In this op-ed published on February 23, 2024 on CIO-Online, Olivier Rafal (Consulting Director Strategy at WeNvision) argues a then-counter-current idea: generative AI is more a matter of technological product than of an AI or data science project. Many organizations get their priorities wrong by seeking to hire data scientists and machine learning engineers.

First argument: building a foundation model from scratch « requires several months, millions of euros, and access to enormous quantities of data ». This only makes sense for players with specific, monetizable datasets — the emblematic example being Bloomberg, which created BloombergGPT to leverage its financial data. For nearly all companies, it is better to rely on commercial LLMs.

You don't necessarily need to be a data scientist, but you do need to understand the basic concepts, have back-office development skills, and strong cloud skills.

— **Olivier Rafal** , cio-online.com

Second argument: the skills mismatch. Rather than data scientists, what's needed are development and integration engineers (back and front), strong cloud skills, and DevOps. A client sums it up: « You don't necessarily need to be a data scientist, but you do need to understand the basic concepts, have back-office development skills, and strong cloud skills. »

Third argument: architecture. Companies should build a plateforme d'IA générative resting on orchestrators and APIs, which makes it « easy to work with the best LLMs on the market and switch between them as they each evolve, without reworking the applications » — a guarantee of independence against vendor lock-in.

Fourth argument, the central one: the shift from project to product. « The platform […] must itself be considered a product », funded not by a one-off investment but by a monthly stream, to sustain continuous iterations and innovation.

Fifth argument: governance. GenAI « has become democratized in an unprecedented way », which generates « as much shadow AI as strong expectations toward the DSI ». Adequate governance must capture the needs of the various business lines, prioritize products by the value they create, and oversee the proper functioning of the whole.

Rafal finally announces a paradigm shift that needs to be accepted: « we are moving from classic algorithmic programming to agents Langchain that handle part of the decisions ». A founding text, it lays down as early as the beginning of 2024 the foundation (product, platform/API, flow-based funding, governance) that later analyses — up to the 2026 FinOps/token financial governance — will only extend.