Two fresh sources published this week offer the clearest picture yet of how generative AI is reshaping work. A Microsoft-backed study of 200,000 Copilot conversations maps exactly which tasks people delegate to AI. The researchers convert these patterns into an “AI applicability score” for every US occupation. Information gathering and writing top the list. Sales, admin and programming roles show the highest overlap with what AI can do.
Meanwhile, a Gizmodo investigation exposes the hidden workforce that keeps these models running. The piece documents the annotation, logistics and moderation work performed by low-paid contractors. One interviewee calls the sector “a new era in forced labour”.
The academic study emphasises augmentation over replacement. Most user requests still treat Copilot as an assistant, not a substitute. But the Gizmodo report quotes executives planning cuts of up to 40 percent. Their reasoning? “AI doesn’t go on strike.” This reveals a disconnect between research caution and boardroom plans.
One finding from the Copilot data challenges conventional wisdom. Wages and education levels show weak correlation with AI exposure. High earners can’t buy protection through credentials. This suggests we need universal retraining budgets, not narrow coding bootcamps.
Labour advocates are pushing for recognition of data annotation as formal employment. They want minimum standards and proper contracts. This would close the loophole that keeps AI’s human infrastructure invisible on corporate reports.
But hard questions remain. Will investors stomach slower rollouts for better labour protections? Can regulators even find, let alone monitor, the sprawling data-labelling networks spanning continents? US jobs data already shows accelerating cuts, with analysts drawing direct lines to AI deployment. Are we watching the first tremors before the earthquake?
Takeaways: The latest studies confirm that AI touches every occupation, with full automation clustering in information and communication roles. Corporate boards are converting early productivity gains into headcount cuts at breakneck speed, creating a widening gap between C-suite promises and workplace reality. Recognising annotation work as formal employment would drag AI’s hidden human infrastructure into the light, forcing companies to account for these workers on their books. The next few quarters will reveal whether the deployment rush can be slowed by any force – regulatory pressure, investor concerns, or worker organisation. The race between AI capability and labour adaptation has begun.