Generative AI needs embedding
A Wharton analysis warns that generative AI alone doesn't create value unless organisations plan for emergence, enablement and embedding—meaning tools must be integrated into workflows to produce measurable outcomes. The write-up frames AI as a capability that requires strategy and process work rather than a plug‑and‑play solution. (knowledge.wharton.upenn.edu)
Generative artificial intelligence can draft text, code and images on command, but a Wharton analysis says companies do not get value from the tools by installing them alone. (knowledge.wharton.upenn.edu) Rahul Kapoor, a Wharton management professor, wrote on April 13, 2026 that business value comes through three stages he calls “emerging,” “enabling” and “embedding.” He argues firms that stop at experimentation or model access will fall behind companies that redesign work around the technology. (knowledge.wharton.upenn.edu) In Kapoor’s framework, “emerging” refers to a new technology’s raw technical promise, “enabling” covers the complementary assets needed to use it, and “embedding” is the step where the tool is built into everyday processes, decisions and incentives. He lists compute, energy, specialized hardware, proprietary data and talent as constraints that can limit returns even when the underlying models improve. (knowledge.wharton.upenn.edu) That argument lands as companies move from pilot projects to return-on-investment questions. McKinsey estimated in 2023 that generative artificial intelligence could add $2.6 trillion to $4.4 trillion annually across 63 use cases, but it also said the gains depend on changes to workflows and worker activities, not software deployment alone. (mckinsey.com) Research from Boston Consulting Group and Harvard Business School found the same pattern at the worker level. In a large experiment published in 2023, consultants using Generative Pre-trained Transformer 4 completed some tasks faster and at higher quality, but users who pushed the tool outside its strengths performed worse than those who did not use it. (bcg.com) Wharton’s point is that generative artificial intelligence behaves less like a plug-in and more like electricity or cloud computing: a general-purpose capability that becomes useful when other systems adapt around it. Kapoor says companies need changes in business models, coordination and organizational design, not just access to a model vendor. (knowledge.wharton.upenn.edu) That view also lines up with other management research. MIT Sloan wrote in 2025 that businesses need to distinguish between generative artificial intelligence and other machine-learning tools, because the right choice depends on the task, the data and the decision being made. (mitsloan.mit.edu) The Wharton essay does not argue against adoption. It argues that the companies most likely to show measurable results will be the ones that connect models to proprietary data, train workers, rewrite processes and decide where human judgment still stays in the loop. (knowledge.wharton.upenn.edu) The practical message is narrower than the hype cycle. Generative artificial intelligence may keep getting better, but Wharton’s analysis says the payoff still depends on the old corporate work of integration, staffing and process design. (knowledge.wharton.upenn.edu)
