Summary
Theo of t3.gg argues that Opus 5.5 is evidence the industry still needs smarter models, reversing his earlier view that AI capability had advanced far enough. He recalls a video from two years ago predicting AI progress would plateau, then traces how coding assistance moved from autocomplete to agents that take a bug screenshot, fix it, test it, and get it merged.
The core of the video is a surprising source of his change of mind: hobbyists using AI to reverse engineer, decompile, and mash together classic games such as Halo and Call of Duty. Theo walks through successive model generations, including Opus 4.5, Fable 5, and Astra, and describes who benefited at each step. Developers gained most from early jumps, while technical people who don’t code were limited. With Opus 5.5, he says, a single well-priced model available on subscription tiers lets people who have never coded or opened Blender build and port things they don’t fully understand, some paying $200 a month for Claude plans to do it.
He also considers what this means for non-technical users, suggesting that the tech-savvy relatives and friends who fix everyone’s computers can use these tools to solve more of their problems and share mods and custom software. A mid-video sponsor segment covers Browserbase, a platform for agent web access with authentication, session persistence, and search and fetch APIs.
📺 Source: Theo – t3․gg · Published October 20, 2026
🏷️ Format: Opinion Editorial
