Descriptions:
Ray Amjad shares his complete AI-assisted coding workflow for 2026, covering the full pipeline from idea to shipped feature. Spec creation starts with a screen recording of a comparable existing product, uploaded to Gemini 1.5 Pro in Google AI Studio to generate an initial PRD. Claude Code’s `ask_user_question` tool then drives a structured interview to refine requirements, while a separate ChatGPT session with deep thinking enabled searches for well-maintained open-source packages compatible with Next.js — surfacing libraries with smaller GitHub followings that Claude Code may not surface on its own.
The bulk of the video is a candid, experience-based comparison of Claude Code and Codex CLI. Amjad describes Claude Code as more conversational and interactive, better for short feedback cycles, quick corrections, and tasks where active engagement matters. Codex CLI, by contrast, spends 10–15 minutes reading broadly across the codebase before making changes, making it more accurate for bugs rooted in architectural decisions or cross-cutting concerns. His current usage split is roughly 80% Claude Code and 20% Codex CLI, with Codex CLI reserved for Sentry error investigations and other tasks that benefit from deep context gathering.
Practical tips include running multiple CLI tool sessions in parallel across different projects using separate terminal windows, validating each development phase manually before proceeding to the next, and staying actively engaged as an orchestrator rather than waiting passively for results. Amjad built his HyperWhisper speech-to-text application writing fewer than three lines of code himself — a concrete illustration of how far agentic coding tools have advanced for independent developers.
📺 Source: Ray Amjad · Published January 02, 2026
🏷️ Format: Workflow Case Study
