Skip to main contentVue-doo Magic AI Development Tricks by Daniel Kelly
- Why AI? Produce code more quickly, surface solutions you might not think of, helps you think at a higher business level
- Using AI to write code — Cursor
- Generates code in multiple files
- Runs commands
- Manage/revert with git
- Writes file to system before accepting changes so you can preview output
- Generating components, tests, or full features
- Create a rules file
- General Workflow Tips
- Generate a task list and ask AI to mark off tasks one at a time
- Don't generate everything from scratch. Use libraries
- Provide screenshots
- Commit early and commit often
- Don't trust the bot
- Don't try to do everything at one time
- Use .cursorrules or project rules for preferences and patterns
- Copy and paste examples of your API payloads, JSON data, etc to pass the LLM
- Make sure MCP is from a secure source as it can run commands on the machine
- Context7 MCP
- Using AI inside of the actual code
- Use Cases
- Extracting text from an image, generating image alts
- Translating text to a different language
- Semantic search with RAG
- Summarizing content
- Generating images and videos
- Totally rethinking how users interact with your app
- Use the right model for the right task to save money
- Provide examples in the prompt
- Ask for JSON-only responses from data (or XML)
- AI SDK
- Unifies API for different providers
- Support for tool calls
- Supports streaming
- AI for TypeScript
- Issues
- Mitigate latency with streaming or queues as it is usually slow
- JSON Reader — stream JSON incrementally
- Chunk larger requests into smaller pieces. Instead of 30 questions, 10 at a time
- Tools
- database.build
- Supabase SQL AI — write queries, debug query issues, setup RLS policies, and more
- Photoshop — generative fill, object removal, generative expand
- Eleven Labs — convert text to (fairly convincing) speech
- Runway — video generation
- danielkelly.io