Manus 101: Getting Started (The AI That Runs the Whole Project)

Most AI tools answer you. Manus works for you. It's the difference between asking a question and handing off a project, and once you feel that difference, you start noticing how many of your "questions" were secretly projects all along.

Here's the plain-English starter: what an agent actually is, the real project I ran through it, the briefing template that gets good work back, and how to review what it hands you.

What Manus is

Manus is an AI agent. Give it a goal ("research this market and tell me the revenue potential") and it plans the steps itself: it searches, reads, compares, compiles, and comes back with a finished deliverable. A chatbot answers in seconds and waits for your next message. Manus disappears for a while and returns with the report.

And the feature that earned it a permanent spot in my stack: it cites its sources. Every claim traces back to where it came from, so I don't feel like it's guessing or hallucinating. For research that feeds real decisions, that's the whole ballgame. It's why, for research tasks, I reach for Manus or Perplexity over Claude or ChatGPT: they're simply better research agents, and I trust what comes back.

The real example

A business partner of mine wants to expand into a market we haven't worked in before. Instead of blocking out a research week, I gave Manus the brief: the business overview, who our customer is, and what we were considering. Then the questions: where's the area of opportunity? What are our blind spots? What do we need to think about that we haven't? What's the revenue potential, and the profit potential?

What came back was the preliminary research I'd otherwise have assigned to a person for a week: sourced, structured, and honest about what it couldn't verify. It didn't make the decision. It meant the decision started from a real map instead of a hunch, and my partner and I spent our meeting arguing about strategy instead of gathering facts.

Getting started (your first hour)

1. Create an account at manus.im.
2. Pick a real project, not a test question. Agents are wasted on "what's the capital of France." Give it something with steps that you actually need. Best first project for almost everyone: competitor research (below).
3. Write the brief like you'd brief a smart contractor. This is the skill. The quality of the brief is the quality of the output, which is the same training principle that runs everything on this site.
4. Let it run. Multi-step projects take real minutes, sometimes longer. This is normal; it's doing the reading.
5. Review like a manager, not a fan. Skim the sources, poke the surprising numbers, and send it back for a revision pass with specific feedback. It takes notes like a team member.

The briefing template (steal this)

PROJECT: [one line: what you want to exist at the end]

CONTEXT: I run [business]. Our customer is [who]. We're considering
[the decision this research feeds]. What I already know/believe:
[2-3 bullets, so you don't get your own assumptions repeated back].

QUESTIONS TO ANSWER:
1. [Where's the opportunity?]
2. [What are our blind spots / what do we need to think about?]
3. [What's the realistic revenue potential? Profit potential?]
4. [Who's already doing this well, and how?]

DELIVERABLE: [a structured report / a comparison table / a one-page
brief I can forward]. Cite your sources for every claim. Flag
anything you couldn't verify instead of smoothing over it, and note
where sources disagree.

Two parts of that template do the heavy lifting. The "what I already know" bullets stop it from spending its effort re-proving your starting point. And "flag what you couldn't verify" is the honesty valve: a good agent report tells you where its map has fog.

Projects worth handing it

• Competitor research: offers, pricing, positioning, and what changed lately, compiled into one comparison. Everyone needs it, nobody enjoys doing it.
• Market entry scouting: the example above. Opportunity, blind spots, revenue and profit potential.
• Vendor shortlists: "find me the five best options for [need], with pricing approaches and known complaints, in a table."
• Content landscape mapping: what's already been written on a topic you're planning, where the gaps are, who owns the conversation.
• Prep dossiers: a partnership or big pitch coming up? Background on the company, the people, their recent moves, their public priorities.

When Manus vs. Perplexity

I go between the two depending on the job, and the split is shape, not quality. Perplexity is question-shaped: ask, get a cited answer or briefing, done in minutes. Manus is project-shaped: brief it, let it run, get a deliverable. My rule of thumb: if you can phrase it in one sentence, it's Perplexity. If briefing it properly takes a paragraph (like the template above), it's Manus.

Mistakes to avoid

• One-line briefs. "Research the fitness market" gets you a Wikipedia summary with better formatting. The template exists because context in = quality out.
• Treating the report as the decision. It's the map, not the judgment. You (and the humans you trust) still make the call; the agent just means you make it informed.
• Skipping the source check. It cites everything precisely so checking is fast. Check the numbers that would change your decision.
• Not sending it back. First drafts from agents are like first drafts from anyone: 70 to 80% there. One specific revision pass ("go deeper on pricing, the competitor section is thin") closes most of the gap.
• Using it for voice work. Reports, yes. Anything that has to sound like you, no. That lives where your rules and examples live.

FAQ

What is an AI agent, in plain English? An AI that doesn't just answer but takes actions toward a goal: planning steps, browsing, reading, compiling, and producing a deliverable without you steering every move.

Is Manus better than ChatGPT? Different job. ChatGPT is a conversation; Manus is a contractor. For multi-step research projects, agents are built for the shape of the work.

Can I trust what Manus produces? Like a strong junior researcher's work: genuinely useful, and you check the sources on anything that costs money. Since every claim is cited, checking takes minutes, not hours.

What should my first Manus project be? Competitor research, using the briefing template above. High value, zero ambiguity about whether the output is useful, and it teaches you the briefing skill everything else builds on.

How long does a project take? Real minutes to considerably longer, depending on scope. The mental model is "assign it and check back," not "wait at the spinner."