Civilization Chronicles
Three AWS workbenches that put machine learning to work
Remember AWS DeepLens, DeepRacer and Lookout for Vision through their cameras, simulated races and visual inspections.
A camera that could recognize a cat, a car learning its way around a track, and a dashboard for finding defects on a production line: these AWS services gave people different ways to see what a machine-learning model could do. Each wrapped a different model in a workflow that made its output visible.
DeepLens: a model inside a camera
AWS DeepLens paired a camera with a managed project workflow. In AWS's original walkthrough, the author registered the device, chose a cat-and-dog recognition template, and created a project combining a pretrained model with a Lambda function. After deployment to the camera, the code processed video frames on the device and published inference output. A template was a starting point: developers could examine and change the function or build their own. AWS's walkthrough shows the actual project steps; it is an operator demonstration, not a test we performed.
AWS lists the managed DeepLens service in full shutdown, with January 31, 2024 as its end-of-support date. That describes the cloud console and device-management service, not the disappearance of every camera or previously exported piece of code. DeepLens archive.
DeepRacer: teach a model to make a lap
DeepRacer made reinforcement learning visible as a racing problem. A participant trained an autonomous-driving model on predefined simulated tracks, evaluated its performance virtually, and could download the model to a small physical car. AWS's 2018 re:Invent workshops let attendees refine models in the simulator and try them on real tracks. The simulator was part of the experience; owning a car was not necessary to understand the cloud workflow. AWS's original announcement describes both parts.
AWS says its managed DeepRacer service has no longer been available in the console since December 15, 2025, in its current FAQ. AWS's separately deployed open-source solution and downloaded models are outside that statement. DeepRacer archive.
Lookout for Vision: turn examples into an inspection workflow
Lookout for Vision addressed a different task: finding visual anomalies in manufactured items. Customers supplied images of acceptable and anomalous examples, and the service supported training and evaluation around those examples. AWS described using it to flag problems such as scratches, missing components or inconsistent patterns, with reports available in a console dashboard. The 2021 availability announcement describes the intended workflow; it does not establish accuracy for every factory or replace a customer's inspection decisions.
AWS lists Lookout for Vision in full shutdown, with October 31, 2025 as its end-of-support date. That applies to the managed service, not cameras or exported materials. Lookout for Vision archive.
Across the three, the point of interaction changed: a camera produced predictions, a simulator produced laps, and a dashboard surfaced inspection results. Each made a model part of a distinct task, with different inputs and an outcome a person could examine.