Summary
Nate Herk runs a structured design comparison between Claude Code and OpenAI’s Codex, giving both tools identical prompts and brand assets to build eight distinct websites, then evaluating the outputs side by side on design quality, UX clarity, and build efficiency.
Across the eight builds, Herk tracks concrete metrics for each tool: number of sub-agents spawned, total build time, output tokens consumed, and dollar cost. In the opening Bowl & Bloom example, Claude Code used four sub-agents, took two hours, generated nearly 500,000 output tokens, and cost $53 — versus Codex’s single agent, 50-minute build, 100,000 tokens, and roughly $20. Similar cost and time gaps appear across subsequent builds. On pure design aesthetics, Herk generally favors Codex’s cleaner layouts and interactive scroll animations, while noting that Claude Code sometimes produces more feature-complete but visually noisier results.
The video is methodologically transparent about its limitations — screen size affects rendering, both tools miss edge cases in interactive elements, and subjective design taste plays a role. Herk ties in his own scroll-animation skill (available as a free download) that both tools can leverage. For teams choosing between Claude Code and Codex for front-end prototyping, the video provides one of the more data-grounded comparisons available, with specific dollar figures and token counts to inform purchasing decisions.
📺 Source: Nate Herk | AI Automation · Published August 26, 2026
🏷️ Format: Comparison
