What Manus Is and What It Can Prepare for a Creator

A chatbot (ChatGPT, Claude or Gemini used for single tasks) works one turn at a time. You type a prompt, it writes an answer, and the next move is yours.

An AI agent takes a goal and works through several steps by itself: it searches, opens pages, reads your files, writes, and hands back finished work.

Mark Zuckerberg, Meta's CEO, described the change in a September 2026 podcast interview. With a chatbot you send one prompt and get one answer. With agents, he said, "you give it goals" and they keep working until the goals are met. He was talking about Meta's own agent, not Manus, and his company sells agents. Use his point to frame the idea, not as a recommendation.

Here it is on a real job. Under your video, a parent asks whether their children, six and eight, can manage the cliff walk and how they'll get back to the car. A chatbot drafts a reply from what it already knows. An agent can open the park's route page and shuttle timetable, note the last shuttle time with a link and the date it checked, and give you a short sourced packet. You check that packet before you write a word.

Manus calls itself an autonomous general AI agent: it plans, carries out and delivers finished work instead of only answering questions.

Each job you give it is a task. Each task runs in its own sandbox, a computer in the cloud with a web browser, a file system and software tools. In the sandbox Manus can browse sites, read the files you upload, run code to analyze data, and save the documents it creates.

For the cliff walk question, a task might go like this:

1. You write a brief: the viewer's question, the park's official pages as sources, and the output you want.
2. Manus proposes a plan.
3. It opens the route page and timetable in its cloud browser and records what each page says.
4. It saves a document with a source table.
5. You read it, open the links and decide what's true.

Because the work happens on a computer you can watch, you can see later which pages it visited and what it did. A later lesson shows how to replay a session.

You'll find plenty of Manus guides online, and many are out of date. Two changes in 2026 matter.

Ownership. Meta announced it was buying Manus in December 2025. On April 27, 2026, China's National Development and Reform Commission (NDRC) prohibited the deal and ordered it withdrawn. Manus announced the separation on August 11, 2026, and on September 1, 2026 said it had resumed independent operations under its founding team. As of October 28, 2026, Meta does not own or run Manus. A page that says "Meta's Manus" is out of date, and that includes some of Manus's own localized pages.

The product. Manus 2.0 shipped on September 28, 2026, with a redesigned workspace. Screenshots and menu names from before then may not match your screen. In this course, when a menu name isn't confirmed for the current version, the step says so and describes it in general terms. If your screen looks different, look for the function, not the exact label.

For creators, this has a practical effect. Old advice about Facebook login or Meta-era connectors may not apply now. Facebook login, for example, has been withdrawn.

Manus 2.0 lets you choose how much effort the agent puts into a task. There are three modes:

• Manus 2.0 Lite for quick work.
• Manus 2.0 for standard work.
• Manus 2.0 Max for hard reasoning or creative work.

How you might use them on a content series (these are illustrations, not rules):

• Lite: tidying a list of viewer questions you've already pulled out into a clean table.
• Manus 2.0: a sourced research packet on the cliff walk and the shuttle times.
• Max: a harder job, such as working out how a dozen sources that disagree relate to one another.

Modes matter because they affect what a task costs, and on the Free plan only one mode can use your daily credits. The next lesson covers credits, and Module 3 covers choosing a mode for each brief. The official docs don't say exactly where the mode picker is on screen, so look for a mode or model selector near the chat box.

Manus produces many kinds of output. These are the ones this course uses, each tied to a project you'll build:

• Sourced research reports. A claim-to-source packet for a viewer's question, where every fact has a link and an access date.
• Data analysis of CSV exports. Manus can read a spreadsheet file (CSV) you export, such as anonymized comments or survey answers, and group what's in it, with charts if you want them.
• Slide decks. Exported as an editable .pptx, a PDF or web slides, with a speaker-notes file. A media kit deck for sponsors is one example.
• One-page websites. For example, a media kit page that uses only your own dated analytics export, kept unpublished until you approve it.
• Documents and spreadsheets. Newsletter drafts, group posts, carousel scripts and source tables.

All of these are drafts and preparation. A deck goes out under your name, so you open every file and check every number before a sponsor sees it.

Manus can do more than this course teaches. These parts are left out on purpose:

• Voice cloning and avatars (for example the ElevenLabs and HeyGen connectors). Your face, voice and likeness stay yours. No agent in this course synthesizes or alters them.
• Cue and payments. Cue is a separate app, invite-only and in early access, and it can give agents a wallet and a phone number. Letting an agent spend money is a high-risk permission.
• Auto-publishing. Manus prepares work, and you publish it. Nothing goes out without your approval.
• The Instagram and Meta Ads connectors. They launched while Meta owned Manus and aren't on the current public connector catalog, so treat them as unconfirmed.
• Detailed video editing in Manus Studio's Video Editor, CapCut or Higgsfield.

One principle runs through every lesson: Manus does the legwork, and you stay the author. It can collect the 5:30 shuttle time and the page it came from. Only you walked those 300 steps with a tired eight-year-old, and only you decide what your audience hears.