Manus tested on research tasks
Manus researched China tech reporters, NYC two-bedroom listings, and Innovators Under 35 candidates for an MIT Technology Review reporter, who steered it with feedback.
It produced a 30-journalist list, a tiered NYC apartment list in under half an hour, and only five solid nominees out of fifty requested.
Overview
How it was set up
Manus general agent with invite access. Uses Claude 3.5 Sonnet and fine-tuned Qwen models. The author watched the Manus's Computer window and took over for captchas.
Worth knowing
Frequent crashes and server overload, paywall and captcha blocks, and a three-hour nomination task that stalled. First answers were lazy until the author pushed back.
A prompt to start from
Compile a list of [number] notable [type of person] covering [topic]. For each, give their current outlet and notable work. Be consistent and thorough, and check for recent employer changes.
Written by Agentic Feats from the source. Not the author's exact prompt.
Sources 1
History
- Source published
- 2025-03-11
- Added to Agentic Feats
- 2026-10-13
- Pick
- No
- Markdown copy
- /c/manus-tested-on-research-tasks.md
More in research
36 featsResultIt created the text file as asked, but after clicking a result it took a screenshot too early and hallucinated the answer.
ResultOperator handled tour search and lesson plan but was slow; adding three grocery items took 15 minutes and an Uber price check took five.
ResultThe draft was usable but weaker than accepted NSF summaries; thinking mode helped, though it began hallucinating and misreading the sources.