SHIPPING AT INFERENCE-SPEED

Summary from the TLDR Dev newsletter

AI-powered coding has accelerated this dev's software development speed, with GPT-5/5.2 Codex being a major breakthrough that allows them to ship code at "inference-speed.” This means they are limited mainly by AI processing time rather than human coding ability. GPT's Codex takes longer to start (sometimes 10-15 minutes reading code), but it produces more reliable results than Claude's Opus, which works faster but often requires fixes. They now rarely read the code they produce, instead focusing on high-level architecture decisions and letting AI handle most implementation details.