Devin renames Neon docs with human cleanup
Neon reports Devin finding old feature terminology across its documentation, but replacement errors required more than twenty manual commits to finish the rename.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Find references to this retired feature name across the docs and propose context-aware edits. Preserve meaning, flag ambiguous cases and present the diff for editorial review before merging.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Daniel Price and Barry Grenon
- Original account
- Read the source ↗
- Published
- 23 Jun 2025
- Source link checked
- 15 Oct 2026
- Reported result
- Neon reports Devin finding old feature terminology across its documentation, but replacement errors required more than twenty manual commits to finish the rename.
- Human involvement
- The docs team reviews generated PRs and manually corrects context and terminology. The source reports more than twenty cleanup commits.
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
- The docs team asks Devin to find old feature names and rename them Instant Restore.
- Devin locates nearly all relevant references across the documentation set.
- The team inspects replacement errors in the generated changes.
Neon's documentation team tried Devin on recurring changelog preparation, issue-driven updates and larger edits. A concrete larger task renamed references to history retention and point-in-time recovery to Instant Restore across its documentation set.
The team gave the agent the terminology change and let it search the repository. Devin found nearly every relevant location, reducing the manual discovery work. Its edits, however, relied too heavily on replacement and created awkward constructions such as a duplicated word in the new name. Price and Grenon report that more than twenty manual commits were needed to polish the result.
This account separates locating affected text from making a contextually correct revision. The discovered references were useful even though the final prose still required substantial human attention. It would be inaccurate to describe the rename as autonomously completed documentation maintenance.
The authors also report other bounded successes, such as starting weekly changelog files from a playbook and finding an appropriate existing screenshot for a small feature document. More ambiguous work could fail silently. Their observed pattern was to delegate clearly scoped repository work and inspect the resulting pull request, with technical writing judgment and final acceptance remaining in the team.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: neon.com · Daniel Price and Barry Grenon
Frequently asked questions
What did the agent find successfully?
Nearly every location using the old feature terminology.
What required human correction?
Direct replacement produced awkward wording, including a repeated word in the new name.
Was the rename completed unattended?
No. The team reports more than twenty manual commits to polish the changes.
More details
Remaining steps
- More than twenty manual commits finish and polish the rename.
About this account
Devin located nearly all old product terms and made replacement edits across documentation, including incorrect repeated wording.
Firsthand account opened and read. Outcome is source-reported, not independently reproduced. Suggested prompt written by Agents In Use. Source reports both useful discovery and extensive manual correction. No claim of autonomous final acceptance.
Credit: Daniel Price and Barry Grenon
Source published 23 Jun 2025
Source checked 15 Oct 2026
Tools and topics
Devin · GitHub · Documentation repo
Devin · GitHub · Documentation repo
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.