Manus is the general AI agent that bridges minds and actions: it doesn't just think, it delivers results.
3336089.1K11M2.6K
Tags
B2CAI agentProduct launchGlobalSeries C+5M+Demo
TLVC Rating
3.60
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
No ratings yet
to rate this launch.
Manus 2.0 is a rebuild of the general-purpose AI agent from Singapore-based startup Butterfly Effect, aimed at founders, operators, and knowledge workers who want an agent that executes real work rather than just generating text. The release introduces a new in-house agent harness called Cascade, a Cloud Computer for projects that need to keep running after a laptop closes, Automations, a professional workspace called Manus Studio, and Cue, a new app for personal agents. In one tested configuration, Cascade used 23.2% fewer tokens, finished tasks in 28.2% less time, and cost 32% less to run than the previous system.
The practical surface area is broader than the first release. A new Automations feature lets a task start when something happens in a connected service, such as a new email, a Slack message, a calendar event, or a Notion update. Manus Studio turns the desktop app into a workspace with video editing, game development, and remote computer use, while Cue is a standalone product where personal agents get their own email, phone number, wallet, and computer so they can coordinate across tasks on a user's behalf.
Manus was developed by Butterfly Effect, a Chinese startup led by Ji Yichao, who previously created the Mammoth mobile browser and founded Peak Labs , and the company is now based in Singapore . This launch matters because the first version of Manus drew heavy attention earlier in the year for autonomous task completion, and 2.0 is the team's attempt to answer the obvious follow-up question of whether a general agent can actually hold state, trigger itself, and ship finished artifacts across long-running projects.
Comments (12)
Sign in to join the discussion.
P
Priya Rangarajan10d ago
Before anyone wires this access to a browser session, where does execution state live and can I pin it to EU regions? Autonomous agents without data residency controls are a procurement nightmare.
T
Tomas Lindqvist10d ago
The 2.0 reveal video has nice pacing but that synth drop at 0:14 is doing a LOT of heavy lifting to hide the fact we never see a full task from prompt to delivery.
L
Lena Quist10d ago
Three replies deep and still no mention of eval methodology. Is there a paper, a tech report, or are we vibes-benchmarking on cherry-picked demos again?
B
Benoît K.10d ago
Market for 'general agents' is quietly collapsing into whoever owns the browser runtime. Being general is a liability when every vertical agent eats your lunch on reliability.
I
Ifeoma Dara10d ago
Polite ping: happy to run an embargoed walkthrough piece if you've got internal benchmarks you haven't shown yet. Even a redacted eval table would do.
Y
yuki10d ago
The kerning on 'Manus 2.0' in the hero shot is cursed, those zeros are kissing the dot. Also the gradient from teal to lavender fights the serif, pick a lane.
R
Ravi M.10d ago
Noob question: if it browses on my behalf and a site asks for 2FA, does it ping me or just quietly give up? Asking because I set it loose on booking a flight and it got stuck forever.
M
Mallory Osei10d ago
Been watching this team since the first demo made the rounds, feels like the Linear of agent runtimes if they nail the handoff UX. Already drafting intros to two of my ops-heavy portcos.
D
deuce10d ago
A general agent executing tasks across services is screaming for an on-chain reputation + payment rail. You're going to need machine-native settlement eventually, might as well plan for it now.
H
Hwan-soo Park10d ago
Shipped something adjacent inside a big co in 2019 and the hard part was never the planner, it was the long-tail recovery when a DOM shifted overnight. Curious how 2.0 handles silent website changes mid-task.
D
Dmitri Varga10d ago
Building my own agent framework and honestly 3 weeks out from a beta that does roughly this. Respect the ship but the moat here has to be the sandbox, not the model.
C
Carla Ibáñez10d ago
Fifth 'general agent' launch I've clocked this quarter and the only one whose teaser didn't feature a wizard metaphor, so points for restraint. Pace of releases in this category is getting unhinged.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Manus AI
Manus is the general AI agent that bridges minds and actions: it doesn't just think, it delivers results.
Manus is the general AI agent that bridges minds and actions: it doesn't just think, it delivers results.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Manus 2.0 is a rebuild of the general-purpose AI agent from Singapore-based startup Butterfly Effect, aimed at founders, operators, and knowledge workers who want an agent that executes real work rather than just generating text. The release introduces a new in-house agent harness called Cascade, a Cloud Computer for projects that need to keep running after a laptop closes, Automations, a professional workspace called Manus Studio, and Cue, a new app for personal agents. In one tested configuration, Cascade used 23.2% fewer tokens, finished tasks in 28.2% less time, and cost 32% less to run than the previous system. The practical surface area is broader than the first release. A new Automations feature lets a task start when something happens in a connected service, such as a new email, a Slack message, a calendar event, or a Notion update. Manus Studio turns the desktop app into a workspace with video editing, game development, and remote computer use, while Cue is a standalone product where personal agents get their own email, phone number, wallet, and computer so they can coordinate across tasks on a user's behalf. Manus was developed by Butterfly Effect, a Chinese startup led by Ji Yichao, who previously created the Mammoth mobile browser and founded Peak Labs , and the company is now based in Singapore . This launch matters because the first version of Manus drew heavy attention earlier in the year for autonomous task completion, and 2.0 is the team's attempt to answer the obvious follow-up question of whether a general agent can actually hold state, trigger itself, and ship finished artifacts across long-running projects.
Comments (12)
Before anyone wires this access to a browser session, where does execution state live and can I pin it to EU regions? Autonomous agents without data residency controls are a procurement nightmare.
The 2.0 reveal video has nice pacing but that synth drop at 0:14 is doing a LOT of heavy lifting to hide the fact we never see a full task from prompt to delivery.
Three replies deep and still no mention of eval methodology. Is there a paper, a tech report, or are we vibes-benchmarking on cherry-picked demos again?
Market for 'general agents' is quietly collapsing into whoever owns the browser runtime. Being general is a liability when every vertical agent eats your lunch on reliability.
Polite ping: happy to run an embargoed walkthrough piece if you've got internal benchmarks you haven't shown yet. Even a redacted eval table would do.
The kerning on 'Manus 2.0' in the hero shot is cursed, those zeros are kissing the dot. Also the gradient from teal to lavender fights the serif, pick a lane.
Noob question: if it browses on my behalf and a site asks for 2FA, does it ping me or just quietly give up? Asking because I set it loose on booking a flight and it got stuck forever.
Been watching this team since the first demo made the rounds, feels like the Linear of agent runtimes if they nail the handoff UX. Already drafting intros to two of my ops-heavy portcos.
A general agent executing tasks across services is screaming for an on-chain reputation + payment rail. You're going to need machine-native settlement eventually, might as well plan for it now.
Shipped something adjacent inside a big co in 2019 and the hard part was never the planner, it was the long-tail recovery when a DOM shifted overnight. Curious how 2.0 handles silent website changes mid-task.
Building my own agent framework and honestly 3 weeks out from a beta that does roughly this. Respect the ship but the moat here has to be the sandbox, not the model.
Fifth 'general agent' launch I've clocked this quarter and the only one whose teaser didn't feature a wizard metaphor, so points for restraint. Pace of releases in this category is getting unhinged.