For years, the data science crowd treated software engineering best practices like optional reading. Version control? Testing? Deployment pipelines? Nice ideas, sure, but not really what the cool kids were doing while they chased Kaggle leaderboard scores. That era is ending, and not a moment too soon.
The problem is obvious once you look at where ML actually lives now. We’re not building toy models for sandbox demos anymore. Machine learning is running against national power grids, supermarket supply chains, and other systems where failure costs real money. The article puts it bluntly: the maturity of the industry hasn’t caught up with the scale of deployment. Google’s own engineering guidance captures the shift perfectly: ‘Do machine learning like the great engineer you are, not like the great machine learning expert you aren’t.’ Translation: stop treating ML like a science fair project.
This is where the DevOps crowd comes in. The infrastructure problem, which used to eat entire quarters of work, is largely solved. Cloud APIs, infrastructure-as-code, Kubeflow, SageMaker — the plumbing exists. The Kaggle survey backs this up: 75% of practitioners already use cloud services, and 45% rely on enterprise platforms like SageMaker, Databricks, or Azure ML Studio. The bottleneck isn’t compute. It’s the process of getting models out of notebooks and into production where assumptions get validated against reality.
So what does the hiring math look like? The author’s answer is blunt: hire more ML-savvy software and DevOps engineers, not more data scientists. That’s a contrarian take that will ruffle some feathers in the data science community, but the logic holds. An okay model running in production beats a brilliant model stuck in a sandbox every single time. Meanwhile, the Transformer architecture keeps spreading beyond language — Vision Transformer and CoAtNet are setting image classification benchmarks with less compute. That’s the ironic part: the models keep getting more general while the real bottleneck is embarrassingly specific. It’s not the math. It’s the engineering.
