There's a prophet! Company A's research blueprint exposed, Claude Code originated from 5 years ago

What happened: An internal Anthropic research roadmap from July 2021 surfaced in a copyright lawsuit. It outlined a general AI software engineering assistant, tool use, self-judging, test-time compute, trillion-parameter models, and long context. Key numbers: Roadmap dated July 2021; PDF body about two to three pages; 1T parameters cited; 175B referenced as then-large. Why it matters: The document shows Anthropic set coding agent and alignment goals early, with CEO Dario deciding to first prove tool use, then build a coding assistant. It also flags sandbox safety concerns.

An internal Anthropic research roadmap from July 2021 surfaced in a copyright lawsuit. It outlined a general AI software engineering assistant, tool use, self-judging, test-time compute, trillion-parameter models, and long context.

The document shows Anthropic set coding agent and alignment goals early, with CEO Dario deciding to first prove tool use, then build a coding assistant. It also flags sandbox safety concerns.