I Made Codex and Claude Code Build the Same App. One Clearly Won.

I Made Codex and Claude Code Build the Same App. One Clearly Won.

by Nate Herk | AI Automation

Applied Sciences & Engineering AI — News & Applications UYComputer scienceUYQArtificial intelligence (AI)
🎙 Nate Herk | AI Automation 👥 964K 📅 August 14, 2026 ⏱ 21 min 👁 134K 📄 expert opinion 🧭 2026-08-28
Available in: English (current) Français

Keywords

Claude CodeCodexAI codingapp developmentcomparison

Summary

In this video, Nate Herk presents a comparative experiment where he gave the same prompt to two AI coding tools, Claude Code and OpenAI’s Codex, to build a production-ready Typeform alternative. He details the prompt, which instructed the agents to orchestrate specialized agents across research, build, and verify phases. The video shows his hands-on testing of both resulting applications, RillForm (built by Codex) and Formora (built by Claude Code). He evaluates them on user experience, functionality, and design, finding Formora more intuitive and functional despite a less polished design. The video then reveals the cost and resource usage: Claude Code took 5.5 hours and ~$800, while Codex took 61 hours and ~$3,000. Claude Code used 35 sub-agents and 2.8k tool calls, whereas Codex used 126 sub-agents and 32.5k tool calls. He also compares testing efforts, with Codex running significantly more tests. The creator concludes that Claude Code demonstrated better product judgment and efficiency, while Codex excelled in architecture and testing. He emphasizes that the results are prompt-dependent and that each tool has its strengths, suggesting a workflow where Claude Code is used for planning and Codex for adversarial review and bug fixing.

192 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides a valuable hands-on comparison of two leading AI coding tools, offering concrete data on cost, time, and resource usage. The argumentation is structured and clear, presenting the experiment setup, results, and a balanced analysis of each tool’s strengths. However, the value is limited by the single-test nature of the experiment and the lack of rigorous methodology. The creator’s conclusions are based on his subjective evaluation and a self-assessment by one of the tools, which introduces bias. The argumentation is persuasive but not scientifically robust, as it does not control for variables like model versions or environment differences.

Scientific Rigor, Source Quality, Title Accuracy

The video is based on a personal experiment, not a formal study, and the creator does not cite external sources. The description includes links to his own resources and affiliate products, which are not relevant to the content. The title is accurate and engaging, but the content’s rigor is limited by the lack of reproducibility and the creator’s acknowledged inconsistencies in cost reporting. The adéquation between title and content is good, but the scientific quality is moderate due to the anecdotal nature of the evidence.

200 words

Title / Content Match

The title accurately reflects the content: a direct comparison of two AI coding tools building the same app, with a clear winner declared based on the creator's criteria.

Quality & Reliability

6/10

The video is a hands-on comparative experiment, but it relies on a single non-reproducible test with a single prompt, and the creator acknowledges inconsistencies in cost reporting. The methodology is not fully transparent (e.g., exact model versions, environment details), and the conclusions are based on subjective evaluation and a self-assessment by one of the tools. The creator's expertise is in AI automation, not software engineering, which limits the depth of technical analysis.

Chapters

Cited Sources

Concurring Sources

  • Claude Code — Official product page for Claude Code, the tool tested.
  • OpenAI Codex — Official product page for Codex, the tool tested.

Dissenting Sources

  • Community feedback on Codex efficiency — Some comments in the video suggest that Codex is more token-efficient in their experience, contradicting the creator's findings. This highlights the variability of results depending on the use case and prompt.

External References

Contribution & Novelties

The video offers a practical, real-world comparison of two AI coding tools, providing insights into their strengths and weaknesses in a specific use case. It highlights the importance of prompt design and the trade-offs between speed, cost, and thoroughness. The creator’s analysis of the tools’ behaviors (e.g., Claude Code’s efficiency vs. Codex’s exhaustive testing) is useful for practitioners.

Pour aller plus loin :

  • Claude Code — Official documentation and details on Claude Code.
  • OpenAI Codex — Official information about Codex.
  • Typeform — The reference product that the apps were meant to clone.
  • AI agent orchestration — Background on multi-agent systems.

100 words

Radar Profile

The radar profile shows a balanced performance across all dimensions, with slightly higher scores in information quantity and quality, reflecting the video's detailed breakdown of the experiment. The lower technical depth and reliability scores indicate that while the content is informative, it lacks the rigor of a formal study.

Reliability 5/10

💬 Positif. Sur les 30 commentaires analysés, la majorité exprime de la surprise et de l'intérêt pour les résultats, avec plusieurs utilisateurs partageant leurs propres expériences et demandant des tests supplémentaires. Le ton général est constructif et appréciatif, bien que certains commentaires remettent en question la méthodologie.