April 1, 2026
The nerds strike back
The revenge of the data scientist
Data scientists clap back: AI isn’t magic, it’s messy—and they own the mop
TLDR: LLMs made adding AI as simple as calling a service, but commenters say the hard part—testing, measuring, and understanding messy data—is where data scientists shine. The thread pits “just call the API” optimism against “prove it works” realism, with most agreeing rigor beats hype and that matters for real-world AI.
The “sexiest job” isn’t dead—it’s plotting a comeback, at least if you believe the comment section. After a talk titled “The Revenge of the Data Scientist,” the crowd rallied around a spicy thesis: calling a big AI model over an API doesn’t kill the job—it makes the messy parts matter more. Think experiments, guardrails, and figuring out what the data even is.
One top-voted voice turned it into school: jamesblonde said to treat your prompts like training data and your checks like test data—simple, blunt, and very teacher-mode. maxwg cheered when there are clear benchmarks (see pg_textsearch), but warned that on brand-new projects, writing good tests is “its own job.” Translation: the easy button isn’t easy.
Then came the reality check: djoldman said the biggest lift is confirming what the data actually is, not what people assume. uduni piled on, saying they get more from just watching an AI agent run than from elaborate “LLM-as-judge” setups—aka bots grading bots. Meanwhile, convexly threw shade at the “prompt engineering” hype: this is the same old loop (define good, measure, iterate), and veterans already have the playbook.
Verdict: the drama is API optimists vs. measurement diehards. But the comments largely agree—the revenge is real because AI still needs grown‑ups who test, debug, and say “no, your data isn’t what you think.”
Key Points
- •The article argues that LLMs and foundation-model APIs reduce reliance on in-house model training but do not eliminate core data science work.
- •Historically, predictive modeling paid the best and was split into the Machine Learning Engineer role.
- •Data scientists’ main contributions include experiment design, generalization testing, debugging stochastic systems, and metrics design.
- •Teams can call LLMs via APIs, but evaluation and experimentation remain necessary to ensure performance on unseen data.
- •The author delivered a talk at PyAI Conf, “The Revenge of the Data Scientist,” presenting examples supporting these claims.