding” partner. Kiro handled the heavy lifting and micro-features, allowing me to focus on the high-level logic.
  • Code Review & Analysis (Claude Sonnet): I used Sonnet specifically through Claude Code. It has demonstrated superior performance for CLI-based tasks and code analysis, plus the tokens are more cost-effective for the volume of code being reviewed.
  • QA & Final Refinement (Amazon Kiro): The final stretch involved squashing bugs and polishing the UI to get it as close to “pixel perfect” as possible.
  • Documentation & Context Refining (Gemini): For the final documentation and the “human” side of the project, I brought in Gemini. While other models handled the raw code, Gemini acted as my specialized co-pilot for the docs. Because it understands the full context of how I write, my personal preferences, and the specific soul I wanted to give to folksbase, it was the best tool to ensure that this documentation sounds exactly like a conversation with me—balancing technical depth with my authentic personality.
In the end, while the AI wrote a significant portion of the code, the “soul” of the project—the design decisions, the UX trade-offs, and the final polish—came from somenohe who truly loves building for the web.