This article is based on reporting originally published by SiliconAngle ahead of AWS re:Invent 2019.
AWS Lowers the Barrier to Machine Learning
In the lead-up to its annual re:Invent conference in late 2019, Amazon Web Services (AWS) announced a set of new features aimed squarely at making machine learning more accessible to mainstream developers—not just data scientists. The central goal, articulated by AWS principal Matt Asay in an official blog post, was to bring Amazon’s machine learning capabilities within reach of anyone who can write SQL.
Until these updates, integrating ML predictions into an application required developers to extract data from storage, transform it into a compatible format, and pipe it into a separate machine learning pipeline. That process added custom code, complicates data governance, and introduced security overhead. The new features were designed to eliminate much of that friction.
Key Feature Updates
ML Predictions Directly from Amazon Aurora and S3
AWS announced that machine learning predictions would soon be executable against data stored in Amazon S3 (the company’s primary object storage service) and Amazon Aurora (its MySQL- and PostgreSQL-compatible managed relational database). This means developers can invoke ML models directly from familiar SQL queries rather than building separate data pipelines.
Similarly, AWS Athena—the serverless interactive query service for analyzing data held in S3—was updated to support ML model invocations inline with standard SQL queries.
QuickSight Gains Machine Learning Visualization
Amazon QuickSight, the company’s business intelligence and data visualization service, was updated to visualize and report on model predictions generated by services such as AWS SageMaker. This lets teams surface AI-driven insights directly in their existing dashboards without needing to build additional tooling to shuttle predictions from SageMaker into a reporting layer.
No Custom Integration Code Required
As Asay wrote in the announcement: “Now anyone who can write SQL can make—and importantly, use—predictions in their applications without any custom code.” By removing the requirement to call ML endpoints from application code, AWS reduced both the implementation effort and the ongoing maintenance burden for development teams.
Why This Matters for DevOps and Development Teams
Historically, operationalizing machine learning models has been one of the harder problems in the ML lifecycle. Even after a model is trained and validated, integrating it into a production application often requires:
- Custom ETL pipelines to move data into a format the model accepts
- API integration code to call inference endpoints
- Additional security controls and IAM policies for each new data path
- Monitoring and logging for prediction latency and errors
By embedding predictions directly into SQL queries against Aurora and Athena, AWS shifted much of this operational complexity into managed infrastructure. For DevOps engineers, this means fewer custom services to maintain and a narrower attack surface for security incidents related to data movement.
Notable Customer Announcements
ProSiebenSat.1 Chooses AWS as Primary Cloud Provider
Alongside the feature announcements, AWS revealed that European satellite broadcaster ProSiebenSat.1 Media SE, headquartered in Germany, had selected AWS as its primary cloud provider. The company stated plans to integrate Amazon’s machine learning services throughout its business operations—including automating internal processes and building personalized products for viewers.
Google Cloud Also Advances AutoML for Manufacturing
On the same day as the AWS announcements, Google provided updates on its own AI efforts, specifically around AutoML Vision—its service for training custom image classification and object detection models without deep ML expertise. Google highlighted adoption by two major industrial enterprises:
- GlobalFoundries: The semiconductor manufacturer is using AutoML Vision to build a visual inspection tool that detects defects in Wafer Map and Scanning Electron Microscope (SEM) images—a task that previously required highly specialized manual review.
- Siemens: The industrial conglomerate is using AutoML Vision to create a “Factory AI service” that accelerates the prototyping of new products by automating parts of the visual quality-control process.
These use cases illustrate a broader industry trend: cloud providers are competing not just on compute pricing and raw ML capability, but on reducing the expertise threshold required to deploy AI in real production environments.
The Bigger Picture: Democratizing ML in the Cloud
The moves by both AWS and Google reflect intensifying competition around ML accessibility. Rather than requiring organizations to hire teams of data scientists or ML engineers to deploy predictions, cloud providers are increasingly offering managed services that let SQL-fluent analysts and general-purpose developers participate in the ML lifecycle.
For DevOps teams, the practical implication is a shifting set of responsibilities: infrastructure-as-code templates now need to account for SageMaker endpoints, Athena workgroups with ML permissions, and QuickSight dataset configurations—all as first-class managed resources rather than bespoke application code.
As these services mature, the engineering effort required to go from a trained model to a production prediction is shrinking—which changes how DevOps practitioners should think about the ML deployment pipeline and where automation investment pays the highest return.