Why do AI rollouts succeed or flop based on training and teamwork, not just which model a company buys?

This explores why the human and organizational side of AI adoption, meaning the training, redesigned workflows, coordination habits and trust structures around the tools, decides whether AI pays off. One caveat first: the corpus doesn't use the phrase "sociotechnical capital" directly. It does keep returning to the idea from several directions, and the agreement across them is striking: the model is the cheap part, and what's valuable is everything a company builds around it.

The clearest economic case comes from the productivity J-curve Why does AI productivity appear weak when investment is high?. Companies adopting GPT-style tools quietly invest heavily in training, new processes and custom software. None of that shows up on the books as an asset, so productivity looks weak at first. Later, once those investments start producing output, productivity looks better than it really is. In other words, sociotechnical capital is real capital that accounting doesn't see. That helps explain why AI's payoff seems to lag the hype. BCG's survey of nearly 12,000 employees points the same way from the ground up: clear strategic direction predicted lasting value better than access to tools did, and companies that redesigned workflows end to end got more out of AI than those that simply handed tools out Does strategic direction matter more than AI tool access?.

The reason this capital is hard to build is a coordination problem you might not expect. Agrawal, Gans and Goldfarb show that AI makes individual decisions better but also more varied, so the different parts of an organization start drifting out of sync Does AI adoption require rethinking how organizations coordinate?. The gains only add up if the organization changes how its parts communicate. Evans makes a related point: making AI tools easier to build doesn't fix the real bottleneck. Most workers don't see their own tasks as automatable, and real adoption depends on decisions that cut across departments and budget cycles Does easier tool-building actually solve enterprise adoption problems?. OpenAI's own enterprise data shows the result: use is uneven, with marketers and early-career staff far ahead of executives Who adopts enterprise AI first and how do they use it?. So the people who set direction are often the ones with the least hands-on experience.

A second, less obvious kind of value is trust. Shneiderman argues that reliability, safety and trust come from three separate sources: good engineering, management culture and independent oversight. Improving one doesn't automatically improve the others Can we separate reliability, safety, and trust into distinct mechanisms?. A related point comes from work on expertise: authority is granted through membership in a community and a track record, and AI can't join that circle on its own Can AI ever gain expert community trust through participation?. That makes the people and institutions that vouch for AI outputs part of the asset. Agentic AI makes this harder still. 76% of executives surveyed see it as a coworker rather than a tool, which means it has to be managed with HR-style and asset-style approaches at the same time Does agentic AI fit the tool-or-coworker framework?.

Here's what you might not have thought to ask: who ends up owning this capital? Brynjolfsson warns that AI designed to replace workers rather than help them shifts bargaining power to whoever controls the technology, and that this can harden into a trap that's hard to escape Does automating labor concentrate power more than augmenting it?. If sociotechnical capital lives in workers' know-how and redesigned teamwork, choosing to augment people preserves it. Choosing to automate them away hands it over. The benchmark-to-GDP gap shows why this matters: agents that win contests still fail at long, real-world professional workflows Why do agent benchmarks not predict real economic value?. The distance between what a model can do and the economic value it creates is largely made up of this organizational capital, and it's the part no vendor can sell you.