Manus AI: how to use the AI agent in 2026 – step by step

Section 01What Manus is and how an agent differs from a chatbot

Manus is a service from a Singapore company that, in March 2025, was one of the first to show the general public a real AI agent. The name is Latin for "hand," and that sums up the idea: the AI got hands. It still answers with text, but it also acts on its own on the web and on a computer.

The difference from a regular chat is easy to see. You ask ChatGPT or Claude: "find me 15 nutritionists who sell challenges and list their prices." The chatbot answers from what it already knows or runs a couple of searches. Some prices will be outdated, some links broken, and you'll still have to check the table by hand.

Manus handles the same request by spinning up a separate computer in the cloud. It opens a browser, searches, visits each site, scrolls through the pricing pages, copies the data, cross-checks it, and builds the spreadsheet. You watch the whole process in a window on the right: which sites it opened, what it read, where it got stuck. In 10–30 minutes, the file arrives.

That gives you the main rule of working with an agent. A chatbot is good when you need to think something through together. An agent is good when the same action has to be repeated many times, the kind of thing you'd otherwise spend a whole evening on: open, find, copy, combine.

If agents are completely new to you, start with the basics: what an AI agent is in plain words. It explains, without jargon, how an agent differs from a bot and from a chat AI.

Section 02What Manus 2.0 can do

Manus 2.0 came out on September 28, 2026. Inside is a new agent engine called Cascade. According to the developers, in one of their tests it used 32% fewer resources and finished tasks 28% faster than the previous version. For you, that means fewer credits spent on the same work. Manus runs in the browser, in a desktop app, and in a mobile app.

There are also tools for advanced users: a cloud computer for projects that need permanent hosting, a video editor, and a mode where the Manus Studio app controls your own computer. Alongside Manus 2.0, a separate app called Cue launched for personal agents, each with its own email address, phone number, and budget. You won't need any of this in the first few weeks. Start with research, spreadsheets, websites, and slides: that's where the agent pays for itself fastest.

Section 03Is it free: credits and plans

Manus runs on credits. Credits are spent while the agent thinks, works in its cloud computer, and calls outside services. A finished task, saved files, and a published website no longer eat credits. Before a task starts, Manus shows an estimated cost, and it's worth checking every time.

Free. You get 300 credits every day. On a free account, daily credits add up to no more than 1,500 a month, and only one task runs at a time. That's enough for a few small research tasks to figure out whether you need an agent at all.

Three credit rules that save money. Monthly credits expire at the end of the billing period and don't roll over. Credits you buy on top last as long as your paid subscription is active. Free bonus credits never expire. And if a task ends up costing more than you had in your balance, the balance goes negative and gets covered by your next top-up.

Which plan should you start with? Free. Run small tasks on it for two or three weeks and watch how many credits your typical work takes. If you hit the limit every day, the $20 plan is the logical next step.

Section 04How to sign up

Signing up takes a couple of minutes. The only real trap is landing on the wrong site.

There's a Manus mobile app, but it's easier to start in a browser on your computer: you can see what the agent is doing, and grabbing the files is simpler.

Section 05Your first task, step by step

Let's take a task almost everyone who sells their services needs: collecting competitors with their prices. By hand, that's an evening. For the agent, it's half an hour without you.

Section 06The task formula for an agent

An agent is diligent but literal. A vague task like "research my competitors" turns into twenty minutes of wandering and a report about nothing. A clear task in six parts gets results on the first try.

Beginners skip the fifth part most often, and that's a mistake. The agent acts on the real internet. If you don't write "don't buy anything and don't submit any forms," it may well fill out a form on a competitor's site to find out the price. The sixth part saves credits: without it, the agent doesn't know when to stop and keeps "improving" the result.

One more habit that pays off: ask the agent to ask you questions first. Add at the end: "if anything in the task is unclear, ask me first, then start." A couple of clarifications at the start cost less than redoing the whole task.

Section 0710 ready tasks for your business

Copy them and swap in your own niche and numbers. Every task is already built on the formula from section 6.

Task #10 is already scheduled work, more on that below in section 10. For how agents like this work inside a business and what else you can hand them, read how to build an AI agent without code.

Section 08Wide Research: competitors and the market

A regular Manus task runs with one agent: it visits sites one by one. With 100 sites, that takes a long time, and by the end the agent can start losing details. Wide Research solves this differently: Manus launches many agents at once, each takes its own part of the list, and at the end everything comes together in one spreadsheet.

Where it helps. Break down every competitor in your niche, across all the search result pages. Collect a hundred places to advertise. Go through a course catalog on a platform and write down prices and programs. Check which of your old partners are still in business.

Competitor analysis is one of the tasks almost any product positioning starts with. One of our clients, Magomed Dibirov, who runs launches, started exactly there: he analyzed competitors, pulled out the selling points, polished his blog, made a lead magnet, and ran ads. He got 200 subscribers, invited them to consultations, and earned 230,000 rubles on his first launch. Back then, competitors were analyzed by hand. Now you can hand the first part of that work, the data collection, to the agent and keep the important part for yourself: the conclusions and the message.

Wide Research eats credits noticeably faster than a regular task, because there are many agents. Always check the cost estimate before you start. On the free plan, start with a list of 20–30 items.

For quick "find and explain" questions with sources, an AI search engine is more convenient. Here's how to use one: Perplexity, the AI search engine with sources. Use Manus when the result is a big spreadsheet or document gathered from dozens of sites.

Section 09Websites and slides in Manus

Manus can build a website from your description and publish it at a link right away. For a challenge landing page, a consultation page, or a simple personal site, that's enough. Write the task using the same formula: which sections, what text, where the button leads, what style.

Three tips that save you rework. Give it your own copy if you have it: the agent will move it onto the page, and you won't have to clean out its generic phrases. Explicitly forbid made-up testimonials, numbers, and case studies: write "leave space for testimonials, don't make up any text." Ask to see the site on a phone: most of your clients will open it there.

Make edits as messages in the same task: "make the button bigger," "change the first screen background to light," "remove the price section." If you need the site for the long haul and with a lead form, think about where it lives and what happens if you cancel your subscription. For a serious project, compare other options: Lovable is a service built specifically for websites, and the guide to vibe coding shows how to build a site with Claude.

Manus makes presentations in a separate slides mode. A two-step approach works well. First, ask for a plan: "suggest a structure for a 15-slide presentation, one idea per slide." Fix the plan. Then: "make the slides from this plan, large text, minimal words." If you need presentations often, also look at Gamma, a service that does nothing but slides.

Section 10Scheduled tasks

The most valuable part of an agent starts when it works without you. In Manus, any task can be set to repeat: every day, every week, at the hour you choose. On paid plans, up to 20 of these tasks can run at once.

What's worth scheduling if you sell services and courses:

• every Monday – a niche news digest with links, like task #10;
• once a week – a price check on your five main competitors: what changed, who launched a promotion;
• once a month – a roundup of fresh audience questions from forums and reviews for your content plan;
• every morning – a summary of new mentions of your name or your school in public sources.

Manus 2.0 also added automations: a task starts when something happens in a connected service. A lead email arrives, and the agent puts together a brief on the person from public sources. A card is added in Notion, and the agent drafts something from it. Start with one such connection and watch for a week how many credits it eats.

Every scheduled task spends credits on every run. Five daily tasks at 100 credits each come to 15,000 credits a month, more than the $40 plan. Do the math up front and schedule only what you actually read. If you need an agent that works around the clock in your Telegram and answers clients, that's a different class of task, for example an AI agent in your Telegram DMs.

Section 11How not to burn your credits

Complaint number one about Manus in reviews: "my credits ran out in two days." Almost always, the cause is how the task was worded. Seven habits that fix the problem:

1. Write the whole task at once. Every clarification after launch is another round of work. Better to spend five minutes on the task text than to correct it three times later.
2. Set a limit. Write "20 competitors" instead of "all competitors." Ask for 5 slides as a test, you can order forty later.
3. Give a finish line. Without it, the agent keeps improving the result and spending credits.
4. Check the estimate before you start. If the number scares you, cut the task down.
5. Stop it if it goes off track. If you see in the window that the agent is reading the wrong sites, hit stop and fix the task.
6. Think in a chat, act in the agent. Discussing ideas, writing posts, and coming up with an offer is cheaper in Claude or ChatGPT. Bring to Manus only what requires action: visiting sites, collecting, publishing.
7. Use your daily credits. 300 credits arrive every day and are spent first, right after bonus credits. Small tasks are easy to run daily, while big ones are worth saving monthly credits for.

Section 12Let Claude write your Manus tasks

Writing a task with all six parts on the first try isn't easy. A handy move is to ask Claude to turn your rough idea into a clear task for the agent. Claude does the thinking for free or within your subscription, and your Manus credits then go only to the actual work. If you don't have Claude yet, sign up at claude.ai. For where to start with Claude, see learning Claude from scratch.

A request you can copy as is:

I ran this request myself. Here's what Claude returned:

Look at what Claude did. It added the goal ("before I revise my prices"), without which the agent doesn't know what to look for. It listed the sources. It laid out the spreadsheet fields so they can be compared. It closed three dangerous gaps: guessing prices, signing up, and buying. It set a clear finish line and asked for a short conclusion. That's exactly the formula from section 6, put together in a few seconds.

If you're not sure you need Manus at all, or whether one chat AI subscription is enough, read the honest comparison Claude vs ChatGPT in 2026: Claude handles many of the tasks in this article on its own, without a separate agent.

Section 13What not to trust the agent with

The agent works on the real internet, on your behalf, with whatever access you've given it. So a few rules are worth following from day one.

• Don't share passwords or text message codes. If a task requires logging into your account, decide whether you're okay with the agent acting there. Public data is enough to start.
• Upload client data anonymized. You can hand over a lead export for analysis without names, phone numbers, and emails: they aren't needed for the math.
• Explicitly forbid purchases, sending emails, and publishing unless the task needs them. And check the result before anything goes out to clients.
• Double-check facts. The agent takes prices, numbers, and legal wording from the internet, and the internet has outdated pages.
• Keep important things on your side. In summer 2026, Manus users were asked to export their data by a certain date because of the ownership change. Keep copies of your spreadsheets and files on your own drive.

Section 14When something goes wrong

Section 15Manus vs Claude, ChatGPT, and Perplexity

These are different tools for different jobs, and "which is better" makes no sense without a task. Here's how I'd split the work.

For many of my clients, the setup is simple: Claude for thinking and writing, the agent for routine data collection. Our school has 10,000+ graduates, and over these years one conclusion holds: the people who win are the ones who hand repetitive work to the machine and keep the decisions for themselves.

Section 16The whole path in a minute

Start today with one action: take task #1 from section 7, swap in your own niche, cut it down to 10 competitors, and run it on the free credits. In half an hour, you'll have a spreadsheet that used to take you an evening, and you'll know whether the agent is worth paying for.