🤖Pay & Market

Most days split between reading recent papers, writing and debugging training code, waiting on and analyzing results from long-running experiments, and discussing findings with collaborators. Contrary to the popular image, a large share of the job is diagnosing why a model is underperforming or a result won't replicate, not brainstorming new architectures from scratch.

For an independent research role at a university or a top industry lab, effectively yes — a PhD remains the field's real credential, since there is no licensing exam. Strong engineers without one increasingly enter through industry residency programs or a public record of open-source work and cited preprints, but titles like 'research scientist' still skew heavily toward doctorate holders.

Parts of it clearly are: literature summarizing, boilerplate experiment code and first-draft related-work sections are already routinely AI-assisted, and labs are openly experimenting with systems that propose and run their own experiments. Framing what a result actually means, and taking responsibility for a published claim, have proven much harder to hand over.

It varies enormously by employer and seniority. The US median for the broader 'computer and information research scientist' category was $140,910 in 2024, but new PhDs joining a large technology company's research lab often start well above that, and a small number of senior researchers at frontier AI labs have reportedly been offered total packages worth seven or eight figures during the 2024–2025 hiring competition.

The lines blur constantly, but a rough distinction holds: a researcher's output is usually a paper, a new method or an insight about why something works, judged by peer review; an engineer's output is usually a working system in production, judged by whether it ships and holds up under real use. Many people do both across a career.

Python dominates, paired with a deep-learning framework — PyTorch is now the most common in research, with JAX popular for large-scale and Google-affiliated work. Beyond code, researchers depend on shared GPU or TPU compute clusters, experiment-tracking tools like Weights & Biases, and arXiv, where most new results circulate as preprints months before formal peer review.