Does negotiating power, not just productivity gains, decide who actually profits when AI makes work more efficient?

This explores whether workers' ability to negotiate, their leverage, decides who gets the gains from new technology over the long run, and what AI does to that leverage. The corpus doesn't have classic labor-economics work on unions or wage bargaining. What it does have is a cluster of AI-economics work that treats bargaining power as the hidden variable deciding whether productivity gains reach paychecks. The clearest statement is Brynjolfsson's argument that the design choice matters: AI built to replace workers rather than augment them hands leverage to whoever owns the technology. That can settle into a self-reinforcing trap, where workers lack the power to change the arrangement that took their power away Does automating labor concentrate power more than augmenting it?.

The long-run wage story follows from that. Anthropic's scenario modeling finds that as AI growth speeds up, labor's share of income falls and capital's rises. Average wages can still climb while knowledge-worker wages stall, because ownership is concentrated and workers can't easily move between occupations Does AI growth inevitably shift wealth away from workers?. Acemoglu reaches a similar headline from a different direction. AI's wage effects may spread more evenly across demographic groups than earlier automation did, but capital income still pulls ahead of labor income Will AI increase inequality less than past automation?. The most extreme version is the AGI-economy argument: once machines can do bottleneck work, a human wage stops reflecting the value the person creates. It falls to whatever it costs to run the compute that could replace them What happens to human wages in an AGI economy?. Put simply, your negotiating ceiling becomes the price of your machine substitute.

There is a counter-current, and it is about where leverage comes from. Gans and Goldfarb point out that when job tasks depend on each other, so one weak link spoils the whole result, automating some tasks concentrates value in the ones humans still do. Standard 'AI exposure' scores may therefore overstate how much displacement weakens workers Does automating some tasks actually increase the value of remaining work?. Being the bottleneck is a form of bargaining power. Similarly, a large consulting experiment found GPT-4 narrowed the gap between weaker and stronger performers, which could compress wage premiums for top talent. That held only when AI use was mandatory Does AI help lower-skilled workers catch up to top performers?.

The timing matters too. Danish administrative data show that two years after ChatGPT launched, earnings and hours hadn't moved, even though tasks had already been reorganized Does AI chatbot adoption change worker pay and hours?. Work changes first and pay follows later, which is when bargaining power gets tested. Firms don't move evenly either. Highly exposed firms replace freelance workers with AI faster and more cheaply than other firms Do firms substitute labor for AI at different rates?, so a worker's leverage may depend as much on their employer's AI capability as on their own skills. Exposure also lands unevenly by gender: in female-dominated fields, lower-paid workers are as exposed as higher-paid ones, with fewer resources to push back Does AI exposure hit low-wage workers harder in some fields?.

One lateral twist you might not expect: bargaining itself is being automated. In AI-agent-to-agent negotiations, the more capable model reliably wins better deals, whether it is buying or selling Does model capability determine who wins in agent negotiations?. If wage and contract negotiations come to run through AI agents, leverage could follow access to the best model rather than collective organization or scarce skills. That is a new route for the power concentration Brynjolfsson warns about.