Model Landscape for Coding

The Future of Code Review: Stop Reviewing Line-by-Line, Start Governing AI Agents - Itamar Friedman

Topic 9 of 11

Published May 4, 2026 · Clip 55:26–1:03:16 (7:50)

Itamar explains that frontier models now excel at different properties, an era of specialization he expects to deepen — GPT-3.5/4 was briefly better at everything, but Sonnet 3.5/3.7 broke that with coding strength, and today Claude and GPT each win on different coding aspects, even subjectively. Companies are building specialized models (tabular data; Qodo fine-tunes OpenAI and Anthropic models for code review and governance). He predicts a possible architecture-breaking moment — transformers aren't the last word, AutoML/neural architecture search will return — and bets open-source models like Kimi and Qwen may innovate first because being number two forces different moves. Qodo uses a cocktail of Anthropic, OpenAI, and Google models depending on deployment and keeps older models when new ones don't beat them.