Macmillan discontinues AI assistant NOA after employee backlash

What NOA was meant to do

NOA, short for No Ordinary Assistant, was an AI chatbot developed by Chaptr, a German software company backed by Holtzbrinck, Macmillan’s parent company. It was introduced as a secure tool for employees to generate emails, draft marketing copy, conduct research, and brainstorm ideas without sharing proprietary data with third parties.

During training sessions, Macmillan staff were warned that NOA could "hallucinate" and required human fact-checking. Employees could adjust its tone—options included "lighthearted," "spartan," "pessimistic," "sarcastic," and "regretful"—and were encouraged to use it for tasks like composing performance review goals.

Why employees pushed back

Macmillan employees described NOA as unreliable and, in some cases, "terrible." Some found its outputs unusable, including evaluations of literary agent submissions generated by Google Gemini, an alternative tool Macmillan later explored.

The name NOA—an acronym for No Ordinary Assistant—was reportedly seen as offensive by human assistants at the company. Staff were also instructed to limit the amount of authors’ manuscripts pasted into the tool, raising concerns about data handling.

Broader industry tensions

Macmillan and other major publishers, including Penguin Random House, continue to explore AI tools like Google Gemini and ChatGPT Enterprise despite ongoing copyright infringement lawsuits against AI companies. Penguin Random House’s German division, for example, is suing OpenAI over alleged violations.

Authors have grown increasingly frustrated with publishers’ AI adoption. Some bestselling writers, who collectively have sold over 30 million copies, report receiving AI-generated marketing copy with fabricated quotes. Others say they could not secure promises from Big Five publishers to avoid using AI in their contracts.

The friction has led some authors to consider self-publishing, questioning the value of traditional publishers if AI handles tasks like editing, marketing, and cover design. One author told WIRED, "What are authors even doing with traditional publishers anymore if they’re going to outsource everything to a machine?"

What’s next for Macmillan

After discontinuing NOA, Macmillan has shifted focus to integrating Google Gemini into its workflows. However, employees who experimented with the tool found its evaluations of submission letters "useless," highlighting ongoing challenges in replacing human judgment with AI.

The company’s experience mirrors broader industry struggles to balance efficiency gains with trust and quality. While some publishers, like Hay House, frame AI as a tool for "creativity" and "innovation," others, such as Penguin Random House, have explicitly barred its use for tasks like publicity copy and cover design.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

Macmillan’s decision to discontinue NOA is a rare example of a publisher pulling back from an AI tool due to employee and author resistance. For founders and operators, this story underscores three lessons:

1. AI tools must prove their worth—quickly. Macmillan’s staff rejected NOA not just on principle but because it failed to deliver reliable results. In creative industries like publishing, where nuance matters, AI-generated outputs can backfire if they require constant human oversight to correct.

2. Trust is fragile. The backlash wasn’t just about NOA’s performance. Authors and employees saw AI as a threat to their roles, and publishers’ opacity about its use deepened the distrust. For startups and incumbents alike, clear communication about how and why AI is being deployed—especially in client-facing or creative work—is critical to maintaining relationships.

3. Copyright and compliance risks are real. Publishers are caught between experimenting with AI and facing lawsuits over its use. For operators in regulated or content-heavy industries, the legal landscape around AI is still uncertain, and the cost of getting it wrong could include reputational damage, not just financial penalties.

Related on Lazyfounder

• More ai news