Have Devin gather research sources, then verify them
Answer.AI tested Devin on paper-finding and transcript research jobs, but it stalled on HTML and produced vague, off-target output.
- What they did
- Answer.AI assigned Devin two research tasks through Slack in a Docker workspace with a browser, editor, and shell. One task was to collect an author’s 25 newest Google Scholar papers with Playwright and avoid paywalled material. The other asked for work on transcript summarization with precise timestamps.
- How it went
- Neither assignment succeeded. During the scraping job, Devin bogged down while decoding HTML and then went idle. On the research job, it came back with loosely related information plus code that missed the central issue. In a set of 20 tasks, the system logged 14 failures and 3 successes.
- Tips
- It appeared to handle vague, pseudocode-style directions better than exact research or scraping jobs. For anything that depends on page parsing or citations, have it surface the source list and ask for approval before it moves on.
Try it yourself with Devin
Look up [author/topic] and compile [number] recent papers, articles, or transcripts that are publicly accessible. For each one, provide a brief summary, the source link, and a note on any access or parsing problem you run into. Then synthesize the main themes across the set. If a page cannot be read cleanly, tell me which source failed and why, and stop for my approval before you submit, send, delete, pay for, or otherwise complete anything irreversible.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Collect meeting minutes into yearly text files
A developer used Claude in Chrome to copy three years of meeting minutes into text files for NotebookLM.
Use an agent to trace papers from a seed set
An agent can start with a few papers, follow citations both ways, and keep narrowing a review until the results stop changing.
Turn videos and papers into searchable Markdown libraries
They used OpenClaw to organize video and paper summaries into two separate Markdown-based libraries that could be searched later.
Use Manus to gather sources and draft a background summary
A researcher let Manus collect web sources on its own and turn them into a structured summary while they kept working.
Use a visual browser to compare online stores
A commenter used ChatGPT Agent as a visual web browser to check many stores for one item and see which ones would ship to their country.
Build a daily Reddit digest of AI tool chatter
Have an agent scan Reddit each day for discussion about new AI tools and write the most talked-about items into a document.