Key Takeaways
- Manus AI launched March 6, 2025, developed by Butterfly Effect Pte Ltd, Singapore
- Meta acquired Manus in December 2025 for a reported $2 to $3 billion
- The agent builds full-stack web apps, mobile apps, and data reports without requiring any coding knowledge
- Non-technical users including an 86-year-old linguist and a Singapore florist independently shipped working AI products using Manus
One year ago, the question was whether an AI could actually do things rather than just describe them. Manus answered that question in one hour on launch day by replicating a product that had taken its own team over a month to build. This is what that first year produced, who it changed, and what is still unfinished.
From Experiment to Acquisition in 12 Months
Manus launched on March 6, 2025, developed by Butterfly Effect Pte Ltd, a Singapore-based company with roots in the Monica.im ecosystem. The reception was immediate enough that VentureBeat described the launch as “a turning point in the development of artificial intelligence,” specifically citing its ability to complete complex tasks without continuous human input.
By December 2025, Meta had acquired Manus for a reported $2 to $3 billion, with exact terms not publicly disclosed. Meta stated it would integrate Manus technology into its products, including Meta AI.
What Manus Actually Does Differently
Most AI tools wait for your next prompt. Manus receives a goal and executes the full chain independently. It breaks complex tasks into subtasks, browses the web, writes and deploys code, processes files, and delivers a finished output.
The architecture behind this is a multi-agent system. Separate sub-agents handle planning, web browsing, coding, and data processing as distinct workstreams. This design allows Manus to adapt mid-task when initial assumptions prove wrong, which a single LLM cannot do reliably.
A feature called “Manus’s Computer” lets users watch the agent work in real time, seeing exactly which sites it visits, which forms it fills, and how it reasons through each decision. That transparency separates it from black-box automation tools that produce outputs without explanation.
Real People, Real Outcomes
The official year-one letter from the Manus team focuses not on benchmarks but on specific users whose constraints changed. Three cases define the range:
- A mother handed Manus a full day’s workload on launch day. One hour later the work was done, and she spent the weekend with her children
- Arkady, 86, a linguist with no coding background, built a fully working AI web app that generates personalized language lessons using direct-thinking methodology
- Noelle, a florist in Singapore, deployed an AI bouquet designer that shows realistic previews and handles checkout, reducing customer drop-off and freeing her to focus on arranging flowers
These are not power users. They are people who previously had no path from idea to working product, and that gap is what Manus targeted from day one.
Core Capabilities in 2026
The official year-one letter confirms the following capabilities Manus developed over its first year:
- Autonomous web research that goes beyond surface-level results and cites sources for each finding
- Full-stack web app development including databases and AI-native features
- Mobile app development, added during year one
- Data visualization from uploaded file inputs
- File creation across documents, spreadsheets, and reports
- Asynchronous execution, meaning users assign a task and check back when it is complete
The platform is accessible via the web app at manus.im and through messaging platforms including Telegram, WhatsApp, LINE, and Slack.
How Manus Compares to ChatGPT and Claude
The three tools serve fundamentally different needs, and choosing between them depends on what you are trying to accomplish.
| Dimension | Manus AI | ChatGPT | Claude |
|---|---|---|---|
| Autonomy | Full: assign and walk away | Requires prompt at each step | Requires prompt at each step |
| Best Use Case | Workflow automation, research, app building | Content creation, brainstorming | Long-doc analysis, deep reasoning |
| Output Type | Files, deployed apps, spreadsheets | Conversational text | Analytical text |
| Technical Requirement | Moderate upfront, then runs autonomously | Lowest barrier | Medium |
The most effective approach uses all three together. Claude analyzes a problem, ChatGPT generates creative directions, and Manus executes the multi-step work autonomously.
Limitations to Consider
Manus is not a production-ready automation system for every context. Users have reported stability issues when processing large volumes of text, and some complex tasks can take close to an hour to complete. That makes it unsuitable as a real-time tool for time-sensitive decisions.
For recurring, business-critical processes that require strict compliance and traceability, structured workflow tools remain more reliable than Manus at this stage. Manus performs strongest on exploratory, one-off, and research-intensive tasks.
What Comes Next
The Manus team has stated two explicit goals for year two. First, reach people who are still locked out of building because of technical barriers, specifically business teams, creators, and everyday users who have never written code. Second, move toward 24/7 always-on operation where an agent works continuously on behalf of a user without requiring check-ins.
The team acknowledges the infrastructure challenge behind that second goal directly, citing the need to keep costs viable at scale as the primary obstacle. That is the work still ahead.
If you are evaluating autonomous AI agents for content research, product prototyping, or workflow automation, Manus at year one has already proven the concept is real. The question is whether your specific use case fits within its current reliability window.
Frequently Asked Questions (FAQs)
When did Manus AI launch and who built it?
Manus AI officially launched on March 6, 2025. It was developed by Butterfly Effect Pte Ltd, a Singapore-based company also known for Monica.im. In December 2025, Meta acquired Manus for a reported $2 to $3 billion, with Meta planning to integrate the technology into its AI products.
What can Manus AI do that ChatGPT cannot?
Manus operates fully autonomously after receiving a single goal. It browses the web, writes and deploys code, processes data, and delivers finished files without requiring follow-up prompts. ChatGPT and Claude both require human guidance at each step of a workflow, making Manus distinctly suited for multi-step execution tasks.
Can non-technical users build apps with Manus AI?
Yes. Manus was designed specifically to remove technical barriers from building. An 86-year-old linguist with no coding background built a fully working AI language learning web app using Manus. A florist in Singapore deployed an AI product designer with checkout functionality using the same approach, both documented in Manus’s official year-one letter.
How does Manus AI handle research tasks?
Manus conducts deep web research by going beyond surface-level results, citing sources for each finding, and compiling structured reports. The agent handles the full research pipeline autonomously from query to formatted deliverable, though complex research assignments can take close to an hour to complete.
What are the main limitations of Manus AI?
Manus has reported stability issues when processing large text volumes. Some tasks run for close to an hour, limiting its use for time-sensitive work. It is not currently a replacement for structured automation systems in business-critical production environments where compliance and traceability are required.
What did Meta plan to do with Manus after acquiring it?
Meta confirmed it would integrate Manus AI technology into its products, including Meta AI. The acquisition was completed in December 2025. Exact financial terms were not publicly disclosed, though reports placed the deal value between $2 and $3 billion.
What is the vision Manus has stated for year two?
The Manus team has publicly stated two goals: expanding access to non-technical users such as business teams and everyday creators, and building toward 24/7 always-on agent operation. The team acknowledged keeping infrastructure costs viable at that scale as the primary challenge to solve.
