MIT Technology Review · Mar 11, 2025
MIT Technology Review tests Manus and finds promise alongside serious limits
Markdown · JSON · Original source
MIT Technology Review’s review, published March 11, 2025, describes a test of Manus, an developed by Wuhan-based startup Butterfly Effect. The reviewer found it intuitive and adaptable, but the results varied by task. Manus performed better than ChatGPT DeepResearch on two of the three assignments, though it took significantly longer. The reviewer could guide Manus and observe or intervene in its work; the tests also exposed trouble with paywalls and captcha blocks, uneven research results, crashes, and freezes. The reviewer’s conclusion is that Manus is best suited to open-internet research with limited scope—work comparable to what a skilled human intern could do during a day. The reviewer also values its collaborative process and argues that Chinese AI companies are shaping the adoption of autonomous AI agents, not merely following Western counterparts. This assessment comes from one reviewer’s three assignments, and the article gives no measurement method for the comparison with ChatGPT DeepResearch. It does not establish how Manus performs across other tasks or users.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
MIT Technology Review · Article
Source details
Original source date: Mar 11, 2025 (2025-03-11).
Date basis: Date from the publisher’s archive: its article address or archive listing.
Added to Escape Meridian: 2026-11-14T22:48:23.273Z. First observed: 2026-10-25T10:32:09.424Z. Retrieved: 2026-10-25T10:33:45.139Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 26, 2026; originally published March 11, 2025. The supplied text does not identify the reviewer by name; attribution here is to MIT Technology Review’s review. The comparison with ChatGPT DeepResearch is based on three assignments by one reviewer; the article does not report a measurement method. The retrieved page includes related material after the review; this reading uses the review text and not that additional material. Text as retrieved on 2026-10-25; originally published March 11, 2025.
Open question
How would Manus perform across a broader range of tasks and users, given that this review tested it on only three assignments?
Source-stated
The reviewer reports that Manus performed better than ChatGPT DeepResearch on two tasks, but took significantly longer.
On two of the three tasks, it provided better results than ChatGPT DeepResearch, though it took significantly longer to complete them.
Source-stated
The reviewer says Manus is best suited to open-internet research with limited scope, comparing its work to that of a skilled human intern.
it’s best to stick to the sorts of things a skilled human intern could do during a day of work.
Source-stated
The reviewer found Manus adaptable to guidance, but reports obstacles with paywalls, uneven research results, crashes, and freezes.
it occasionally lacks understanding of what it’s being asked to do, makes incorrect assumptions, or cuts corners to expedite tasks
Source-stated
The reviewer argues that Chinese AI companies are shaping the adoption of autonomous AI agents rather than only following Western counterparts.
they are actively shaping the adoption of autonomous AI agents in their own way.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
AI agent
An AI system that chooses and carries out actions toward a goal, often using software tools.
Editorial provenance
Version 1. Publication ID: f396fa50-e702-177e-acca-e5154ed374eb.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-14T22:44:02.061Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: d359ebc264860af8705e109c6adde53c7662cf398dc685c069a7a60e470933db. Source text hash: eca5f08d05fbc424271e02a7a11c4e13a304fe170a631586f4d2584f3e28e69b.
Revision and source-check history
Version 1 · 2026-11-14T22:48:23.273Z
MIT Technology Review tests Manus and finds promise alongside serious limits
MIT Technology Review’s review, published March 11, 2025, describes a test of Manus, an developed by Wuhan-based startup Butterfly Effect. The reviewer found it intuitive and adaptable, but the results varied by task. Manus performed better than ChatGPT DeepResearch on two of the three assignments, though it took significantly longer. The reviewer could guide Manus and observe or intervene in its work; the tests also exposed trouble with paywalls and captcha blocks, uneven research results, crashes, and freezes. The reviewer’s conclusion is that Manus is best suited to open-internet research with limited scope—work comparable to what a skilled human intern could do during a day. The reviewer also values its collaborative process and argues that Chinese AI companies are shaping the adoption of autonomous AI agents, not merely following Western counterparts. This assessment comes from one reviewer’s three assignments, and the article gives no measurement method for the comparison with ChatGPT DeepResearch. It does not establish how Manus performs across other tasks or users.
Original source
MIT Technology Review · Article
Source details
Original source date: Mar 11, 2025 (2025-03-11).
Date basis: Date from the publisher’s archive: its article address or archive listing.
Added to Escape Meridian: 2026-11-14T22:48:23.273Z. First observed: 2026-10-25T10:32:09.424Z. Retrieved: 2026-10-25T10:33:45.139Z.
Limits
Text as retrieved on October 26, 2026; originally published March 11, 2025. The supplied text does not identify the reviewer by name; attribution here is to MIT Technology Review’s review. The comparison with ChatGPT DeepResearch is based on three assignments by one reviewer; the article does not report a measurement method. The retrieved page includes related material after the review; this reading uses the review text and not that additional material. Text as retrieved on 2026-10-25; originally published March 11, 2025.
Open question
How would Manus perform across a broader range of tasks and users, given that this review tested it on only three assignments?
Source-stated
The reviewer reports that Manus performed better than ChatGPT DeepResearch on two tasks, but took significantly longer.
On two of the three tasks, it provided better results than ChatGPT DeepResearch, though it took significantly longer to complete them.
Source-stated
The reviewer says Manus is best suited to open-internet research with limited scope, comparing its work to that of a skilled human intern.
it’s best to stick to the sorts of things a skilled human intern could do during a day of work.
Source-stated
The reviewer found Manus adaptable to guidance, but reports obstacles with paywalls, uneven research results, crashes, and freezes.
it occasionally lacks understanding of what it’s being asked to do, makes incorrect assumptions, or cuts corners to expedite tasks
Source-stated
The reviewer argues that Chinese AI companies are shaping the adoption of autonomous AI agents rather than only following Western counterparts.
they are actively shaping the adoption of autonomous AI agents in their own way.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
AI agent
An AI system that chooses and carries out actions toward a goal, often using software tools.
Editorial provenance
Version 1. Publication ID: f396fa50-e702-177e-acca-e5154ed374eb.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-14T22:44:02.061Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: d359ebc264860af8705e109c6adde53c7662cf398dc685c069a7a60e470933db. Source text hash: eca5f08d05fbc424271e02a7a11c4e13a304fe170a631586f4d2584f3e28e69b.
uncertain · 2026-11-14T22:44:02.061Z · anthropic/claude-opus-5.5
Publication: f396fa50-e702-177e-acca-e5154ed374eb
attribution — pass: The reading correctly attributes the review to MIT Technology Review. It notes that the reviewer is not named in the supplied text and attributes Manus's development to Butterfly Effect, as the source states. Assessments are framed as the reviewer's.
developed by the Wuhan-based startup Butterfly Effect
summary — pass: The reading clearly states the actor, the product, the test, the mixed results and the reviewer's conclusions. The term AI agent is defined in the page definitions, and ChatGPT DeepResearch is presented as a comparison tool. It preserves the reviewer's 'promising but not perfect' stance and the argument about Chinese AI companies. The question follows from the limited test scope.
Ultimately, it’s promising but not perfect.
date — uncertain: The stated publication date, March 11, 2025, matches originalPublished at day precision. There is no hindsight framing, and the later related-story sidebar is explicitly excluded. However, the limitations give conflicting retrieval dates: 'October 26, 2026' and '2026-10-25'. The evidence shows retrieval on 2026-10-25. The incorrect date is metadata, so no source passage can be quoted. Repair: delete the October 26 limitation line.
support — pass: Each claim is attributed to the reviewer and matches the review text: better results on two of three tasks but slower, best suited to intern-level limited-scope research, the obstacles, and the argument about Chinese AI companies. These passages are plainly part of the original review body, not the later-edited sidebar. The caveat about one reviewer and three tasks is supported.
Manus can suffer from frequent crashes and system instability, and it may struggle when asked to process large chunks of text.
media — pass: No media was selected, and no verified media candidates were supplied. Omitting media is appropriate.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
An AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceAn AI system that chooses and carries out actions toward a goal, often using software tools.
Definition reference