978: Should A New Coder Use AI?

Episode Summary

Executive Summary: This potluck episode covers how to use AI coding tools effectively without becoming dependent on them, when different pagination patterns make sense, how much AI support is appropriate for newer developers, gaps in formal CS education versus real-world web work, an XSS attempt against the show, career advice for outdated stacks, balancing side projects with family, and a brief look at RAG/local LLMs plus podcast programming decisions.

Main Topics: Using multiple AI agents responsibly: The hosts argue that the hard parts of AI-assisted coding are planning, context-setting, and review—not code generation itself. They favor one agent at a time for important work, overlapping it with other tasks, and using multiple agents mainly for low-risk or repetitive jobs. Pagination vs infinite scroll: They explain offset pagination, cursor-based pagination, and infinite scroll, emphasizing that feeds suit infinite scroll while large, searchable catalogs and frequently changing content often need pagination or cursor-based approaches. How new coders should use AI: For beginners, AI should be a research and learning accelerator, not a replacement for understanding. The hosts stress that developers still need to know how systems work, debug effectively, and validate AI output. Education gaps and real-world web development: A listener asks what school did not teach that matters day to day. The hosts say university provides theory and thinking skills, but practical tools, libraries, deliverability, and evolving workflows are learned on the job and through the community. XSS attempt and web security basics: They inspect a malicious potluck submission that attempted cross-site scripting via a script tag and cookie exfiltration. This is used to explain how XSS works and why frameworks and sanitization matter. Career growth from outdated stacks: A listener stuck on PHP and jQuery is advised to skill up on modern stacks and build side projects aligned with current hiring demand, such as React, Prisma/Drizzle, Zod, Tailwind, and AI integrations. Managing side projects, family, and learning: They describe balancing kids, work, and hobbies by making side projects useful to daily life, learning in bursts, and using asynchronous time. The hosts also talk about hobby devices and electronics as practical outlets.

Key Arguments: AI coding tools are most valuable when the developer does the thinking up front: define the problem, establish patterns, and provide strong context before asking the model to generate code. Reviewing AI output carefully is essential because models can duplicate utilities, drift from conventions, and create hidden maintenance debt. Multiple agents can be useful, but mainly when tasks are low-risk, repetitive, or well-scoped; important production work still benefits from a hands-on, highly supervised approach. Infinite scroll works best for endless feeds, while pagination is better for large catalogs, linkable pages, and content that users need to navigate directly. Cursor-based pagination is preferable when data changes frequently because offset pagination can shift items between pages and produce inconsistent results. New developers should use AI to learn faster, break down documentation, debug, and research—rather than to avoid understanding programming fundamentals. Formal education gives foundational thinking skills, but most practical web knowledge comes from mentorship, on-the-job experience, and keeping up with the ecosystem. A real XSS attack can steal accessible client-side data if user input is inserted as HTML without sanitization; modern frameworks reduce this risk but do not eliminate it. To move from an outdated stack into a larger-company role, developers should demonstrate modern tooling and current expectations through side projects and updated skills. The best side projects are often ones that solve personal or family problems, making them easier to sustain and more valuable. Local/smaller LLMs are often better for narrow tasks than for open-ended chat, especially in browser-based or Transformers.js setups. Podcast guest/host decisions should be organic and based on fit; content that feels too different may belong on a separate channel rather than the main feed.

Data Points: Episode format: Potluck Q&A - The episode is framed as listener questions answered by the hosts. Approximate episode scale: Nearly 1,000 episodes - Used as an example of why the main Syntax episode listing should not be infinite scroll. Listeners mentioned in pagination example: Three cards per row - Used to justify page sizing and row-based multiples in the Syntax UI. Time to get to the bottom of a shopping list: About 10 minutes - The hosts describe browsing Restoration Hardware with infinite scroll. Time spent trying to recover scroll position: About 3 minutes - They explain how accidental clicks make infinite scroll frustrating without URL state. Python familiarity of learner: Basic level of Python - The listener asking about AI-assisted learning mentions this background. Experience since graduation: About half a year out of school - The listener running an LLC and client work says they are roughly six months post-graduation. Experience in role before seeking larger company: About 5 years - The PHP/jQuery front-end developer has worked at a small software house for five years. Podcast team/production reference: 50% of time - One host says a large portion of his job is exploring and learning through side projects. Devotional/hobby device update interval: Every 15 minutes - The e-ink terminal device cycles through different screens on this interval. Specific UI example: 20% volume - A simple AI-assisted fix described reducing audio volume to 20%.

Pivotal Quotes: "some of y'all have AI doing everything for you and are still lazy" — Scott: Used to argue that AI only helps if developers do the necessary planning, setup, and review work. "You should be using it to help you save time, not to do things for you" — Wes: Core advice to newer developers on how to integrate AI into learning and coding. "school teaches you the theory, how to think, how to research, how to tackle things" — Wes: Explaining why universities may not teach every modern tool, but still matter for foundational reasoning.

Implications: Listeners should treat AI as a multiplier, not a substitute for judgment. The episode reinforces that strong fundamentals, modern tooling, careful review, and intentional product choices still matter more than hype.