What problems do companies solve with generative AI?

The same MIT Industrial Performance Center report sorts the generative AI applications it found across its sample of 20-plus companies into three recurring organizational problems. The bottleneck problem is "where workers are responsible for a growing number of simple tasks that get in the way of higher value-added work," which generative AI tools target by "speeding up near-routine tasks." The cafeteria problem "emerges in a process that requires workers to consult experts from various domains and integrate their input into a product, document, or idea," and organizations use generative AI "to predict what those domain experts might have said based on what they have produced in the past." The learning curve problem is "the extra time and effort novices require to complete a complex task in a new domain," addressed by tools "directed to help workers to perform as if they had more experience – and to develop expertise – in new domains."

Each label names a different bottleneck in how work gets done, not a different capability of the underlying model. The bottleneck problem is about throughput: routine steps crowd out higher-value work regardless of who or what performs them. The cafeteria problem is about synthesis: integrating several experts' input is slow because people, not just information, must be gathered, so the report frames generative AI's role as predicting an absent expert's contribution from that expert's past output. The learning curve problem is about tenure: novices lack the situational knowledge experienced workers accumulated over time, so the tool stands in for accumulated experience rather than for raw capability. The report presents the three as the common threads it found across four industry groups, not as mutually exclusive categories or as counts of how many companies fall into each.

This is a taxonomy of the problem an organization is solving when it reaches for generative AI, which sits beside taxonomies in the library that classify a different axis of the same technology. How does generative AI actually change worker skills? classifies what happens to a worker's skill afterward, not what problem the deployment targeted; a cafeteria-problem tool could register as skill erosion or skill development depending on how workers end up using it. How are national lab staff actually using generative AI? classifies the deployment's modality — copilot versus workflow agent — rather than the organizational problem it addresses; a cafeteria-problem use case (predicting an absent expert's input) reads as a copilot modality in that framework. Does generative AI adoption actually eliminate jobs or reshape them? describes what all three problems add up to once the tooling is in place across a company: workers overseeing rather than executing.

The three-problem framework comes from interviews conducted primarily with managers and AI leaders at large incumbent firms, not from a count of deployments or frontline-worker reports, so it says which problems organizations described solving, not how often each one succeeded or how workers experienced the result. The report does not quantify how many of the 20-plus companies fall into each category, nor whether a given deployment addresses more than one problem at once. The taxonomy is best read as the framing the researchers found useful for describing what they saw in legacy firms between 2023 and 2025, not as an exhaustive or measured classification of generative AI's organizational uses.