Generative AI · Long read
Beyond the DemoNew doors, new rules
Generative AI puts within reach what most companies could not do before. But its projects do not behave like the ones your organisation knows how to run, and RAG is the clearest example.
Summary
Generative AI projects do not behave like the projects your organisation knows how to run. This article follows one example throughout: a system that lets staff ask questions in plain language and get answers drawn from the company's own documents. The technical name is RAG; what matters is what such a project reveals.
GenAI makes the technology affordable, but it makes the management harder. The real work lies less in the AI than in the state of your knowledge: scattered, outdated, contradictory, often owned by no one, and partly never written down. Quality is never simply "done"; it is a level of reliability you must define, measure and choose to accept. The business case cannot be proven up front, only built step by step.
The answer is neither to wait nor to "just do it and see." The disciplines it calls for are not new: governance, learning in short cycles, funding in stages, deciding explicitly what risk is acceptable. What is new is applying them to GenAI, and deciding who owns them. Fix what only you own, and let AI help with the rest.
This article is written for two readers: members of executive committees, who must take decisions they are not used to taking, and CTOs and CIOs, whose role changes as a result.