Manus

An autonomous AI agent "based on Claude" (Anthropic), used by Ethan Mollick as a reference point for what agentic systems can do.

In this source

Mollick "vibe-benchmarked" Manus (organizational benchmarking) by handing it a fictional startup's description and projected financials (from an unpublished Wharton business simulation) plus a checklist. In a couple of prompts it produced a website, a PowerPoint pitch deck, a business-model analysis, and a test of the financial assumptions. Mollick judged the 45-page analysis "very solid… far less mistakes, and far more thorough, than what I would expect from talented students," while exposing weak spots — image generation and "a tendency to extrapolate answers without asking." It became his baseline for comparing each new agentic system.

Connections

• Agent Autonomy Spectrum — a concrete autonomous agent; the "extrapolate without asking" failure mode is an argument for the semi-synchronous middle.
• Anthropic / Claude — built on Anthropic's models.
• Leadership, Lab, and Crowd — featured in Mollick's "build your own benchmarks" practice.

Sources

• Making AI Work — Leadership, Lab, and Crowd.