Devin prepares Fern docs for human revision
Fern writer Devin Logan reports using a Slack agent built on Devin to draft documentation PRs, then reviewing and restructuring them with engineers and Claude Code.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Use the authorized engineering thread to draft a documentation PR. Preserve existing structure, identify the exact new behavior and include a scoped changelog entry. Flag terminology and assumptions for human review.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Devin Logan
- Original account
- Read the source ↗
- Published
- 10 Mar 2026
- Source link checked
- 15 Oct 2026
- Reported result
- Fern writer Devin Logan reports using a Slack agent built on Devin to draft documentation PRs, then reviewing and restructuring them with engineers and Claude Code.
- Human involvement
- Logan reviews and edits every page; engineers provide technical corrections. Fern Writer drafts and handles repository changes under that review.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Logan gives Fern Writer a Slack discussion about PDF-export filtering.
- The Devin-based agent creates a PR updating the existing documentation.
- Logan pulls the branch and uses Claude Code for structural and changelog revisions.
Devin Logan, Fern's human technical writer, uses Fern Writer, a Slack-based agent built on Cognition's Devin. The similar names refer to different participants. Logan gives the agent engineering context and receives documentation changes as pull requests.
A PDF-export update shows the division of work. Fern Writer reads an internal discussion about filtering exports by product and version, then proposes an update to the existing documentation page. Logan pulls the branch because the new material belongs inside an existing customization step rather than as a separate step. Claude Code helps perform that local revision and create a changelog entry.
Human review catches important details. The first changelog draft covers the entire PDF-export feature rather than the new filtering capability. The wording also confuses a dropdown with a different product feature. Logan supplies corrections and keeps iterating.
The account includes a larger authentication-documentation restructure where an engineer corrects the conceptual categories before approval. Logan explicitly says every page is reviewed and often extensively edited. The useful outcome is a repository draft and mechanical file work prepared for expert revision, not documentation that writes or validates its own technical truth.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: buildwithfern.com · Devin Logan
Frequently asked questions
Is Devin Logan the AI agent?
No. Devin Logan is the human technical writer; Fern Writer is built on Cognition's Devin.
What did the initial PDF change get wrong?
It placed filtering as a separate step and omitted the desired changelog; later wording also needed correction.
Who validates technical meaning?
Logan and the engineers review the documentation and correct its conceptual framing.
More details
Remaining steps
- Human review corrects the scope, feature names and technical wording.
About this account
Fern Writer prepared a documentation PR; Claude Code later reorganized the filtering instructions and drafted a corrected changelog through human-directed revisions.
Firsthand account opened and read. Outcome is source-reported, not independently reproduced. Suggested prompt written by Agents In Use. Mixed-agent account; Devin is the verified engine behind Fern Writer. Human authorship and technical review retained.
Credit: Devin Logan
Source published 10 Mar 2026
Source checked 15 Oct 2026
Tools and topics
Devin · Fern Writer · Slack · Claude Code
Devin · Fern Writer · Slack
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.