Build presentations and spreadsheets from web research
A user had Agent Mode build a presentation, draft an eBook and turn web research into spreadsheets; the author said the design was not award-winning but saved hours. Task: Daria Cupareanu used ChatGPT Agent Mode to execute three distinct workflows: generating an AI-in-education presentation from a minimal prompt, drafting an eBook by synthesizing uploaded Substack posts with approved external sources, and crawling the Y Combinator directory into a Google Sheet. The agent autonomously browsed websites, manipulated files, and created documents, while the human provided initial prompts, uploaded assets, and reviewed outputs. This workflow demonstrates the agent's ability to handle multi-step research and content assembly tasks.; tools: ChatGPT Agent. Human role: Source evidence is unavailable regarding specific human roles.; limitations: Tasks consumed the Plus plan's 40 monthly agent messages and took 10–30 minutes each. Output quality heavily depends on precise upfront instructions regarding structure, entry counts, and design assets. The agent may stop early if limits aren't specified, and reliability issues like misclicks or CAPTCHA blocks can require manual intervention. It is best used for drafting rather than final production..
Tools
Published on Agent Work Atlas:
Overview
A user had Agent Mode build a presentation, draft an eBook and turn web research into spreadsheets; the author said the design was not award-winning but saved hours.
Workflow
Daria Cupareanu used ChatGPT Agent Mode to execute three distinct workflows: generating an AI-in-education presentation from a minimal prompt, drafting an eBook by synthesizing uploaded Substack posts with approved external sources, and crawling the Y Combinator directory into a Google Sheet. The agent autonomously browsed websites, manipulated files, and created documents, while the human provided initial prompts, uploaded assets, and reviewed outputs. This workflow demonstrates the agent's ability to handle multi-step research and content assembly tasks.
Reported result
The presentation was functional but lacked award-winning design quality. The eBook draft resulted in 11 pages instead of the requested 20 and was not publish-ready, highlighting dependency on input specificity. The Y Combinator spreadsheet was raw but accurate enough for use, though it stopped at 20 entries due to unspecified limits. Overall, the agent saved hours of manual research and assembly but required human refinement for final quality.
Human role
Source evidence is unavailable regarding specific human roles.
Limitations
Tasks consumed the Plus plan's 40 monthly agent messages and took 10–30 minutes each. Output quality heavily depends on precise upfront instructions regarding structure, entry counts, and design assets. The agent may stop early if limits aren't specified, and reliability issues like misclicks or CAPTCHA blocks can require manual intervention. It is best used for drafting rather than final production.
Try it
Act as a research assistant using ChatGPT Agent Mode. I need you to create a lead list from [replaceable source URL, e.g., Y Combinator directory]. 1. Visit the URL and extract [replaceable fields, e.g., company name, website, tagline] for the first [replaceable number, e.g., 50] entries. 2. Create a Google Sheet in my Drive with these columns. 3. For each company, search for a public contact email and append it to the sheet. 4. Stop if you encounter a CAPTCHA or login wall and report the status. Before executing, confirm the target number of entries and output format. Do not proceed without my confirmation on the scope to avoid wasting agent credits.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.