Ep002 800,000 Lines Later: Keeping Humans in the Loop with Agent Swarms
カートのアイテムが多すぎます
カートに追加できませんでした。
ウィッシュリストに追加できませんでした。
ほしい物リストの削除に失敗しました。
ポッドキャストのフォローに失敗しました
ポッドキャストのフォロー解除に失敗しました
Ep002 800,000 Lines Later: Keeping Humans in the Loop with Agent Swarms
-
ナレーター：
-
著者：
Bruce is joined by Nathan Foreman, a principal engineer at Roller. Bruce and Nathan have worked together since their Cookdown days, and Nathan was the one who first showed Bruce how to "write code" with AI around a barbecue in Nashville.
They trace Nathan's journey from early GitHub Copilot to multi-agent workflows using Spec Kit, Gastown and Claude agent teams. They also cover what happened when he let agents approve their own pull requests and came back to 800,000 lines of working but unmaintainable code. Along the way they discuss why AI over-engineers, and how architecture decision records keep agents on track. They get into the "governance dial" between reading every law and getting sued, and the unsolved problem of handling secrets with computer-use agents. They're also candid about burnout, the checkpoints teams lose when one person can do every role, and Nathan's hard-won advice: look before you build.