Manus Step-by-Step: How to Delegate Tasks Without Burning Credits

👁 116 reads

Manus is an agent, not a chatbot: you give it a goal, and it plans the steps, browses pages, runs code, and delivers a finished result - a report, presentation, or website. It works on its own cloud computer, so you can delegate a task and come back for the result. This tutorial shows you how to delegate tasks without burning credits on trial and error.

These steps are based on official Manus materials, including the beginner’s command guide. I covered how Manus differs from regular chatbots in our Manus guide.

Before You Start: Describe Your Task in Five Sentences

Before you delegate your first major task to Manus, try describing it in five sentences: what should be created, why, from what sources, in what format, and what to avoid. If you can’t do this, the agent won’t know what you want either, and it will burn credits guessing. This description also becomes a ready-to-use prompt you can paste in.

Prepare the files and links the agent should consider before you start. Including them upfront saves back-and-forth questions during execution.

Step 1: Chat vs Agent Mode

Manus has two working modes. Chat mode is for quick answers. Agent mode executes tasks independently: it opens pages, scrolls through them, copies fragments, and runs code. You can watch what it’s doing in real time. For simple questions, chat is enough - save the agent for tasks that require multiple steps.

Step 2: First Task - Narrow and Specific

Official materials recommend starting with a narrow task, such as market research ending in a report. In your prompt, specify the goal, constraints, and what the finished result looks like. Instead of “research competitors,” write: find five online programming course providers, compare their prices, course length, and class format, deliver the result in a table with links to sources.

You can attach files or links for Manus to consider in the task.

Step 3: Delegate the Entire Task

The Manus beginner’s guide suggests an interesting approach: instead of dictating every step, give the agent responsibility for the whole task and ask it to determine what needs to be done, ask you questions, and lead the process. Manus will ask for details before it starts, which saves credits spent on wrong assumptions.

Step 4: Presentations

Manus builds presentations from descriptions or research results: it structures the story, writes content, and formats slides for editing. You edit the text directly, and you can fix individual slides with a separate command. Manus can also connect to Google Drive, Notion, or local files and compile one presentation from multiple documents.

Step 5: Time and Credits

Simple presentations take a few minutes, large reports and websites can take up to half an hour. Credit consumption depends on task complexity. The most important warning from the documentation: if a task exhausts credits halfway through, it stops and leaves an incomplete result. That’s why you should start with smaller tasks until you know what they cost.

For complex tasks, Manus can divide the work into parts executed in parallel. You don’t enable this mode manually - the agent decides on its own.

Step 6: Recurring Tasks

Tasks you repeat can be set up once and Manus will execute them on schedule: daily reports, weekly reviews, monthly analyses. You describe the task specifically and set the schedule, and you’ll find the list of active tasks in settings. There you can pause, modify, or review results from previous runs.

Template for a Good Prompt

• Goal: what should be created.
• Scope: how many sources, what market, what time period.
• Output format: table, report, presentation, website.
• Constraints: what not to do, what sources not to use.
• Request for questions before starting work.

Common Mistakes

• Vague prompt without specifying output format.
• First task too large, exhausting credits halfway through.
• Not checking sources in the finished report.

How to Evaluate Your First Result

After your first task, check three things. Does the result have the format you requested, for example a table instead of long text. Do the sources exist and say what the report attributes to them. Open at least three links. How many credits did the task consume, because this lets you plan larger ones.

If the result was good, save the prompt. You’ll delegate similar tasks in the future by just changing the topic, and if they repeat regularly, you’ll set them up as recurring tasks.

In such tasks, it’s worth immediately requesting sources for each table row. This makes checking the result take just a few minutes: you open links for prices that seem suspiciously low or high and compare. Without sources, you’d have to repeat the entire research to verify the report, which defeats the purpose of delegating it to an agent.

Over time, it’s worth building your own list of proven prompts. A prompt that once gave a good result is more valuable than a new one written from scratch, because you know its cost in credits and output quality.

// read moreManus AI Pricing: Credits, Plans & How Many Tasks You GetGuidesManus: AI Agent That Completes Tasks Instead of AnsweringAI ToolsManus AI: 7 Tasks It Handles Well and Where It FailsAI Tools

Kinetyra