Rebuilding My Personal Website with Codex and Claude: A Practical Review
Ben's Bites · rss · 2026-08-28
Ben Tossell shares his detailed process and experience rebuilding his personal website using AI coding assistants (Codex and Claude).
Design and Iteration:
- Initially attempted to mimic an Agent app interface, having Codex generate a prototype. He established a rapid iteration loop by taking screenshots, annotating feedback in tldraw, and sending it back to the model for adjustments.
- Found that Claude is better at inferring design intent, whereas GPT requires more specific feedback.
- Experienced "scope creep" during development by adding too many interactions, ultimately deciding to scrap it and return to a simpler canvas style.
Technical Implementation and Deployment:
- Content Management: Configured the site to pull content from individual Markdown files for easy maintenance.
- Deployment: Used here.now for rapid deployment. The agent directly configured Vercel DNS records to point the domain to the new site, eliminating manual setup.
- Animation Details: Guided the agent via text prompts to create a walking mascot animation, iteratively refining limb movements and interaction logic.
Key Takeaways:
- Real requirements often only become clear after seeing a working version, hence the practice of asking agents for multiple versions for comparison.
- The "design-annotate-fix" loop is an efficient workflow for AI-assisted development.
More from coding & agent
- Guide: Using Claude Design Inside Claude Code for rapid iteration — tomcrawshaw01 · 2026-08-28
- KateBench details: 86% automation rate reveals consistency issues — every · 2026-08-28
- The Universal Loop Structure for Agents and Workflows — DavidKPiano · 2026-08-28
- Take: APIs, CLIs, and MCP Are Best Understood as Cost Optimization Mechanisms — curious_vii · 2026-08-28
- Tencent Hunyuan Hy4 Fixes 17 Bugs for Just $3.13 in Real-World Test — PawelHuryn · 2026-08-28
- Dev simulates eclipses in custom light physics engine — measure_plan · 2026-08-28