Agents In Use
What’s new todayLast update
— use cases added today

Devin builds an unanswered-email dashboard

Jamesin Seidel reports Devin researching Gmail documentation and building, testing and deploying an unanswered-email dashboard after he supplied working authentication himself.

FROM THE SOURCE

Original prompt

Shared by the source author. See source notes for context.

Can you also make sure that the emails are in my "Primary" inbox?

Evidence at a glance

Sourced account
Source author
Jamesin Seidel
Original account
Read the source ↗
Published
11 Dec 2024
Source link checked
15 Oct 2026
Reported result
Jamesin Seidel reports Devin researching Gmail documentation and building, testing and deploying an unanswered-email dashboard after he supplied working authentication himself.
Human involvement
Seidel implements authentication, gives follow-up constraints and reviews the interface. Supabase-backed ignore functionality was unfinished in this account.

A linked account supports attribution. It does not mean we independently reproduced its result.

Report an issue with this caseOpen correction email

Opens your email app with the case link included. Nothing is sent automatically.

How the agent did it

  1. Seidel asks Devin to reproduce an unanswered-email view using Gmail.
  2. After authentication fails, Seidel writes that code himself with Cursor.
  3. Devin researches Gmail documentation and implements recent-message and Primary-inbox filtering.

Jamesin Seidel wanted to recreate an unanswered-email view that he missed after changing CRM tools. He gave Devin the goal of finding important Gmail messages that still needed a response. The first obstacle was authentication, and the agent struggled long enough that Seidel stopped and wrote that portion himself using Cursor.

With authentication working, Devin implemented a function to identify messages from the previous week. Seidel then asked it to restrict the search to the Primary inbox. He describes watching the agent open a browser, research Google's developer documentation and implement filtering and thread-response logic using the Gmail API.

The next request was a frontend. Seidel reports that Devin wrote it, tested it locally and deployed it to a public link. The deployed interface is the concrete outcome of the account. A later Supabase-backed ignore function was still being worked on while he wrote, so it should not be counted as finished functionality.

The useful workflow includes a significant human rescue at the boundary of account access. It also left Seidel uncomfortable with the distance between the agent's internal work and his own view of the code. This was his initial experience over roughly twelve hours, rather than a sustained reliability or email-classification benchmark.

Prompt
Original prompt

Source: jamesin.substack.com · Jamesin Seidel

Frequently asked questions

Did Devin finish authentication unaided?

No. In says he gave up on that part and wrote it himself using Cursor.

What did the browser research resolve?

It helped implement Primary-inbox filtering and tracking replies within Gmail threads.

Was the ignore function complete?

The article says Devin was still connecting Supabase and adding that behavior when it was written.

More details

Remaining steps

  1. It creates a frontend, tests it locally and deploys a public interface.

About this account

Devin researched Gmail developer documentation, implemented message filtering and thread-response logic, created and locally tested the frontend and deployed a public interface.

Firsthand account opened and read. Outcome is source-reported, not independently reproduced. First twelve-hour experiment. Human supplied authentication; later database-backed ignore feature still in progress. Original prompt box contains a short excerpt of the author's published actual-run instructions.

Credit: Jamesin Seidel

Source published 11 Dec 2024

Source checked 15 Oct 2026

Tools and topics

Devin · Gmail API · Browser

Devin · Gmail API · Browser

Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.