How can you use agents and keep an interesting data science job?
I don’t want to become CRO, Chief Reviewer Officer, doing only reviews of AI generated code. I also don’t want to become CFO, Chief bug-Fixing Officer, doing only emergency bug fix because things are not going as planned in production. Without giving in to FOMO, I want to be up to date with the ongoing working method. Is it possible to conciliate all these wishes? (spoiler: yes, come to see how!)
The answer I propose is two fold:
- Agents are only doing what they are told to. It means that we have to find a way to make them do state of the art without having to repeat oneself during each prompt about the how-to. This can be done through skills. Furthermore, it means that as humans we still have to find the good questions that will generate things and open doors, instead of only “dry” questions.
- We should limit the amount of code they have to create. Just like any other developer, huge amounts of code is painful to review and keep track. This can be done through abstractions.
Marie Sacksick
Head of Product Management at Probabl, Marie is also co-organizer of Women in Machine Learning and Data Science Paris.