978: Should A New Coder Use AI?

What It Covers

Scott and Wes answer listener questions about running multiple AI coding agents simultaneously, whether beginners should use AI tools, different pagination strategies, transitioning from outdated tech stacks like PHP and jQuery, managing side projects with family responsibilities, and implementing RAG search with transformers JS for personal portfolio sites.

Key Questions Answered

• •AI Agent Management: Running multiple AI agents simultaneously creates more overhead in planning and code review than actual generation time. The longest parts are defining requirements upfront with proper context, code styles, and problem specifications, then thoroughly reviewing output for duplicated utilities and pattern violations. Most developers overlap agents by planning next tasks while one runs rather than managing five concurrent agents.
• •AI for New Developers: Beginners should use AI as a research and debugging accelerator rather than a code generation crutch. Use it to understand error messages, explore how systems interact, and break down complex documentation into diagrams and bulleted lists. Avoid simply prompting features without understanding implementation, similar to using calculators before learning multiplication—you need foundational problem-solving skills that come from working through concepts manually first.
• •Pagination Implementation: Offset pagination works for static content with page numbers, showing items 1-10 then 11-20. Cursor-based pagination prevents duplicate content when data updates frequently by using the last item's ID as reference point rather than numeric offsets. Infinite scroll needs URL updates via push state to preserve scroll position on refresh, preventing user frustration from losing their place after accidental navigation.
• •Upskilling from Legacy Stacks: Developers working with PHP and jQuery need to build side projects using current employer-desired technologies rather than personal preferences. Focus on React, Drizzle or Prisma ORMs, Tailwind CSS, Zod validation schemas, and AI API integration. These represent keyword-stuffable skills that pass recruiter filters and demonstrate understanding of modern patterns like validation schemas and transactional email services beyond basic SMTP.
• •Side Project Time Management: Successful side project execution requires diving deep into one topic until mastery, then moving to the next rather than maintaining ten simultaneous interests. Eliminate activities that don't align with personal interests—not knowing pop culture or sports creates hours for technical exploration. Asynchronous learning through podcasts during commutes and dead time compounds knowledge without dedicated blocks. Build projects that solve real family problems to justify time investment.
• •RAG Implementation Challenges: Retrieval augmented generation fails when context windows are too narrow. When vectorizing documents like resumes, retrieve not just the matching utterance but two sentences before and after to provide surrounding context. Small in-browser models from transformers JS work better for specific tasks like toxicity detection or translation than general chat, which requires larger models with better reasoning capabilities to properly synthesize retrieved context.