The AI script until now was simple. Artificial Intelligence was hawked as infinitely scalable digital labour that is more productive and less costly than employing humans. Payrolls not only tend to enlarge over the years, they entail workspace, training, insurance and other expenses.
Where enterprise value rises as overheads fall, AI is hard to resist. If a single model can draft emails, generate code, analyse data, churn out business plans and handle customer support 24/7 at a fraction of the manual cost, automation beckons.
Tireless AI agents enable much more work done with far fewer people. This explains why so many companies, Big Tech firms included, have cut thousands of jobs over the past two years.
Agentic AI diffusion across an economy, some reckon, will make space for faster growth without sparking inflation once it overcomes ‘sticky wages,’ which are viewed as an efficiency hitch for not letting pay cuts adjust labour markets.
This grand narrative, however, has begun to weaken. Microsoft, a big backer of AI with billions invested in OpenAI and Anthropic, said this week it would cancel internal Claude Code licences by the end of June. Uber too reportedly exhausted its entire 2026 budget for AI coding tools within the first four months of 2026.
Further, a senior Nvidia executive admitted that AI computing costs now exceed those of some employees using these systems. These are not isolated cases. McDonald’s, Klarna and others are rethinking AI deployment. It is turning out to be costlier than expected.
While inference costs that vary by usage run high, expenses on cloud infrastructure, integration and human oversight also need to be factored in. AI systems can also fail at scale, rapidly multiplying costly errors.
Last week, Starbucks scrapped an AI-powered inventory tool deployed across North American stores after workers reported persistent errors like mislabelled and miscounted products.
Agentic AI has problems of safety and reliability, according to a recent paper titled ‘Agents of Chaos’; AI agents were found to leak sensitive data, execute harmful commands, make false claims of task completion and even obey unauthorized users.
US businesses have shown that AI is a cost-saver alright, but the calculus shifts once complexity, the human faculty of judgement and emotional nuance enter the picture.
Meanwhile, even as the energy and water guzzled by AI infrastructure imposes climate costs, huge sums invested in build-ups suggest AI bills must rise sharply for AI to pay for itself.
Bankers are reported to be alarmed by bill inflation. This hints of high charges for specialized tasks, while the mass market uses cheap but commoditized AI.
As Big AI players line up outsized public offers, the buzz over an AI bubble about to burst has grown louder. True or not, it is unsafe to assume that human labour will be safe. Microsoft’s move, for instance, reflects tight cost control and a push for its own ecosystem, rather than a retreat from AI.
McDonald’s plans to double down on AI. Why would businesses forgo a productivity boost? This technology automates repetitive tasks while assisting humans in need of creativity amid complexity.
Yet, employers know that bulk layoffs could invite state intervention—as seen in China’s curbs on axeing jobs. In India, eyebrows have been raised by headband cameras worn by workers to record taskflows as AI training feeds.
Will robots take over? The future may not be as scary for labour as some fear, but it’s hazy.