The job titles we use in data science have always been a messy, evolving taxonomy, but the last two years have scrambled the categories entirely. A recent exploration from Towards Data Science unpacks roles like AI Engineer, Applied Scientist, and Forward Deployed Engineer, each emerging from the same root: the Large Language Model explosion. Our take is straightforward: these new titles are not just rebranding. They represent a fundamental split in how teams translate model capability into business value, and most organizations are still organizing around the wrong distinctions.
If you are building an AI-native workflow, the old separation between "data scientist" and "software engineer" no longer holds. The new roles demand overlapping skills, but they differ sharply in where they sit between research and production. An AI Engineer focuses on integrating models into products, handling latency, cost, and reliability. An Applied Scientist bridges experimental work and deployment, often validating whether a novel approach actually improves a system. Forward Deployed Engineers embed directly with customers, adapting models to messy real-world data. This is not a semantic debate; it is a hiring and team-structure problem. If you staff your project with three Applied Scientists when you need one Forward Deployed Engineer and two AI Engineers, your model may be elegant but your product will stall. The same logic applies to how we assess these systems. Our related article on Assessing Coding Agents Without Ground Truth via Consistency Mapping shows that even evaluating model reliability often requires new methods, because the old benchmarks were built for static tasks, not dynamic agents.
The practical consequence for our readers is clear: your next hire should not be a generic "data scientist" unless you are prepared to redefine the role on day one. Instead, map the specific bottleneck in your pipeline. Is it model development, integration, or customer-specific tuning? The answer dictates the title. This also reshapes how teams collaborate. When you shift resources between these roles, the metrics you track must change too. Our piece on When you shift marketing spend, does your MMM finally tell the truth? makes a parallel point: a model trained on the wrong inputs will always give misleading outputs. Similarly, a team organized around outdated roles will misallocate its talent.
One specific detail to watch is how these roles handle the "last mile" of deployment. Forward Deployed Engineers exist precisely because models fail in production in ways that lab testing never reveals. The most interesting open question is whether organizations will invest in this role early enough, or wait until their first high-profile deployment crashes against a customer's real data. That decision, hire for the last mile before you need it, will separate the teams that iterate quickly from those that rebuild.