AWS re:Invent 2019

December 9, 2019 • #

AWS’s re:Invent conference just wrapped last week. Since we’re so deep into AWS technologies, I keep an eye out each year on the trends visible in Amazon’s product launches. They move at breathtaking speed to fill out their offering suite and keep their current momentum as the leader in the cloud space. They’re really nailing the bundling & scale economics that the likes of Microsoft and Oracle were so successful at in years past. When going upmarket, having a product for every problem outweighs the need for having the highest quality in any individual product line. Enterprises often value the ability to buy everythign they need from a single vendor higher than the quality of the products (what Ben Thompson has referred to as the “one throat to choke” phenomenon).

Here are a handful of the announcements I found most interesting, in no particular order:

AWS Outposts

AWS has finally relented to the customer base that’s been reluctant to move to the cloud for the past decade. With the scale they have now they’ve been able to productize a managed service that puts an “AWS in-a-box” type of modular system into a customer’s datacenter, ideally giving the best of all worlds of security, compliance, and exposure to the AWS services and APIs. It’ll be interesting to see what kind of adoption this gets.

SageMaker Studio

SageMaker is their service for creating, training, and deploying ML models. It’s really an umbrella brand name for about a dozen sub-products for various pieces of the ML workflow. Studio intended to be a full “IDE”-style interface for working with everything you’ve built in SM. Clear indication that this is one of their big strategic plays going forward: lowering the barrier to doing ML and having customers new to the space learning with and expanding from the AWS platform from the start.

Rekognition Custom Labels

Rekognition is AWS’s computer vision service, with endpoints for analyzing video and image data for objects, sentiment, content moderation, and search. One of the barriers for image classification tasks has been the ability to tailor the models to recognize other domain-specific content (like “what kind of part is this?” from a list of parts the customer builds). It now lets you upload your own custom labeled image datasets for training custom Rekognition models.

Amazon Builders Library

This isn’t really a service or expansion on one like the others in the list. This is more a knowledge base of content from Amazon engineers on how they internally build and operate software at scale.