Frameworkless frontend with AI agents, quality gates, and repeatable delivery, from web to Electron

How do you bring AI into the development cycle without losing technical control, and without turning your codebase into a "slot machine"? In this talk I walk through a concrete workflow built on ai-coding-frontend-template: we go from specifications to implementation with verifiable steps, minimal diffs, and genuinely repeatable outcomes.

We build UI with Vanilla JavaScript (ES Modules), Custom Elements v1 in the light DOM, and zero runtime dependencies, cutting complexity and lock-in without sacrificing performance or maintainability. Then we widen the scope: the same approach applied to an Electron desktop app, covering packaging, asset management, a secure bridge between UI and the system (IPC), and a clean path for file system access and native integrations, without throwing away the frontend architecture.

We will dive into multi-agent operating rules, micro-step planning, and automated quality gates (lint + E2E) that run both locally and in CI. The focus is engineering-first: just method, traceability, and reliability.

A practical session for devs that want to turn specs into real features with a measurable, governable, and scalable process, on web and desktop.

Requirements:
- Modern JavaScript: ES Modules, async/await, DOM APIs, events, fetch.
- Component fundamentals: Custom Elements, light DOM, attributes, lifecycle, basic accessibility.
- Comfortable with CLI and Node: install deps, run scripts, read logs.
- Practical interest in AI prompting and vibe coding, with quality-first discipline.