Hi everyone,
I have a general question: in agentic AI projects, how do you typically draw the boundary between a data scientist and an AI/ML engineer? What criteria do you use to decide ownership and responsibilities across the lifecycle (design, building, evaluation, deployment, and ongoing improvement)?
What exactly does a “data scientist” do?
Data Scientist owns the “why” → hypothesis testing, experiment design, eval metrics, iterative improvement (A/B tests that drove my 18% conversion lifts).
AI/ML Engineer owns the “how” → production pipelines, deployment (SageMaker/Databricks APIs), scaling, monitoring (MLflow reproducibility).
Ownership splits at deployment readiness: DS validates the model works in controlled evals, Engineer builds the agentic orchestration, RAG retrieval, prompt chains, and CI/CD. Post-deploy, DS owns metric drift detection, Engineer owns infra/scaling.
Clear handoff: DS delivers validated model artifacts + eval framework; Engineer packages into production agents. Works because I did both ends-to-end.
2 Likes
Thanks for the background.
Often in my real-world experience, there is no clear boundary between those roles. Experienced engineers may end up doing most of what you’ve identified as “data science”. Then it’s called being a “systems engineer”.
How many roles (and titles) are involved depends on the size of the team and the complexity of the tasks.
1 Like
I agree. The role, title, and their responsibilities can mean different things to different people. But I wanted to add that this is due to the novelty of AI in general so a lot of AI-based roles (and even concepts) could have different meanings, not because companies are misinformed or anything like that.
3 Likes