A recent study published in the Harvard Business Review reveals that artificial intelligence does not lessen employee workloads but rather intensifies them. Despite expectations that AI tools would reduce routine tasks such as drafting documents and summarizing information, research indicates that employees are experiencing increased demands, leading to higher workloads, burnout, and lower work quality. The rapid pace of work often results in cognitive fatigue and diminished decision-making abilities Many companies are currently facing challenges in encouraging widespread adoption of AI, as they hoped these tools would free up time for higher-value tasks.
A global cohort of over 100 artificial intelligence experts has published the second international AI safety report, highlighting significant uncertainty surrounding AI’s development and its associated risks. The report, released ahead of the AI Impact Summit in India, notes that while general-purpose AI capabilities have improved, predicting their behavior and potential dangers remains challenging. The report outlines various threats posed by AI, including impacts on employment, human autonomy, and environmental concerns. It emphasizes the need for more research to understand these risks and the effectiveness of mitigation strategies. The findings suggest that without proactive measures, AI could exacerbate economic inequalities, shifting earnings from labor to capital owners, particularly benefiting high-income countries with skilled workforces.
A recent study by researchers from the Oxford Internet Institute and the Nuffield Department of Primary Care Health Sciences reveals that AI chatbots may pose risks to patients by providing inadequate medical advice. Those who utilized large language models for medical decision-making performed no better than those relying on personal knowledge or internet searches. The study found that while these models, such as GPT-4, are proficient at structured tasks, they often fail in interactive scenarios, resulting in potentially dangerous advice. The authors conclude that current AI chatbots are not equipped for real-world medical decision-making, highlighting the need for further advancements before their deployment as public medical assistants.
Why do we care?
We built output accelerators and forgot the control systems.
The HBR workload findings and the Oxford medical results are the same failure mode in two environments. AI lowers the cost of producing an answer—drafts, summaries, recommendations—so organizations raise expectations. Employees don’t get time back; they get higher throughput targets plus a new tax: verification, rework, and exception handling when the model is subtly wrong. That’s how you get burnout and declining quality even while “productivity” metrics look better on paper.
Now add the international AI safety report’s point: we can’t reliably predict behavior at scale. That’s not a gap we close next quarter—it’s a structural constraint. If the system is powerful but not fully predictable, you can’t govern it the way most businesses govern software.
And the distributional warning matters for clients: the upside concentrates where capital, skills, and platform ownership already are. The downside—displacement and wage pressure—lands locally. That turns AI from a tool decision into an operating-model decision.
For MSPs, the commercial risk is simple: if you sell AI as “do more with less,” your client may associate your rollout with higher workload, lower quality, and attrition when renewals come due. In regulated environments, that turns into liability. The winning offer isn’t “deploy ChatGPT.” It’s the governance layer: workload caps, quality gates, escalation rules, and a measurement plan that proves AI is improving outcomes instead of just increasing volume.