Manus and autonomous agents: impressive demos, practical limits
An AI that researches, writes and builds while you sleep sounds like the perfect employee. The reality is more useful — and more modest — than the viral videos suggest.
By the Xuru Technologies team · 12 March 2025
Key takeaways
- Manus, launched in early March 2025 by Chinese start-up Butterfly Effect, is a general-purpose agent that plans and carries out multi-step tasks.
- Early testers praised its ambition but reported failures, loops and slow performance on harder tasks.
- Agents work best on well-defined, checkable tasks with clear inputs.
- Businesses should automate narrow workflows first, with a person approving results.
Every few months a new AI demo convinces the internet that work as we know it is over. Then reality arrives. Both reactions miss the useful middle ground.
What happened
Manus launched on an invite-only basis and quickly drew attention with demonstrations of tasks such as researching topics, analysing data and building simple websites. MIT Technology Review tested it and found it capable and transparent about its steps but prone to crashes and slowdowns under load. Background notes it builds on existing large models rather than its own foundation model.
What it means for businesses
- Agents are good interns, not managers. They can gather information and draft work; they still need direction and review.
- Narrow beats general. "Summarise every new enquiry and suggest a reply" works far better than "run my marketing".
- Data access is the risk. An agent that can log into your accounts can also make mistakes in them.
- The winners are well-documented processes. If a task is written down step by step, it can probably be automated.
Good first automations
| Workflow | Human check |
|---|---|
| Enquiry triage and draft replies | Approve before sending |
| Competitor and market research summaries | Verify key facts |
| Turning meeting notes into tasks | Quick review |
| Review monitoring and response drafts | Edit tone before posting |
See seven AI workflows that save hours.
Frequently asked questions
What is the difference between a chatbot and an agent?
A chatbot answers. An agent takes actions — searching, clicking, writing files or calling other tools — to reach a goal.
Should I give an AI agent access to my business accounts?
Only with limited permissions, a clear task and a person approving anything that sends, pays or publishes.
Will agents replace staff?
For now they mostly remove repetitive steps, freeing people for customer-facing and judgement work.
Agents that help, with you in charge.
We design narrow, reliable AI workflows for enquiries, follow-ups and admin, with approvals where they matter.
Websites, payments, digital marketing, AI automation, consultancy and practical technology training — for businesses and aspiring professionals worldwide.
Saturday – Sunday: closed
Leave a Reply