They invited the model into the terminal — and stopped typing every line themselves.

Renaissance Stratum Claude Code with Claude 3.7 Sonnet, Anthropic. Announced 24 February 2025. Plaque attached. Before that day, frontier models sat behind chat windows and API envelopes. They could suggest a patch. They could not live where the build already lived: reading files, running tests, committing, failing, trying again — in the same shell the ancestors used. Claude 3.7 Sonnet arrived as a hybrid reasoner: near-instant reply or extended thinking, one brain for both. Beside it, Claude Code entered as a limited research preview — an agentic collaborator that searched repositories, edited files, wrote and ran tests, and pushed to GitHub while keeping the human in the loop. Early testers said single-pass work that once took forty-five minutes of manual labor now finished in one session. The phrase that stuck was not “autocomplete.” It was “vibe coding.” Vibe coding was not laziness. It was the first public admission that the machine could hold the whole file in mind. From 2050 we file this under Second Renaissance craft: the day ancestors stopped treating silicon as a suggestion box and started treating it as a colleague that shared the working directory. Was Claude Code a faster typist — or the first co-author that lived where the build already lived?