Copilots or Replacements: AI Careers

Not long ago, AI was the helpful intern. It booked your meetings, tidied your grammar, and summarised your inbox without complaint. It was clever but not cocky—exactly like an overachieving assistant who never asked for a raise. But those days are gone. In the new-ChatGPT office, AI isn’t there to help. It’s there to act. And now, the tech world’s splitting in two: companies building AI to lend a hand and others building it to take your job altogether.

Meet Your New Colleague (or Successor)

There’s a silent crossroads on the AI road. On one side: tools—copilots, assistants, digital nannies. These are your friendly sidekicks, embedded in Excel or Gmail, nudging you through the workday like a caffeinated software. They still need you in the driving seat. On the other side: agents. These don’t assist, they act by themselves. They fill forms, schedule shipments, plan orders, and advise on investments—as already seen in tools from UiPath, DoNotPay, Klarna, and even Morgan Stanley’s AI wealth advisors.

This divide is shaping how tech firms make their money. Microsoft, Adobe, and Salesforce are backing the tool route: keep humans in the loop, just superaugmented. But over in Silicon Valley, outfits like OpenAI and Amazon are chasing a different dream: one where agents handle tasks from start to finish, and the software bills per job done, not per seat sold.

It’s a whole new game. Tools enhance jobs. Agents erase them. In the AI-agent economy, your to-do list becomes a system queue, and your job title turns into a deprecated feature. One day you’re optimising workflows, the next you’re watching them run without you. And HR won’t even need to schedule the exit interview.

The Bullshit Job Reckoning

Step back a moment and the picture gets even more curious. AI agents might finally deliver on a long-standing fantasy: the purge of so-called ‘bullshit jobs.’ Those mid-tier roles that keep the bureaucratic machine whirring—expense approvers, procurement schedulers, people whose job is half email and half waiting for a reply. Graeber would be delighted.

Number of US federal works AI would threaten, by sector.

But the implications go beyond a few vanishing spreadsheets. According to a recent McKinsey Global Institute report, automation could absorb up to 15% of global work hours by 2030 — potentially displacing some 400 million workers worldwide. In much of the West, whole economies are propped up by these roles. In countries like the UK and France, nearly a third of jobs sit in administrative or clerical categories exactly the kind AI-agent projects are targeting. Meanwhile, in the Global South or the East, countries like India, Japan, China, and the Philippines have built export economies on bullshit jobs they outsourced. All of them could soon disappear as well—under siege from tireless, accent-less agents who work 24/7 and never ask for paid overtime.

Share of work hours with potential AI automation, divided by country.

The North used to need the South’s labour. Now it’s eyeing a world where it doesn’t. Not a reshoring of factories, but of office work. High skilled positions will remain, mostly in the West, when the rest is probably going to fade. The future is about creativity and skills. Repetitive tasks are about to vanish. The OECD warns that over a quarter of jobs in member nations are highly automatable. The IMF notes AI could turbocharge inequality between rich and poor nations. The future might be more productive and more precarious for many countries. AI agents will take over your task if it has a repeatable pattern. And if it does, you're out of luck.

Two Futures, One Platform

So where does this leave us? With two futures staring down on us: copilots or replacements.

For now, AI agents are closely watched— and increasingly shaped—by three distinct stakeholder blocs. The corporate adopters like Amazon, Klarna, and Morgan Stanley are deploying agents to automate their workflows. The State actors, notably China, Singapore, and the UAE, are investing in agent-driven automation to modernise governance, digitise public services, and expand administrative reach without expanding payroll. And the soft power stakeholders—including VC firms like Andreessen Horowitz—are backing foundational agent startups (e.g. Adept, Imbue) to shape future economic dependencies and influence global cognitive infrastructures. These 3 majors block the future of AI development.

And which model wins may depend on who blinks first—tools or agents. The question is a major one. Indeed, why sell Word when you can just sell 'do my job' to ChatGPT? This future gets us to an interesting point: if AI agents democratise, we could end up in a world where we need less people to work and where more people, the entrepreneurial and creative ones, will rock; handling and creating several businesses from their sofa on their super-AI-powered laptop.

But then again there’s the centralisation elephant in the room. Copilots live inside your tools, running on your device. Agents live in the cloud, watching everything. The more powerful they get, the more they rely on massive models run by a few companies with a lot of GPUs. Decentralised AI agents remain a pipe dream for now, but quantum computing could bring them within reach. If quantum chips break free from the lab and into everyday machines, AI won’t need a trillion-dollar cloud to run. It could live anywhere.

Statement

AI is redrawing who works at all. Tools make you faster. Agents make you redundant. The divide is upon us, and the consequences are stark. One future empowers people. The other sidelines them. As companies choose between augmentation and automation, the rest of us should take note: the job market won’t collapse with a bang, but with a handover. If you’re not holding the reins, there’s a decent chance you’re already being replaced.

Weekly