Delegation

The study explicitly models consumers delegating hotel search and purchase decisions to LLM agents acting on their behalf across 5,000 simulated sessions.

The POLICY condition is designed to let users pre-delegate decisions via standing rules, but the study finds users resist full delegation, preferring to retain case-by-case judgment.

ChatGPT Work routes knowledge worker intent through Codex as a background execution harness, abstracting agentic coding work from users who don't think of themselves as writing code.

Users delegate open-ended online tasks to an AI agent that autonomously navigates websites using Playwright.

Torvalds delegated grunt-work debugging tasks to the AI while retaining strategic direction and the decision to persist.

The human delegates distinct functional responsibilities to named agent roles rather than issuing ad-hoc prompts, mirroring executive delegation to functional leaders.

Users delegate routine coding tasks, git workflows, and codebase navigation to Claude via natural language terminal commands.

The article argues orchestrators should follow explicit ground rules for when to delegate to subagents, treating delegation as a mechanism to protect working memory rather than purely for parallelism.

Discusses how vibecoding at scale shifts intent-expression from code-writing to high-level delegation, raising questions about where human judgment must remain.

The developer delegates work to a fleet of agents rather than a single assistant, requiring new models of intent distribution.