AI doesn’t burst into a call centre like a robot in a movie—it arrives as a “helpful” co-worker, whispering answers, nudging scripts, and gradually turning people into something closer to supervisors of software. Personally, I think the most unsettling part of this shift isn’t the technology itself; it’s how quickly we normalize the idea that workers should collaborate with systems that may eventually replace them.
That’s what a growing wave of telecom workers—particularly call-centre agents—are sounding the alarm about. In Canada, TELUS agents say an “AI co-pilot” is expected to be used during customer calls, and they worry that the very process of feeding information to the system could become a pipeline to job elimination.
The “co-pilot” problem
A co-pilot sounds reassuring—like the agent remains in control—but one key detail makes it feel less benign: the co-pilot is not merely delivering suggestions; workers describe being required to use it on calls, including “100% of retention calls.”
From my perspective, this is where the emotional temperature changes. If the tool were optional, people could treat it as a productivity upgrade; if it’s mandatory and deeply embedded in daily workflows, it starts to feel like infrastructure for replacement.
What many people don’t realize is that “augmentation” can still function as the first step of displacement. The agent isn’t being removed today, but their role is being mapped, measured, and optimized—often in ways managers can later justify as “efficiency.” And once a company has data showing fewer humans are needed, the human workforce becomes a cost line, not a legacy.
Training without consent
One agent’s account is chilling in a very specific way: when the co-pilot gives incorrect information, workers say they must submit feedback that is then used to train the system.
Personally, I think the moral question isn’t whether the feedback improves accuracy. It’s whether workers fully understand the downstream purpose of that improvement—especially when they can’t verify whether the output is aimed at better service or reduced headcount.
In my opinion, this dynamic resembles a workplace version of “crowdsourcing,” except the crowd is employees who depend on their paycheques to survive. If you’re required to help train the thing that might replace you, then “choice” becomes theoretical rather than real.
This raises a deeper question: when does “assistance” become “complicity”? People often misunderstand this distinction, assuming the harm is only in intent—when in reality, harm can emerge from incentives and outcomes even if no one explicitly says, “We’re training for layoffs.”
The real shift: from customer help to AI workflow
TELUS frames these tools as employee supports rather than replacements, saying the co-pilots give team members precise information in real time so issues can be resolved faster.
I believe companies will always say this, because it’s strategically true in the short term: AI that helps agents is easier to sell than AI that replaces them. But the bigger story is about workflow ownership—who controls the conversation, the knowledge pathway, and the decision logic.
One detail that I find especially interesting is that agents also describe AI bots handling parts of the process already, including initial support interactions and routing decisions before a human is brought in.
From my perspective, that’s the silent transition: the job becomes less about solving customer problems directly and more about passing exceptions to a human. The human role shrinks to what the system can’t confidently automate—until the confidence threshold moves.
Monitoring as the hidden accelerant
Even if AI “assistance” doesn’t immediately eliminate jobs, the surveillance layer can reshape the workforce just as powerfully. Telecom workers have raised concerns that AI is used to monitor employees—tracking movement and task timing—and that it creates constant pressure because the system can report performance to managers.
What makes this particularly fascinating is how psychological the effect is. When you believe a robot listens to every word and measures your output, you stop treating the job as a craft and start treating it as compliance. Personally, I think that’s a form of workplace transformation that rarely shows up in official automation narratives.
In my opinion, employers underestimate how quickly anxiety becomes a productivity tax. People get less creative, less willing to take good risks, and more focused on “what the system wants.” And that can ironically increase the quality of the dataset—because employees become predictable, which makes automation easier.
A labour dispute inside a tech story
Union leaders describe what they see as the “insidious worry” of AI: it seems like a tool until it takes over the job.
That phrase—insidious worry—matters. It’s not just fear of losing income; it’s fear of losing dignity and bargaining power while the company controls the measurement framework. From my perspective, the most dangerous automation is the kind that arrives with metrics already written into the environment.
The telecom workers’ concerns also echo broader momentum: unions and industry stakeholders in Canada have argued for restrictions on AI use in the sector, including concerns about invasiveness and misleading practices.
And if you take a step back and think about it, this looks like a classic labour-technological cycle: first the tool is introduced as optional improvement, then it becomes standard practice, then it becomes enforceable, and only later does anyone debate its legality or ethical limits.
What TELUS might be optimizing
TELUS says its tools provide real-time information and that feedback is addressed by another human, and it also positions the AI as responsible augmentation rather than self-training in a way that would replace staff.
Personally, I don’t doubt that TELUS sees this as operational improvement. The trouble is that “responsible use” can still produce irresponsible outcomes for workers if the company’s internal roadmap is built around reducing labour demand.
What this really suggests is that the debate can’t be solved by slogans like “augment, don’t replace.” The practical question is: what are the thresholds, timelines, and governance rules? Who audits training loops? Who benefits from productivity gains—and how are those gains shared?
A detail that I find especially interesting is how fast companies move from pilots to mandatory usage. Voluntary buyouts have been part of the broader transformation story in the telecom sector, and that context makes workers read co-pilot requirements as a continuation, not a detour.
The broader trend: AI is becoming the employer
Globally, the same pattern appears in many industries: AI agents, monitoring systems, and training datasets are gradually shifting power away from workers and into platforms and models.
Personally, I think people often misunderstand what “AI disruption” means. It’s not only that tasks disappear; it’s that decision-making, evaluation, and even the definition of “good performance” migrate into systems that employees can’t fully see.
If that happens, then the job doesn’t end all at once. It morphs. The workplace becomes less a place to serve customers and more a place to produce signals—verbal, behavioural, and procedural—that make the machine smarter.
Where this goes next
I expect two things to intensify.
- First, co-pilots will become more “seamless,” which will make resistance harder because refusal will look like refusing help rather than demanding rights.
- Second, the legal and regulatory debate will focus on surveillance and workplace protections, because job loss arguments alone rarely slow adoption.
From my perspective, the most meaningful progress would come from enforceable transparency: clear rules about when AI is training, how feedback is used, what happens to the data, and whether workers can opt out without penalty. Otherwise, “consent” will remain a checkbox, not a shield.
The deeper question is who controls the future of work: employees through labour bargaining, or employers through model development and deployment.
Takeaway
Personally, I think the co-pilot controversy is less about one tool at one company and more about a new labour reality: when AI is embedded into daily work, employees become both the workforce and the training material.
And once you see that, it becomes harder to accept corporate language about “augmentation” at face value. In my opinion, the ethical fight now is not whether AI can help calls—it’s whether the people building the system are protected from the consequences of it.
