Manus AI Will Handle It? How (Not) to Retrieve GDP Data from the World Bank Open Data

Manus is a general-purpose AI agent designed to carry out multi-step tasks on behalf of users. It promises to understand instructions, break down complex goals, and deliver results—whether that means writing code, automating research workflows, or, in our case, retrieving public data from the web. Touted as the world’s first truly autonomous AI agent, Manus was launched with the bold claim that it could handle a wide variety of tasks with ease, without requiring manual intervention. We tested it with a seemingly simple challenge: collect GDP per capita data for Hungary, Poland, Slovakia, and Czechia from the World Bank Open Data platform, covering the years 2000 to 2020. The goal was to download the dataset, export it to Excel (countries as rows, years as columns), generate a line chart with trends, and provide a short comparative analysis. What happened was less impressive. In this test, the outcome was a clear failure: Manus kept “working” for nearly an hour and a half before finally admitting it couldn’t complete the task and was waiting for us to provide the data instead.

Prompt

To evaluate its real-world usefulness, we gave Manus a structured, multi-step task: gather GDP per capita data (in current US dollars) for Hungary, Poland, Slovakia, and Czechia, covering the years 2000 to 2020, using the World Bank Open Data platform. Once retrieved, the data was to be exported into an Excel spreadsheet—formatted with countries as rows and years as columns. The next step was to generate a line chart showing the GDP trends for each country, and finally, to produce a short analysis.

Output

The test began at 17:15 with the prompt clearly outlining each step of the task. At first glance, Manus appeared to understand the assignment perfectly. It confidently restated the goal, mentioned the World Bank source, and promised regular progress updates.

At first, it looked like Manus was progressing smoothly—announcing that it was formatting the data for Excel export. But soon, the process got stuck in a loop of vague “text editor issues.” Despite claiming to generate a script, no actual file or output appeared.

Manus then moved on to a supposed diagnostic phase to “fix the GDP data collection step.” But nothing changed—just another long sequence of identical error messages. By this point, the task had been running for over 45 minutes (17:58), with no sign of actual data, charts, or analysis.

By 18:22, Manus shifted from simulation to clarification. It now asked whether we could provide a data source—either a CSV file or an API—so it could proceed. We told it to use the World Bank website. In response, it agreed and promised to fix the processing step.

It briefly seemed like Manus was making progress. The interface showed it navigating to the World Bank DataBank portal, selecting the relevant countries and indicators, and preparing for data download. On the surface, this looked promising—like the agent had finally located the right dataset. But once again, this activity ended in the familiar loop of “handling text editor issue” messages, with no actual data retrieved and no further steps completed.

Finally, at 18:35—roughly an hour and a half after the task began—Manus gave up. It informed us that it would wait for us to download the GDP data from the World Bank DataBank. Once we had the file, it said, it could help with processing, charting, and analysis. In short: the AI agent handed the core of the task back to the user.

Recommendations

Although this test failed, it’s still possible that the Manus agent could perform better with different datasets. The platform shows potential—particularly in areas like code generation and structured output formatting—but its ability to autonomously retrieve and process data from public web sources is currently limited.