Goldman Sachs CEO says AI is changing work, not eliminating jobs

Goldman Sachs will bring on roughly 2,400 to 2,500 interns this year, with a comparable cohort of permanent new joiners arriving in July. For a bank whose CEO has spent the past several months arguing that AI's impact on employment is being dramatically overstated, the hiring numbers are the point.

Speaking on Bloomberg's Odd Lots podcast this month, CEO David Solomon acknowledged that Goldman's hiring will "contract a little" over the next three years but framed that as a modest recalibration rather than structural retreat.

The bank brought in north of 3,000 interns in 2021; this year's intake lands closer to pre-COVID levels. Net headcount for 2026 is still expected to increase.

Solomon's position puts him at the opposite end of the debate from Anthropic CEO Dario Amodei, who has warned of an entry-level wipe-out across knowledge work industries.

He sits closer to Apollo Chief Economist Torsten Slok, who sees no real evidence of AI-driven layoffs materialising today, and Uber COO Andrew Macdonald, who has argued that the economics of automating tasks with AI are harder to justify than they appear.

Goldman has relied on its own chief economist, Jan Hatzius, for analysis, and he points towards temporary dislocations without pointing to any structural increase in unemployment levels in the long run.

There was an internal memo in October indicating slowing hiring and even possible downsizing, with Goldman's typical 3% to 5% reduction based on performance every year shifted up to Q2.

However, the bank has made it clear that OneGS 3.0 is about capacity building, not about reducing headcount, as AI allows the same group to do much more than before.

What worries Solomon more than this year’s hiring figures is the lack of learning on the part of junior bankers in a world where everything is answered by artificial intelligence within seconds.

As he once started his job in finance with no real-time price feeds, having to compare the prices of securities manually by copying them out of The Wall Street Journal and making computations by hand on graph paper.