[Panel] Vibe Coding: Yay or Nay?
summary
Panelists debate concrete gains and risks: prototypes that once took weeks now appear in hours using Cursor, Claude Code and GPT-5, a walking-phone voice-transcriber gets built and runs, and an IIT course uses GPT-5 mini to auto-evaluate student repos and generate follow-up tasks. Problems include untraceable agent decisions, hallucinated Terraform/infrastructure, brittle autogenerated tests, and licensing/IP uncertainty. Practical fixes proposed: store prompts.md and spec.md in repos, generate tests first, commit prompts and git history for provenance, and prefer code for determinism.
AI-written summary · about
chapters (19)
- 0:00Panel introduction
- 0:44What is vibe coding?
- 3:12Vibe vs AI coding differences
- 6:09Tools and ethical concerns
- 8:40Impact on new developers
- 10:51Vibe evaluation in education
- 12:22LLM evolution and testing
- 13:10Prompting strategies
- 15:23Why vibe coding is attractive
- 17:09Dependency on LLMs
- 18:36Model improvements
- 20:02Interface preferences
- 22:37Vibe coding loops
- 24:44Debugging and iteration
- 25:49Tool limitations
- 26:53Codebase management
- 27:58Responsible coding practices
- 29:25Responsible vibe coding
- 32:20Q&A session
