Being a Vue Developer in the Age of AI Coding Assistants by Erik Hanchett
One of the first things brought up is how we don't write code anymore. Although, even before AI, I can't say I wrote all my code by hand, not in the way I think about it. I did lots of copy pasta, tweaking, deleting, using snippets to generate templates.
Our expertise and taste are more valuable than ever before (takeaway: keep learning)
Where to begin?
Pick a harness and a model (Kiro, Claude Code, Cursor, etc)
Local models
CLI over UI, Type over Talk
Create rules for the AI. Include projects purpose, architecture, coding standards. Keep it brief. Living document
Use skills (github.com/vuejs-ai/skills)
github.com/antfu/skills
Be careful about online skills. This is a solid attack vector (skills.sh has many regular skills but again be careful)
Create agents or plugins
Setup an Agent Workflow - research, requirements, design, implementation, test
Don't go past 2 agents running at the same time
Setup Guardrails
Tests or Vitests
Linting
CI Checks
Code Coverage
Error Tracking
Pre-commit hooks
Code Reviews are the real bottlenecks
Spec Driven Development
Spec-Kit
Claude plan mode
grill-me is a cool spec that asks clarifying questions
Vibe code prototypes
Action Items
Take a look at some of the links ahead
There's a Claude Code skill for creating skills.
Based on the code review checkpoint, how can we make code reviews better and more efficient? What is a good workflow?
I found for a lot of items, I could have Claude create seed data for different scenarios to take a look at them in the UI (the 'empty' case, a bunch of data, standard use, etc). I wonder if there is a way to streamline this
I would ask Claude to summarize changes, tell me how to test changes (if not provided by the developer), walk me through code, etc.