In July 2020, founder Sharif Shameem posted a video on Twitter showing his tool debuild.co, where he typed a plain English sentence into a text box and GPT-3 wrote a working React component with a constructor, state management, and two buttons that changed a number on the screen. “I just built a functioning React app by describing what I wanted to GPT-3. I’m still in awe,” Shameem wrote. At the time, GPT-3 had only been available to a small circle of developers for a few weeks. The demo spread quickly, becoming one of the moments people point to as the beginning of the current AI wave. Debuild never grew into a lasting company; Shameem later founded Lexica before joining OpenAI’s models team. Today, current agentic coding models are given open-ended tasks and work largely unsupervised, reading error logs and installing dependencies. Anthropic’s Claude Opus 4.6 had held the longest task-completion time horizon tracked by METR, with a fifty percent success rate on tasks that would take a human roughly fourteen and a half hours. In one demonstration, sixteen instances of Claude Opus 4.6 working in parallel wrote a C compiler in Rust from scratch, capable of compiling the Linux kernel. OpenAI has said its coding model, GPT-5.3-Codex, was the first version of Codex instrumental in building itself, with early checkpoints used internally to debug their own