REX 2024

1 year ago I wrote a post about what we had seen / done in GenAI at Coddity and with some clients, and had drawn some insights out of it. 1 year later, I wouldn’t change much of it (self-congratulation), but if last year I was amusing myself formulating hypotheses, some of them have turned into certainties.

⚙️ Technical

All GenAI projects look alike and need the same components. It is essential to develop and maintain a common software foundation while avoiding “vendor lock”, so as to quickly benefit from the (numerous) new things, test them (and keep the pace).

For the more technical among you, Langchain brings nothing, and we do without it. 🤷‍♂️

👷‍♀️ Business

Business understanding is paramount, and the success of a GenAI project depends less on the code or on the performance of the model used than on the curiosity and the interest of the person who is going to implement the use case.

You have a project exposed to competition? Be more afraid of the domain expert who understands how an LLM or an agentic system works than of a team of excellent developers and / or data scientists. However, this type of profile will be rare to find, both internally and externally.

🛠️ Production

Monitoring is essential to improve performance. That means hours of reading the results produced and of modifying prompts. Hence the importance of having a modular technical platform with good observability. The load this represents must not be underestimated.

💵 The market for IT services companies

There is one, but it won’t be the goose that lays the golden eggs.

On the large-corporation side, I think everything will be done in-house, with small teams that will develop and maintain the software foundation mentioned above, so no need for an entire floor of developers. All the more so as the subject is too political to be handed to outsiders.

On the other hand, for mid-sized companies, the need is strong, but I think it has to go through “SaaS” with consumption-based billing. But in both cases, staff-augmentation assignments will be rare, and a large share will concern assignments that are more technical business-side support or PO, on transforming domain know-how in order to automate it.

🏭 Impact on organizations

If you think that deploying a Copilot will be enough to transform an organization, you’re not ready for what follows.

Oops, character limit reached, the rest very soon!