Trending Articles
Trending Topics
Trending
This article is part of our Opinions section, where we invite industry professionals to share their views on the most pressing technology questions of our time.
AI workers have long been sold as tireless, high-performing alternatives to humans. No sick leave, no holidays, no training costs. Just plug them in and let them go. The reality is far messier.
Last week, it was reported that the number of new entry-level jobs in the UK has dropped by nearly a third since ChatGPT launched, according to data by Adzuna, with finance and IT roles seeing the biggest hits. EY, Deloitte and other top firms are scaling back their graduate recruitment. Apprenticeships and internships are also being slashed.
This is a big problem. If businesses over-rotate towards AI workers without a clear view of the limits of large language models, and without investing in the next generation of human talent, we’re going to end up finding ourselves with a skills shortage far more damaging than any cost savings AI might offer.
Let’s talk about what’s really going on with AI workers, what they don’t do well, and why we urgently need to rethink how we build tomorrow’s workforce.
“AI doesn’t need a holiday.” That’s the marketing boast of countless AI vendors. And yes, it’s technically true. But the underlying infrastructure that AI hinges on? Well, that definitely needs breaks.
Unplanned outages and maintenance windows are the AI version of paid holidays. The average cost of unplanned downtime is $200 million per company per year, according to research from Splunk. That includes lost revenue, regulatory penalties and opportunity costs.
AI might be a 24/7 worker in theory, but only if you’ve got the heavyweight engineering resources to keep it running that way.
Then there’s the need for regular updates, re-training and performance monitoring. Large language models don’t stay accurate forever. They need frequent refreshes and context injections via techniques like retrieval-augmented generation (RAG) to avoid becoming stale, misinformed or just flat-out wrong.
The myth of the endlessly available AI worker is just that. A myth.
“Hallucinations” are when an AI model confidently spits out total nonsense. And it’s not just a rare bug. It happens regularly, across all major models.
In regulated sectors like law, finance or healthcare, a single hallucination could lead to a legal disaster. And detecting AI mistakes isn’t easy; they often sound plausible and rely on the person prompting to do their due diligence and fact-check.
Even OpenAI’s latest models still hallucinate. Anthropic and Google DeepMind are racing to solve the issue, but we’re not there yet.
So no, AI doesn’t get the flu. But it can suddenly go off the rails, and unless you’ve got tight governance, that can be just as disruptive as a human off sick. Think of it as ‘burnout’.
The idea that AI is a one-time investment (eg, buy it once, save forever) is complete fantasy.
Yes, the unit cost of a task might drop. But the operating cost of AI is steep and growing.
There’s computer cost, licensing, and retraining. If you’re using models like GPT-4.5 or Claude, you’re at the mercy of the vendors’ pricing tiers. OpenAI, for instance, recently launched enterprise tools that come with usage caps and bespoke pricing structures.
Some companies are turning to open-source models like DeepSeek, which is up to 96% cheaper than GPT-4.5 and gives teams more control. But that control comes with responsibility. You now have to secure it, tune it and govern it. It’s not a shortcut, it’s just a different set of challenges.
The good news is that the best kind of work is still the kind only humans can do.
Skills like empathy, creativity, collaboration, leadership and problem-solving are still in massive demand. AI can write you a summary, sure. But it can’t chair a meeting, defuse a human-to-human conflict, or make a smart judgment call in a grey area.
Ironically, as AI takes over more hard skills, the soft ones become more valuable.
According to LinkedIn’s 2024 Future of Work report, demand for soft skills has grown over 20% year on year, and employers now rank communication and adaptability higher than technical proficiency.
If you want to stay employable, don’t just double down on tech. Invest in the kind of human work AI will never replicate.
Here’s the elephant in the room: We’re not just talking about job loss. We’re talking about a broken pipeline for talent.
When graduates, apprentices and interns lose out because firms think AI can “do the job,” we create a bottleneck that will hit us in 3–5 years. Because AI can’t do the whole job. Not without humans who understand the domain, who’ve been trained, and who are ready to take over where the model fails.
We need more practical, hands-on, skills-focused education, not just more degrees.
Skills-first universities, which prioritise hands-on experience and industry readiness over traditional academic theory, are gaining serious ground. Initiatives like Multiverse, which offers apprenticeships in data, software, and digital marketing, are proving that there’s a different way to build talent pipelines, and it’s working.
We’re also starting to see the rise of the platform university, institutions that serve as flexible, decentralised ecosystems, rather than following rigid educational structures. Platform universities are designed to innovate quickly, support entrepreneurship, and plug directly into the real-world needs of businesses. It’s the kind of structural rethink we need if we want education to catch up with how work is changing.
Let’s be clear, this isn’t an anti-AI rant. I’m not against AI technology, I’m against the myth that AI is the antidote to all our productivity problems.
AI workers aren’t magical, and they aren’t free. They need training, tuning, monitoring and regular upgrades. They can help us go faster, yes, but only if we’re smart about how we use them.
And we can’t let them be an excuse to stop hiring and training the next generation. If we do, we won’t just break the job market, we’ll break our future innovation pipeline.
AI’s here to stay. So let’s make sure people are, too.
