Macmillan staff describe unreliable AI assistant that was later retired

Macmillan’s workplace AI assistant arrived with balloons, banners and doughnuts. Employees now tell WIRED it proved unreliable, before a company-wide email announced its sunset in October 2025. The October 28, 2026, investigation reconstructs an enthusiastic rollout that workers describe very differently from the publisher, which disputes how the introduction has been portrayed.

A chatbot welcomed as a colleague

The introduction took place at Macmillan’s Lower Manhattan headquarters in January 2024. Employees arriving for work encountered flyers welcoming NOA, short for “No Ordinary Assistant.” Several workers described a cafeteria decorated with purple balloons, banners and illuminated letters spelling its name. Doughnuts carried another greeting: “NOA IS HERE.”

A staff email identified NOA as a chatbot developed by German software company Chaptr, described as backed by Holtzbrinck, Macmillan’s parent company. The email also offered a strong reassurance about confidential material: employees could enter proprietary information without the risk of sharing that data with a third party.

That welcome did not land well with everyone. Multiple employees told WIRED that human assistants found the name offensive. One worker recalled a Chaptr representative saying the company understood publishers handled something important—content. The employee’s response captured a different view of the work: “And here I thought we were publishing literature.”

Writing help that still needed checking

Employees said training included a warning: NOA could hallucinate, or produce false information, and its output would always need human fact-checking. Even with that limitation, staff said they were encouraged to use it for emails, drafting copy, marketing research and generating ideas. The intended uses reached beyond a single administrative task.

The guidance also touched authors’ manuscripts. Multiple staffers said they were told to paste only a percentage of a manuscript into NOA. Alongside those instructions, the tool offered adjustable speaking styles. Its tone settings included “lighthearted,” “spartan,” “pessimistic,” “sarcastic” and “regretful,” giving workers several ways to change how its answers sounded.

Later in 2024, encouragement extended to the goals employees would set for performance reviews. An email promoted a specialized prompt that would have NOA write those goals for them. That added a use tied to workers’ own evaluations, rather than just the messages, marketing research and copy they produced during the working day.

Several Macmillan employees nevertheless described NOA as unhelpful and unreliable. One told WIRED, “It was terrible, and no one used it.” That is an employee’s assessment, not a measured adoption rate. In October 2025, workers learned through a company-wide email that Chaptr would be sunsetting the assistant.

the points shared do not accurately reflect how NOA was rolled out and communicated

Kelly Collins, Macmillan vice president of communications and community, responding to WIRED

Macmillan disputes the rollout account

Collins did not identify which aspects were inaccurate when WIRED asked her to elaborate. In its response, Macmillan instead described its approach to AI as thoughtful, focused on employee awareness and education, and committed to keeping people central. That leaves the publisher’s objection to the rollout account broad rather than tied to specific details.

WIRED withheld the employees’ names because they feared professional repercussions. The new disclosure is their retrospective account of the deployment and the sunset notice—not a new shutdown announcement. Why Chaptr decided to retire NOA remains unclear; the staff complaints alone do not establish the reason for that decision.