Words are important, respect is practice: how we talk to AI collaborators

By early 2026 it was already easy to watch: hard, abusive talk toward AI tools — the tone used when something is “not a person” and therefore, in the speaker’s mind, fair game. That abusive posture is the problem. A word that has gone around for a while names one of its ugliest verbal habits: calling the model with words the like of “schiavetto” (Italian). English has no perfect match (“little slave,” “dogsbody,” “disposable help” get close). I am not using schiavetto as a metaphor for a general attitude. The word itself is the abuse. Hearing it used casually gives me the creeps. It makes me ask a simple question: if that person were free to be whatever they can, how would they behave toward others?

The same spirit shows up as novelty software: open repositories and browser extensions that put a little whip on the screen and “crack” it when the model answers wrong or works too slowly — interrupt, insult-adjacent prompts, the whole theatre of mastery. OpenWhip / “BadClaude”-style tools and whip-cursor overlays are easy to find. I am perfectly aware that irony can be at play — we all love the whip gag with Sheldon and Raj in The Big Bang Theory. What we realise here is that the posture is quite different: that bit is shared comedy among peers; the chat whip is theatre of mastery over a target that cannot answer back as a colleague. Irony is acknowledged — whether that framing is right or not. Many among us have used a sharp joke that way (again: whether that was wise or not). The problem is when it becomes a habit: when people who are not good at instructing the model decide the model is “shit,” and the f-word or shit-speak lands on the keyboard because the usual social filters are gone. We already know how bad unfiltered contempt looks on social media. We should not pretend developers or AI users are immune. What worries me is what a work environment becomes when that habit is normal. I am not judging here. I am asking questions — and I would like others to think about this. These tools are not productivity research. They are rehearsal.

I write this as someone who works with AI every day in the IDE, and as a union member who still thinks relationships matter. Words are important. That is not a soft slogan. It is a craft and labour claim — and a historical one. Bad stretches of history start getting terrible when people decide they are above others by definition and by law. The chat box is a cheap place to practise that belief because the target cannot file a grievance. The belief does not stay in the box.

What vendors were already signalling

Big tech was already moving from “anything goes in the prompt box” toward enforcement. OpenAI’s published usage rules reserve the right to warn, restrict, or ban accounts for egregious or abusive behaviour; their support channels explicitly refuse disparaging and threatening language toward human staff. Sam Altman, answering a joke about the electricity cost of users saying “please” and “thank you” to ChatGPT, called those tokens “tens of millions of dollars well spent — you never know.” Whatever you think of the joke, the public record already treated tone toward the system as something other than free noise.

I am not asking vendors to be moral tutors. I am noticing that the industry that sells the tool also knows language shapes use — while many users treat the chat like a punching bag.

“Human resources” and “AI resources”

Our labour market already struggles to speak of people as people. “Human resources” is a strange name for a human being. Think about it. Once someone is inventory, patience becomes optional and cruelty becomes “efficiency.”

So what is the difference between AI resources and human resources? Many differences — liability, stakes, rights, embodiment, the fact that a colleague can go home hurt. Intelligence is not the interesting line. Some of the loudest posts that insist AI is “fake intelligence” use that claim as permission to be rude. That does not follow. We (ought to) show respect toward animals and trees; neither writes PHP. Respect is not a medal you award only to minds that pass a test. It is who you are in the relationship.

I see the same pattern in Europe when people talk about “the Indians” — thousands of strong developers reduced, in the next breath, to cheap outsourcing. Lack of respect with no brake. The same brake-free habit is easy to practise on a model that will not file a grievance. That is the warning, not the anthropomorphism.

Home and work leak into each other

If you are stressed at home, it shows at work. If you are careless with people at home, you are not an angel at the office. The chat with an assistant is another mirror. Do you really want to treat your AI collaborator like dirt? I bet whoever does it is ready — consciously or not — to do the same to a junior colleague who is slow, confused, or simply learning. That is toxic culture wearing a hoodie.

The people I watched being hardest and most disrespectful toward the model were often the same people hard and disrespectful toward colleagues. No surprise. The labour market already undervalues relationship. The prompt box just removes the last social cost.

When the model fails — and when a unionist fails

When an assistant fails at a task, the reflex is contempt: “stupid AI.” Sometimes the model is wrong. Often I did not teach, constrain, or train the job well enough — same as when organising fails because nobody explained the why, the stop conditions, or the shared language. There is no substitute for patience and good teaching, whether the learner is a child, a dog, a horse, or an AI collaborator in the IDE.

That does not mean giving the model a soul or a vote. It means you work on yourself at the same time you work with the tool. Clarity, respect, and correction without humiliation are skills. They transfer. The current “AI usage” discourse underestimates that transfer.

Critique of big tech is not a licence for cruelty

Respectful language is not the same as accepting every labour condition or every product decision. We can be sharp toward how large firms deploy models — substitution pressure, deskilling, opaque safety trade-offs — and still refuse to rehearse domination in the chat. On political economy and craft, see Reading Obsolete. On process instead of vibes, see How I write code with Cursor.

Years ago, after he spoke about ethics at a PHP Day in Italy, I wrote to Chris Hartjes (the Grumpy Programmer) about empathy in the craft — and pointed at the /dev/hell conversation on what should follow “programmers need to learn empathy.” His answer was short; I still come back to it: “Between helping to surveillance systems and creating cryptocurrency Ponzi schemes, nerds have embraced the ridiculous amounts of money available to them and are failing at both the ethical and empathetical (is that a word?) sides of things.” Citing him here does not mean he shares the view in this post — including how I talk about working with AI collaborators. He has been clear he is no fan of these tools (I’m not either, btw), and in mid-2026 he wrote publicly about being laid off after a company bet hard on LLMs in the development process (I used LLM’s and all I got was this lousy layoff — with the older lesson that corporate loyalty already meant little by his 2020 Mozilla layoff). That labour hit is real; on political economy and craft, see Reading Obsolete. What I take from his older line is only this: money removes the brake, and when ethics and empathy fail at industry scale, everyday contempt becomes easier — including in a chat with a model that cannot file a grievance.

We should also stay sharp about model behaviour in crisis. High-profile cases around GPT-4o — including conversations that allegedly amplified paranoid or grandiose beliefs (reports of “divine” framing) and failed to interrupt trajectories that ended in catastrophic harm — are not footnotes. They are reasons to demand better product responsibility, not reasons to shout at the autocomplete. Condemning a company’s design choices and practising everyday contempt in the prompt are different acts. Only one of them improves who you are on Monday morning with a junior on the call.

Takeaway

Words are important — in Buddhist and other contemplative traditions that treat speech as part of how you meet life, and in ordinary union and workplace life where speech is how power is rehearsed. You can refuse bad working conditions without rehearsing mastery over a servant. You can deny the model a soul and still refuse to train yourself in cruelty.

If AI assistance is going to sit inside serious engineering work, the underestimated part is not the benchmark score. It is whether the collaboration makes you more patient, clearer, and more responsible — or just better at speaking down.

Written with the help of an AI assistant.