How I Used MCP in Development#
Tools used: Claude, Cursor, Notion, Slack, Figma
- Slack was used to surface new tasks that needed to be created
- Each feature was documented in Notion
- The Notion content was then used to generate corresponding Figma files
- Figma components were read and turned into code (markup + internal business logic)
- Before deployment, test code was generated and QA/QC was run
- The resulting output was summarized and stored back into Notion
The result?
The output wasn’t perfectly smooth, but it was workable enough to demonstrate end-to-end automation. While the generated results are generally not production-ready, it’s surprisingly viable for early-stage MVPs. This workflow shows real promise as a communication and handoff tool between planners/POs, designers, and developers—more effective than many existing methods today.
Recent Thoughts on AI#
Will developers be largely replaced in the next 1–2 years?#
Why I think so:
- After seeing what MCP can already do, I believe most of what we currently build on the web can be software-generated
- Increasingly, companies are freezing or cutting entry-level developer hiring
I wonder how long it will take for this shift to reach Korea—or directly impact people like us. It might not be that far off.
Vercel’s v0.dev is said to be underhyped
I’ve only tried the dashboard briefly, but it’s great for building FE MVPs. Still, for a more fully-featured app, I felt refine.ai might be a better fit based on recent experience.