Vellum Pulls Its AI Automation Feature a Day After Launch

Vellum announced the feature on its official blog with a careful opening: “we have not built AI into Vellum.” A day later the same blog carried a second post, “We’ve Removed Vellum Automation”, in which the three-person team wrote that it had read the emails, the Threads posts and “a plethora of 🤮 emojis”, and that it had underestimated how strongly authors would object to Vellum working with AI assistants at all. Indie Author Magazine reported the reversal on 6 November.

The episode is a small case study for everyone who builds software for writers, from formatting tools to reading apps. Vellum shipped a sensitive feature to its entire customer base at once and learned the reaction in public. Teams that launch a minimum viable product to a small opt-in group first usually hear that kind of objection while it is still cheap to change course.

For authors the story is less about one app than about a question every writing tool will now face: where does help with the business of publishing end and AI in the creative work begin?

What did Vellum Automation do?

Automation used a local connection, built on the Model Context Protocol, the same plumbing we described in Apple’s AI assistant for Pages, so that an assistant on the author's own computer could send instructions to Vellum. According to Vellum’s help pages, it worked with supported desktop assistant apps and required a Vellum Press licence.

The examples Vellum gave were production chores, not prose:

• check that every image in a book has an accessibility description;
• generate new Kindle epub files for all books in a folder;
• move the About the Author page into a shared template and reuse it across a series;
• convert old italic text messages to Vellum’s Text Conversation style and rebuild the files.

Vinci Books, a publisher managing thousands of titles, tested it early and called it “revolutionary for our production department.”

Why was it removed so fast?

Many authors use Vellum precisely because it feels like a craftsman's tool, and the word “AI” on its release notes read to them as a change of side. Vellum’s removal post says the feature was built for writers with dozens or hundreds of books who asked for help with repetitive tasks, “and definitely not to encourage AI-generated writing.” That distinction did not survive first contact with the announcement.

The mood in publishing explains part of it. Authors are watching the Anthropic settlement payouts, the AI disclosure rows after the Goncourt and a steady flow of scraped-book scandals. In that climate, any tool that opens a door to an AI assistant is judged on what the door could let in, not on what it was built for.

What does it mean for authors?

Nothing changes in your files. If you updated to 4.2, update again to 4.2.1; your books, styles and the new word count history stay as they are. If you were one of the authors who liked the idea, the practical loss is time: series-wide fixes go back to being done book by book.

The larger signal is that writers now hold real leverage over their tools. A vocal user base reversed a product decision in about 24 hours. Expect other vendors to label AI features more precisely, keep them strictly opt-in and test them quietly before they appear in a changelog.

What should you do now?

1. Update to Vellum 4.2.1 through Check for Updates if you installed 4.2.
2. Write down your own rule for where AI may touch your work, if anywhere. Our guide to AI tools for fiction writers separates admin uses from creative ones.
3. Read release notes before updating production tools in the middle of a launch.
4. Tell your tool makers what you want. This week proved they listen.

Massimo Marino Author Editorial Team

Independent reader guides, science notes and writing craft.