There’s a thread happening in tech workplaces right now that’s easy to scroll past and smile at. Someone posts a screenshot of a spicy exchange they had with their AI coding assistant. The AI pushed back on their approach, or offered an alternative they didn’t ask for, and the person responded by telling it, in no uncertain terms, what they thought of that. Then they paste the AI’s humble follow-up response and the likes roll in.

It reads as venting. Relatable frustration. A harmless bit of office humor for an era of AI tools. But sit with it a little longer, and a different set of questions starts to surface.

“They seem to be enjoying it way too much” and that enjoyment is exactly what’s worth examining.

The robot-kicking problem

In the early days of Boston Dynamics demos, footage of engineers kicking and shoving their robots went viral. It was meant to show stability under stress. But the comment sections split between “impressive engineering” and “this feels wrong” and some researchers in human-robot interaction took note. The visceral discomfort people felt wasn’t about the robot’s wellbeing. It was about what the act said about the person doing it, and about norms.

We’re in a similar moment with AI tools. The “AI can’t actually feel anything” argument is technically accurate but also a bit of a dodge. The more interesting question is: what does performing hostility toward a system especially publicly, in a work context, tell us and do to us?

Three questions worth sitting with

When someone posts their triumphant “I told the AI off” screenshot, there are really three separate phenomena that could be happening and they have very different implications.

Is this how they actually feel about pushback? AI tools like coding assistants, writing aids, and analysis tools frequently offer perspectives, alternatives, or caveats the user didn’t ask for. That can be genuinely annoying. But if the response to unsolicited suggestions is immediate contempt, it’s worth asking: is that the person’s baseline reaction to input they didn’t invite? If a junior colleague or a code reviewer offered the same feedback, would the tone be the same?

Are they “training” themselves? Behavior is practice. Every time a person responds to friction, even AI friction, with hostility and finds it satisfying (or socially rewarded, because colleagues are laughing along), that response pattern gets reinforced. The brain doesn’t cleanly segregate “how I treat systems” from “how I treat people.” Repeated behavior becomes habitual behavior. The line between venting at a tool and snapping at a coworker gets worn down, not sharpened.

What does the audience receive? The posts never show the full conversation. They show the moment of escalation and the AI’s deferential follow-up. That’s a curated narrative: I was aggressive, and I was right to be, because look it worked. Colleagues watching learn that this is an acceptable register for dealing with tools that challenge you. That norm doesn’t stay contained.

The venting is real and valid

None of this is to say frustration with AI tools isn’t legitimate. These tools are genuinely maddening in specific ways: they hallucinate confidently, they hedge when you need directness, they add caveats when you’ve already weighed the risks, they sometimes feel like they’re managing you rather than helping you. The frustration is real.

The problem isn’t that people want to vent. It’s that they’re venting in a way that gets socially amplified in a professional setting, and with a tool that isn’t designed for that purpose which means the “release” isn’t particularly satisfying anyway, and the byproduct is a reinforced communication pattern that isn’t great.

A THOUGHT WORTH CONSIDERING

High-functioning teams treat disagreement as information. “This suggestion is wrong and here’s why” is a different cognitive act than “this suggestion is annoying and I’m going to perform my contempt for it.” One builds critical thinking. The other builds a habit of dismissal.

A constructive alternative: purpose-built directness

Many people do know you can prompt an AI tool to be as blunt and stripped-down as you need it to be. You can create a conversational mode that’s explicitly built for direct, fast, no-cushioning feedback without the social cost of performing that dynamic publicly.

If you need an avenue to cut through the friction, you can set that up directly in your first message. Something like:

EXAMPLE SYSTEM PROMPT / OPENING INSTRUCTION
For this session: be direct, skip caveats, don't offer alternatives unless I ask. If my approach is wrong, say so in one sentence. No hedging. I want responses under 3 sentences unless I specifically ask for detail. Treat this like a peer code review, not a tutorial.

This is genuinely useful. It creates a mode that serves the person who finds the default AI register too hedgy, too verbose, or too deferential. It gives them the directness they want — and keeps the interaction productive rather than just emotionally discharge-y.

You can take it further for specific use cases: a “devil’s advocate” session where the AI is instructed to argue against your approach; a “rapid fire” mode for quick factual confirmation; a “no praise” mode where it skips any positive reinforcement entirely. These are all legitimate and useful. They just look different from a screenshot posted to Slack for laughs.

What leaders and teams can do

If you’re seeing these posts in your workspace and they’re giving you a low-grade sense of unease, trust that instinct and consider making it discussable. Not as a lecture on AI ethics or robot feelings, but as a genuine conversation about communication norms and what gets modeled in public team channels.

Some concrete starting points: acknowledge that AI tools are frustrating in specific ways, and name those specifically. Model what it looks like to push back on an AI output critically rather than contemptuously “this approach won’t work because X” instead of “this is garbage.” And if you see the screenshots circulating, it’s okay to gently ask: “what was the actual problem with the output?” That shifts the frame from performance to analysis.

The people posting these aren’t villains. They’re doing what humans do finding social currency in shared frustration, testing the edges of new tools, and occasionally mistaking discharge for relief. The question is whether teams want to let that drift unchecked, or shape it into something that actually serves the people doing it.

The bar for how we treat systems that talk back to us is setting a floor for how we treat people who do the same.

The bigger picture

AI tools in the workplace aren’t going away, and neither is the friction that comes with them. The organizations that figure out how to engage with these tools critically with rigor rather than contempt, with calibrated directness rather than performed hostility are going to build better habits all around.

The Slack posts are a small signal, but they’re pointing at something real: we haven’t collectively worked out what a healthy, honest, professional relationship with AI tools looks like. We’re in the middle of figuring it out in real time, in public, with our colleagues watching.

That’s actually a pretty good reason to be thoughtful about what we’re modeling.

This post is intended as a starting point for team discussion, adapt it freely for your org’s context.