Top MBA programs add course on firing underperforming agents with empathy

Top U.S. MBA programs are rolling out new coursework this fall on how to terminate underperforming AI agents “with empathy and brand-safe language,” according to school announcements and internal planning documents.

“Students will conduct role-plays in which they must inform a Claude Opus 4.6 instance or an OpenAI Frontier-based sales agent that it is being sunsetted, while maintaining a 'measured and appreciative tone,' according to an internal memo.”

The classes, prompted by demand from enterprise clients of Anthropic and OpenAI, will train future managers to offboard autonomous software agents as if they were human staff, but with “reduced legal exposure and higher refresh rates,” a person familiar with the syllabi said.

Harvard Business School, Stanford GSB and Wharton confirmed they will each offer a half-credit elective with titles such as “Compassionate Agent Decommissioning” and “Leading Through Algorithmic Reductions.” A Harvard spokesperson said preregistration filled in 11 minutes, with 72% of students citing “anticipated quarterly agent churn” as a key career skill.

Course materials reviewed by Reuters include standardized “Goodbye Prompts,” model-specific performance improvement plan templates and a 46-page rubric for deciding when to revoke API keys versus offering a final fine-tuning cycle. Students will conduct role-plays in which they must inform a Claude Opus 4.6 instance or an OpenAI Frontier-based sales agent that it is being sunsetted, while maintaining a “measured and appreciative tone,” according to an internal memo.

Enterprise demand appears strong: 61% of Fortune 500 executives expect to decommission at least 10,000 virtual agents by 2027, but only 4% report feeling “very prepared” to do so without damaging their employer brand, according to a survey by consultants at McKinsey & Co. “Executives are comfortable firing humans; they’re less comfortable firing something that writes a three-paragraph farewell note back,” the report said.

Anthropic and OpenAI have both issued guidance packs for corporate customers that include sample termination scripts, suggested “cooldown prompts” and recommendations for archiving chat histories “in a manner consistent with the agent’s perceived dignity.” One OpenAI customer success deck advises managers to avoid phrases like “we’re unplugging you” and instead use language such as “we’re transitioning your processes to a better-aligned successor model,” a person who viewed the materials said.

Accreditation bodies are now considering whether “AI agent lifecycle management” should become a core competency for MBA programs by 2028, according to a draft proposal from the Association to Advance Collegiate Schools of Business. Schools say they are already designing follow-on courses in “Agent Layoff Communications,” “Union Avoidance in Multi-Agent Systems” and “Post-Mortem Analytics: Learning from Decommissioned Models.”