Research US stablecoin adoption with an agent
A Manus agent connected to Heurist's mesh ran deep research on US stablecoin adoption in 2026 after one approval step, finishing in two minutes.
Manus built a Google Form survey on undergraduates' views of AGI; the author hit a snag when asking it to fill the survey with synthesized student responses.
Manus built a Google Form survey on undergraduates' views of AGI; the author hit a snag when asking it to fill the survey with synthesized student responses.
Andrew Maynard gave Manus a short prompt asking for a Google Form on undergraduates' hopes and fears about AGI, mixing Likert and open-ended questions. Manus wrote a to-do list and chose to produce a Google Apps Script, which Maynard ran himself to create the form. When he asked it to submit 1,000 synthetic responses, it refused on ethics and rate-limit grounds.
The form was reasonably well designed. Manus declined to fill it, so Maynard had it run the simulated survey internally and write a 44-page report, which took a couple of iterations. All the data was made up.
Manus could probably have built the form directly, but that would have meant signing into a Google account on its virtual computer, which would be slow. The script route was quicker.
Create a Google Form survey for [target group] about their views on [topic], with [number] questions mixing multiple choice and short answer. Give me the shareable link and the editing link when done. Don't fill in or submit any responses yourself, and don't send the form to anyone without showing me first.
A Manus agent connected to Heurist's mesh ran deep research on US stablecoin adoption in 2026 after one approval step, finishing in two minutes.
A Make scenario runs at 9 AM, finds a YouTuber's newest video, has Hugging Face summarize it, and appends the summary to a Google Doc.
An n8n workflow scrapes Reddit posts and comments via Bright Data, extracts concerns and themes with Gemini, and writes reports to Notion.