Two very different pictures of AI and jobs arrived this week, and reading them side by side is more revealing than either alone. In the United States, fresh earnings reports show Big Tech headcount drifting almost sideways: Alphabet actually grew, adding nearly 12,000 people to reach 198,933, while Meta shed more than 2,000 to 75,472 and Microsoft ended June with 223,000, some 5,000 fewer than a year earlier. In Johannesburg, meanwhile, the story is not about a rounding error in a giant's payroll. It is about the bottom of the ladder quietly being sawn off.
Across South Africa's call-centre and business-process outsourcing hubs, generative AI now handles the routine work that used to fill entire shifts: live-chat triage, password resets, frequently asked questions. Analysts stress this is not yet wholesale replacement; the complex, emotional and compliance-heavy calls still need people. But the volume of simple work is shrinking, and that simple work is exactly what entry-level jobs are made of. PwC's 2026 barometer found the share of local postings demanding AI skills nearly doubling to 3 percent even as broader conditions weakened.
Why this matters more in South Africa than in Seattle comes down to what those jobs do. In a country with one of the world's highest unemployment rates, a call-centre headset has long been the first rung on the formal-economy ladder for young people. McKinsey estimates automation could displace 3.3 million South African jobs by 2030 while creating 1.8 million, with more than half a million losses in manufacturing alone. Old Mutual Investments warns of the deeper danger: if you remove the first rung, social mobility stalls and wealth concentrates, especially when the AI systems doing the displacing are owned by a handful of foreign firms.
That ownership point is the thread linking the two pictures. The American giants trimming or holding headcount are the same companies selling the tools that automate a Johannesburg call centre. Alphabet's hiring and Meta's layoffs are both, in their way, bets on AI: Meta explicitly moved some 7,000 staff onto AI initiatives while cutting elsewhere, and is spending up to $145 billion this year. The productivity those investments promise lands as a flat payroll in California and a vanishing first job in Gauteng.
It would be easy to overstate this, and the honest data resists a clean panic. A Yale Budget Lab assessment found no broad, measurable labour-market disruption from AI globally since ChatGPT's release, and the World Economic Forum still projects far more jobs created than destroyed by 2030, 170 million against 92 million. The aggregate may well hold up. But aggregates hide distribution, and the distribution is the whole story: pain concentrated at the entry level, in poorer regions, among younger and lower-skilled workers, can be invisible in a global average and devastating in a single labour market.
The Brookings Institution's advice to African governments captures the tension neatly: sequence AI adoption rather than rush it, so the first rung is not removed before another is built. The same caution applies everywhere. The question raised by this week's numbers is not whether AI destroys more jobs than it makes. It is who gets the new jobs, and whether they are standing anywhere near the people who lost the old ones.