Accidentally Left My AI Writing Assistant Set to "Passive-Aggressive Mode" During Performance Reviews

The Managerial Burden

As a newly promoted engineering manager, I found myself facing one of management's most dreaded responsibilities: annual performance reviews. Twelve detailed evaluations needed to be written, each requiring thoughtful analysis of the past year's accomplishments, constructive feedback on areas for improvement, and specific goals for the coming year.

After spending three hours crafting just one review—endlessly rewriting sentences to strike the perfect balance between honesty and encouragement—I remembered that our company had recently subscribed to WriteWise AI, an artificial intelligence writing assistant specifically designed for workplace documentation.

The AI Writing Assistant

WriteWise AI promised to "transform bullet points into comprehensive professional documents while maintaining your personal tone." Perfect for my needs! The interface was sleek and offered multiple "communication styles" to choose from:

• Supportive Coach
• Objective Evaluator
• Growth Mindset Mentor
• Direct Communicator
• Data-Driven Analyst

And at the very bottom of the dropdown, barely visible without scrolling: "Passive-Aggressive Colleague (BETA)."

Curious, I hovered over the information icon, which explained: "This experimental mode demonstrates how NOT to provide feedback. For training purposes only."

I chuckled and selected "Supportive Coach" instead. For my first review, I input my bullet points about Tom's performance:

• Consistently delivered projects on time
• Good team player, helps others when asked
• Technical skills are strong in current stack
• Could be more proactive in suggesting solutions
• Documentation sometimes lacks detail
• Should work on communication with non-technical teams

WriteWise transformed these points into a thoughtful, encouraging review that maintained honesty while emphasizing growth opportunities. Impressed, I proceeded to work through all twelve reviews, following the same process.

The Email That Revealed Everything

I submitted all twelve reviews to HR by the deadline, feeling pleased with my efficiency. The next morning, I received an alarming email from the HR director:

Confused, I pulled up the submitted documents. To my horror, I discovered that at some point during the process, the AI assistant's mode had switched to "Passive-Aggressive Colleague"—possibly when I'd clicked between browser tabs or accidentally hit the dropdown while typing.

Instead of supportive, constructive feedback, I had submitted brutally honest assessments wrapped in a thin veneer of professional courtesy—all delivered with the subtle knife-twisting art of passive-aggression.

The Passive-Aggressive Masterpieces

To understand the magnitude of the disaster, compare what I intended to write versus what was actually submitted:

For Tom, the Reliable Developer

What I thought I submitted: "Tom consistently delivers quality work on schedule and is a valuable team player who readily assists colleagues when asked. His technical capabilities in our current stack are strong. To continue his professional growth, Tom could focus on taking more initiative in proposing solutions rather than waiting for direction. Additionally, enhancing documentation detail and improving communication with non-technical stakeholders would increase his overall effectiveness."

What was actually submitted: "Tom manages to complete his assigned tasks without causing significant delays, which is apparently worth mentioning in today's professional climate. When directly asked—and only when explicitly asked—he will help teammates, though one wonders why this requires a request in the first place. His technical skills are adequate for his current responsibilities, though one might hope for more after three years in the same role.

If Tom could somehow find the initiative to suggest solutions before problems become urgent team discussions, that would be a refreshing change of pace. His documentation, when he remembers to complete it, often leaves readers on an exciting treasure hunt for actually useful information. Perhaps in the coming year, Tom might make the extraordinary effort to communicate with non-technical teams in a language resembling English rather than expecting everyone else to decode his technical jargon—but realistic goals are important."

For Sarah, the New UX Designer

What I thought I submitted: "Sarah has made an impressive impact since joining the team six months ago. Her design solutions show creativity and user-centricity, and she's quickly integrated into our workflow. To build on this strong start, Sarah would benefit from gaining a deeper understanding of our technical constraints and from providing more advance notice when design changes are required."

What was actually submitted: "Sarah has certainly made her presence known since arriving six months ago. Her designs are undeniably creative—sometimes impressively disconnected from technical reality, but creative nonetheless. The speed with which she's inserted herself into every workflow conversation is remarkable.

Moving forward, it might be revolutionary if Sarah developed even a passing interest in technical constraints before designing solutions that require rewriting our entire codebase. Similarly, while we all appreciate the adrenaline rush of last-minute design changes, some team members' cardiovascular health might benefit from receiving more than three hours' notice before major UI revisions. But perhaps that's just how things were done at her previous company, which she mentions approximately 12.7 times per meeting."

The Mathematical Improbability

The probability of this occurring was so low it could be expressed as:

Yet somehow, I had achieved the impossible—a perfect storm of technological mishap and human oversight.

The Damage Control Meeting

The meeting with HR was about as pleasant as one might expect. I explained the AI assistant situation, showing the settings and how the mode must have switched without my knowledge.

"So you're saying an AI wrote these... brutally honest assessments?" the HR director asked skeptically.

"Yes, but based on my bullet points," I admitted. "The AI just... amplified the negative aspects and added the passive-aggressive tone."

"The thing is," she said, leaning forward, "several employees have already seen their reviews—and the reactions are not what we expected."

The Unexpected Reactions

As it turned out, the passive-aggressive reviews had produced some surprising responses:

Tom's Response:

What I expected: Anger, demotivation, possibly a resignation letter.

What happened: Tom requested a meeting where he said, "That review was harsh but... not wrong. I've gotten comfortable. I'm already working on a proposal for improving our API documentation and signed up for a communication workshop."

Sarah's Feedback:

What I expected: Tears, HR complaints, team conflict.

What happened: Sarah laughed and replied, "Fair point about the technical constraints. My previous company had a larger engineering team—sorry for mentioning it so often. I've scheduled time with the backend devs to understand our limitations better."

The Team-Wide Impact

Word spread quickly about the "brutally honest" reviews. Rather than the morale disaster I feared, something unusual happened:

1. Increased Self-Awareness: Team members began openly discussing their blind spots and improvement areas
2. More Direct Communication: The usual corporate euphemisms decreased in day-to-day feedback
3. Proactive Improvement: Several team members initiated their own development plans without prompting
4. Unexpected Humor: "That's passive-aggressive AI worthy" became shorthand for "hard truth delivered unnecessarily harshly"

During our next team meeting, Michael, a usually quiet developer, even said: "Could we maybe have a channel for anonymous feedback that's this direct but without the snark? Sometimes it's easier to hear what you need to improve when it's not wrapped in so many layers of niceness that you can ignore it."

The HR Compromise

After much discussion, HR and I reached a compromise:

1. I would personally meet with each team member to discuss their actual performance assessment
2. The passive-aggressive reviews would be officially expunged from their records
3. However, employees could request to keep specific parts of the feedback if they found it valuable
4. I would take an additional management training course on "effective feedback delivery"

To my surprise, nine out of twelve team members opted to keep portions of their passive-aggressive reviews, claiming they were "more useful" than traditional corporate feedback.

The Tool Settings Adjustment

WriteWise AI contacted our company after we reported the incident. Their product team was both horrified and intrigued by what had happened:

"The Passive-Aggressive Colleague mode was never intended for actual use," their representative explained. "It was created for our internal development team to demonstrate feedback anti-patterns. We've now moved experimental modes behind a developer flag and added a confirmation dialog when switching between modes."

They also requested anonymized data about the team's reactions, noting: "We obviously don't recommend passive-aggressive communication, but the directness aspect clearly fulfilled an unmet need in performance feedback."

Lessons Learned About AI and Feedback

This disaster-turned-learning-opportunity taught me several valuable lessons about both AI tools and human communication:

1. Always verify AI output before submitting it, especially for sensitive communications
2. Check your settings regularly when using AI assistants across multiple documents
3. Sometimes conventional feedback is too soft to actually drive change
4. Different people respond to different feedback styles - some appreciate direct criticism more than gentle suggestions
5. Humor and unexpectedness can sometimes cut through the typical defensiveness that feedback engenders

"The most effective feedback occurs in the narrow space between what's comfortable to hear and what's necessary to know."

My New Feedback Approach

I've since developed a new approach to feedback that I call "Direct but Supportive" - aimed at capturing the clarity of those passive-aggressive reviews without the unhelpful snark:

1. State observations clearly without cushioning them in excessive positivity
2. Provide specific examples rather than generalizations
3. Connect improvement areas directly to impact on team and outcomes
4. Suggest concrete next steps for development
5. Deliver with genuine support, not artificial cheerleading

The Legacy of the Passive-Aggressive AI

A year later, this incident has become part of our team folklore. Performance has actually improved across multiple metrics, and we've developed a culture of more direct, honest feedback—though delivered with respect rather than passive-aggressive barbs.

As for WriteWise AI? They've since introduced a new mode called "Constructively Direct" that aims to capture the clarity of the passive-aggressive mode without the unhelpful attitude.

And whenever I open the assistant now, it defaults to "Supportive Coach" with a small notification that reads: "Are you ABSOLUTELY SURE this is the mode you want? We wouldn't want another 'incident,' would we?"

Even AI, it seems, can master passive-aggression when properly trained.