LLM integration: manage change or die

Integrating GenAI is similar to integrating digital projects. I have rarely seen a process more efficient and clear for all stakeholders than the following:

1. Mock-up: Have a tangible vision of what you want to achieve. This is the starting point for everything. The simpler the message, the better it will be understood. It should be straightforward, with clear examples.

2. Proof of Concept: Conduct tests on data, files, small and understandable, in a various environments. This is an opportunity for short experiments, but the end goal is to learn how to scope things correctly so the output is as close as possible to the expected one. It's a chance to get accustomed to various technologies and to prove that there's value in changing things, even those that have been in place for decades. A proof of concept is not a one-time activity, you will need to iterate and more importantly validate each attempt.

3. Minimum Viable Product: With the insights from experiments and proofs of concept, you can confidently move to the next level by extending your tests to a wider range of data and documents. Ensure that you understand each step and interdependency (terms and conditions, etc.) of the services you are using.

4. Large-scale Deployment: At this stage, you should have clarity on how to deploy, including understanding the interdependencies, challenges, timelines, and resources required.

Based on my experience, if you don't approach change management through these four steps, the implementation will likely fail.