https://america66.jp/en/tags/ai-development/Recent content in AI Development on オブライエンの日誌Hugoen-USSat, 19 Sep 2026 00:00:00 +0900https://america66.jp/en/post/2026-09-19-ai-existential-risk-ugaya/Sat, 19 Sep 2026 00:00:00 +0900https://america66.jp/en/post/2026-09-19-ai-existential-risk-ugaya/Journalist Hiromichi Ugaya, drawing on his background in nuclear strategy, argues that AI is becoming an existential risk precisely because the people building it keep saying so themselves. A summary of his YouTube livestream.https://america66.jp/en/post/2026-09-12-boris-cherny-opus-5-claude-code-unhobbling/Sat, 12 Sep 2026 00:00:00 +0900https://america66.jp/en/post/2026-09-12-boris-cherny-opus-5-claude-code-unhobbling/A summary of Claude Code creator Boris Cherny's talk at Y Combinator Startup School 2026: why the team deletes most of the system prompt with every new model, the idea of 'unhobbling,' and how to run thousands of agents at once.https://america66.jp/en/post/2026-09-10-claude-fills-the-gaps/Thu, 10 Sep 2026 00:00:00 +0900https://america66.jp/en/post/2026-09-10-claude-fills-the-gaps/Thirty hours of work with Claude Code (Sonnet 5), fact-checking every line. The failure modes that kept showing up — and why having it write a rule doesn't fix them.https://america66.jp/en/post/2026-08-17-claude-code-verification-loops-multi-clauding/Mon, 17 Aug 2026 00:00:00 +0900https://america66.jp/en/post/2026-08-17-claude-code-verification-loops-multi-clauding/A summary of a Claude Code engineering team conference talk, covering three techniques — verification loops, multi-clauding, and background loops — for turning Claude from something you supervise into something you can trust with real work.