Manus AI: A Guide With 5 Practical Examples

After processing this, here’s the result: Coffee shop sale analysis.

I quite like the overall presentation of this one, and I even sent it to my friend.

The part that I was most interested in was the “Time Analysis”. From the times I’ve been there, it always felt like mornings were quiet while afternoons were quite busy. I was curious to see what the impact would be of shifting the opening hours to a later schedule.

Unfortunately, something went wrong with the way the data was processed by Manus AI. By putting the time chart it generated next to the real one, we can see that it doesn’t match.

At first, I thought that maybe it just shifted the data because of timezone issues, but it’s more than data.

Not everything is wrong, though. For example, the total revenue we see on the first page matches the real revenue. However, the mistake with the times made me skeptical about trusting the rest of the results. Imagine making business decisions based on those results without double-checking. The consequences could be really bad.

One way in which I feel this was useful is that it gave my friend inspiration on what data to look at. If you ever use Manus for this sort of task, always make sure to double-check the results.

2. S&P 500 Predictions

It's hard to overlook the recent downturn in the economy, a situation that has been influenced, in part, by tariffs imposed by the United States. The fluctuations have been notably reflected in the S&P 500, which has experienced significant declines. The S&P 500 is a stock market index that includes 500 of the largest publicly traded companies in the United States, providing a snapshot of the market's overall performance.

Given these circumstances, I was curious to explore if Manus AI could offer projections on how long it might take for the market to recover. I used the following prompt:

“Analyze the current downturn in the S&P 500, considering recent global tariffs imposed by Donald Trump. Using historical data on similar economic events and market downturns, identify potential scenarios for how low the S&P 500 could go (i.e., expected minimum levels), and estimate a timeline for recovery to previous levels. Include historical comparisons, relevant economic indicators, and confidence ranges where applicable.”

Before we analyze the results, please be aware that this is not financial advice. It’s just an experiment to see how Manus AI handles such a complex question that involves analyzing historical data and combining it with current data to predict possible scenarios.

Manus AI analyzed several potential scenarios from mild to extreme. Based on the results, it created and hosted a website with the results (which you can access here), containing several sections.

Some of the sections include:

• An overview of the results from its research.
• A summary of the current situation.
• Historical data of similar situations.
• Analysis of potential minimum levels the S&P 500 could reach in each scenario.
• Recovery timelines for each scenario.
• A page with a few interactive charts that aggregate its findings.
• A conclusion explaining what it thinks is the most likely situation.

Approach breakdown

Here’s a breakdown of how Manus AI generated these results:

• First, it gathered current S&P 500 data using the YahooFinance API, which provided the most up-to-date market information. It wrote a Python script that extracted the current index value, recent high, and calculated the current decline.
• To understand tariff policies, it researched multiple sources, including Reuters and BBC articles, which revealed the comprehensive nature of the new tariffs: a baseline 10% on all imports, up to 60% on Chinese imports, and targeted rates for specific countries.
• Then, it identified and analyzed three key historical tariff events: the 2018-2019 US-China trade war, Bush's 2002 steel tariffs, and the 1930 Smoot-Hawley Tariff Act. For each, it examined market impacts, recovery patterns, and economic consequences. The 2018-2019 trade war provided the most relevant comparison, showing how markets initially declined but recovered after policy resolution and Federal Reserve intervention.
• To calculate potential minimum levels, it created a Python script that developed four scenarios (Mild, Moderate, Severe, and Extreme) based on historical tariff-related declines. It used Monte Carlo simulations with 10,000 iterations to generate confidence intervals, which showed the S&P 500 would most likely bottom between 4,300 and 4,900 (50% confidence interval).
• For recovery timeline estimates, it wrote another script that analyzed historical recovery patterns and used statistical modeling to project four recovery scenarios. The most probable outcome (50% probability) suggested a 9-month recovery period from the market bottom, with a U-shaped pattern:

3. Product Recommendation

Next, I was wondering how Manus AI would fare on a product recommendation task. Finding the best product to buy can be time-consuming and it’s something I hate to do.

I live in Portugal, and some rooms in my house get very humid during the winter. I wanted to buy a dehumidifier, and I was curious to see if Manus AI could help me do research and provide some recommendations on what to buy.

This is the prompt I used:

“​I reside in Portugal and am experiencing high humidity levels in my home, often exceeding 90%. I need assistance in selecting a dehumidifier suitable for a room up to 25 square meters. The dehumidifiers should be available for purchase in Portugal. Please create a comprehensive report that includes:​

- Product Recommendations: Identify several dehumidifier models appropriate for a 25-square-meter room, considering various price points.​

- Product Comparisons: Compare the recommended models based on key features such as dehumidification capacity, energy efficiency, noise levels, water tank capacity, and additional functionalities (e.g., air purification, laundry mode).​

- Availability in Portugal: Ensure that the recommended dehumidifiers are available for purchase in Portugal, either through local retailers or online platforms that ship to Portugal.​

- Online Reviews: Incorporate insights from online reviews to assess the reliability and performance of each model.”

Here’s the result: Dehumidifier product recommendations.

Note that this is just for testing purposes; I didn’t try any of those products, so I have no idea whether or not the results are accurate.

How Manus AI built the recommendations

When using Manus AI, we can see a step-by-step breakdown of what it is doing. To build the product recommendations, it looked at data from:

• Local retailers.
• Customer review organizations that gather reviews from consumers about products.
• Forum and Reddit threads about people discussing the products.
• Technical resources to match the product specifications to my specific request.
• Climate information.

These sources give me some confidence in the results because the steps are quite close to what I would do if I were to do this research manually.

4. Finding a Job

Manus AI is good at searching the web and combining the findings into an easy-to-use website. Let’s see if we can use it to find a job. I uploaded my CV and wrote this prompt:

“I am a full-stack engineer seeking a new job opportunity. Please find job listings that meet the following criteria:

Job Titles: Full-stack Engineer, Teacher, or Researcher.

Salary: Minimum of USD 100,000 per year.

Work Environment: Fully remote.

Tech Stack Preference: React (front-end) and Python (back-end).

Interest Area: I’m interested in quantum computing and open to positions related to it, even though I have no prior experience in the field.

Provide a list of suitable job openings, and include direct links to the job postings or application pages. Prioritize jobs that align most closely with my preferences.”

This is what it created: Job posting results.

At first sight, it seems like the result is quite impressive. It looks like it generated a website tailored to my needs with searchable job postings that match my criteria. However, after further inspection, I noticed that it only contains 12 job postings.

Moreover, only one of them contains a link to the company website. For all the others, this button is disabled.

Out of all the examples, this was the most disappointing one. I thought the task would be easier to accomplish than others because it doesn’t involve any calculations. It only asks Manus AI to search the web for jobs that match the given criteria and aggregate the results.

5. Learning a New Skill

In this final example, I tested Manus AI on learning a new skill. To be able to better asses the results, I selected a skill that I already know: Astro Photography.

This is the prompt I used:

“I want to learn astrophotography using my Nikon Z7II camera. Search the web and design a comprehensive, beginner-to-advanced course specifically tailored to my camera and needs. The course should include:

- Fundamentals of Astrophotography

- Camera-Specific Guidance

- Essential Gear

- Shooting Techniques

- Post-Processing Workflow

- Learning Resources

- Checklists & Planning Tools”

Out of the examples, this was by far the one that took Manus AI the most time to complete. It also took significantly more credits to accomplish than other tasks:

This is what Manus AI built: Astrophotography course.

There’s a lot to unpack here, so I’ll summarize the main aspects of the content it generated:

• The course covers a lot of topics and has the necessary information to get someone started.
• The sections are too succinct. For instance, it will say that a specific setting is recommended, but won’t explain how to set it.
• While the information is good, the delivery isn’t. The UI is hard to use because most content is behind a collapsible section and requires many clicks to access.
• Another issue with the delivery is that it doesn’t tell a story, and it feels like an instruction manual without any visual aid.
• It contains a section with external resources like courses and YouTube videos, which are quite useful.

As with most experiments, I think it is a good starting point, but definitely not a final product. In this particular case, I think the raw information it gathered is more useful than the actual website it created to browse that information.

Conclusion

In evaluating Manus AI, I find myself with a mix of positive and negative impressions. On the positive side, Manus AI excels at gathering and condensing large amounts of information efficiently. This automated assistance significantly cuts down on time, helping us focus on analysis instead of data collection.

However, the trust we place in Manus AI's results is tempered by occasional inaccuracies. When examining something familiar, like coffee shop data analysis, we can personally verify the findings and occasionally spot errors. But when it comes to more complex cases where we lack the expertise, can we trust the results?

I think that Manus AI is better as a tool to bootstrap an analysis, but I wouldn’t blindly trust it and make decisions based on the raw results.