The NANDA project research at MIT has identified a critical gap between the potential of generative AI and its actual effectiveness in business. According to the report "The GenAI Divide: State of AI in Business 2025," the vast majority of implementations fail to achieve their stated goals in the short term.
What Happened
According to "The GenAI Divide: State of AI in Business 2025" report, 95% of custom or embedded corporate AI projects do not yield measurable profit or sustainable productivity growth within their first six months of operation. The primary technical reason cited for this failure is the "memory gap" — the inability of modern LLMs to effectively retain context and adapt to specific organizational knowledge.
Context
The problem lies in the architectural staticity of current models. While businesses expect AI to be deeply integrated into processes, existing LLMs are limited in their ability to work with long-term memory. This makes using them as simple "wrappers" without complex engineering overhead ineffective for solving real corporate tasks.
Why It Matters for the Industry
For the industry, this signifies a shift in focus from developing increasingly powerful general-purpose models to creating systems with advanced long-term knowledge storage mechanisms. An increased demand is expected for specialized RAG systems, context management tools, and evaluation (evals) solutions for memory quality in production environments.
Why It Matters for Users
For users and executives, this is a signal that the era of simply plugging ChatGPT into workflows is over. AI success will now depend not on the power of the model used, but on the quality of its integration into specific work cycles and the system's ability to learn from an organization's internal data.
Sources
Author
Look at AI, Editorial Team