More How I AI episodes

Claude Opus 4.6 vs. GPT-5.3 Codex: How I shipped 93,000 lines of code in 5 days thumbnail

Claude Opus 4.6 vs. GPT-5.3 Codex: How I shipped 93,000 lines of code in 5 days

Published 11 Feb 2026

Duration: 00:30:12

Recent advancements in AI coding models, including OpenAI's Codex and Anthropic's Opus 4.6, are discussed in the context of their effectiveness and limitations in software development.

Episode Description

I put the newest AI coding models from OpenAI and Anthropic head-to-head, testing them on real engineering work Im actually doing. I compare GPT-5.3 C...

Overview

The podcast explores recent advancements in AI coding models, specifically OpenAI's Codex and Anthropic's Opus 4.6 and Opus 4.6 Fast, and their impact on software development. It highlights Codex's proficiency in code review, architectural advice, and generating high-quality code, though it struggles with creative coding tasks. Opus 4.6 excels in creating functional and visually appealing components, though its initial outputs required refinement through iterative prompting and detailed design guidelines to achieve polished results. A practical example involved using these models to redesign a website, which uncovered challenges such as overfitting to user prompts and the importance of clear instructions in achieving desired outcomes.

The discussion also examines how Codex and Opus can be used together in development workflows, with Codex handling initial code generation and Opus refining and enhancing the output. The podcast touches on various practical considerations, including cost, token usage, and the importance of selecting the right model based on specific project needs and budget constraints. Overall, the conversation emphasizes the potential benefits of integrating multiple AI coding models to create more efficient and effective development processes.

Recent Episodes of How I AI

16 Mar 2026 From journalist to iOS developer: How LinkedIns editor builds with Claude Code | Daniel Roth

AI agents like Bob the Builder, Ray the Reviewer, and The VibeCoder streamline software development by combining coding, security, and user preferences, enabling non-experts to build apps with tools like Claude and Cursor while addressing enterprise security challenges, iterative testing, and human-AI collaboration to balance automation with quality and compliance.

More How I AI episodes