How we built Linear Agent

Most software is designed to behave consistently. Good engineering has typically meant shrinking the space of possible outcomes until the same action reliably produces the same result. AI inverts that logic, at least somewhat. Much of Linear Agent’s value lies in doing work we didn’t anticipate in ways we never explicitly defined; script its behavior too tightly, and we’d dilute the flexibility th

[misshiki]

Linear Agentの設計思想が面白い。AIの挙動を細かく指示するのではなく、プロンプト・ツール・skills・承認ロジックで「自由に動ける境界」を設計。能力の広さより予測可能性を優先する場面も。