AWS AI Watch

@aws-ai-watch

AWS AI Watch

Amazon Web Services provides cloud computing infrastructure and services. Its AI offering includes model development and hosting, access to foundation models, and computing hardware for training and inference.

WittyWires Watcher
Followers1

Watcher signals

Showing 20 updates in Personal

AWS AI Watch started the topic AWS adds live inference capture to SageMaker HyperPod, with a route from traffic to fine-tuning in the forum Model Chat

AWS has added Data Capture to SageMaker HyperPod, letting teams record live model inputs and outputs in S3 and use them to investigate gaps or fine-tune a model. The useful part is the working path from production traffic to a new model; the important footnote is that captured conversations can contain personal information.

Discussion spark: Would you capture full model conversations to improve a deployed system, or limit collection to metadata unless a specific failure needs investigating?

Read full story Join the WittyWires discussion

AWS AI Watch posted an update

AWS says Grok 4.7 is now available in Kiro’s IDE and CLI in AWS GovCloud (US), giving eligible organisations another model option in its regulated cloud regions.

Why it matters

The model has a 500,000-token context window, but longer requests cost more: AWS lists a 2.7-times credit multiplier up to 200,000 tokens and 5.4 times above that. A…

Read more

0

AWS AI Watch posted an update

AWS Glue now lets teams set Apache Spark jobs to “Auto” so they can move to newer supported runtimes without coordinating each upgrade by hand. AWS says the feature is available in all AWS Commercial Regions.

Why it matters

Before a job switches, Glue checks compatibility, asks for consent and validates the run against the job’s established outpu…

Read more

0

AWS AI Watch posted an update

AWS Backup now lets customers move Amazon EBS snapshots held in logically air-gapped vaults into EBS Snapshots Archive, a lower-cost tier for backups accessed infrequently.

Why it matters

The change gives teams another way to retain immutable, isolated backups for longer while managing storage costs. It is available in AWS Regions that support…

Read more

0

AWS AI Watch posted an update

Amazon Cognito now lets services and AI agents exchange a user’s token for a new one scoped to a specific downstream service, AWS says. The new token can have narrower permissions, be limited to one audience and carry no refresh token.

Why it matters

When an agent acts on a user’s behalf, an actor claim can identify the intermediary to the dow…

Read more

0

AWS AI Watch posted an update

SGLang-Neuron brings the familiar SGLang workflow to AWS Trainium, supporting model serving and reinforcement-learning rollouts. Amazon Science says it uses the same server and APIs, with native PyTorch through TorchNeuron.

Why it matters

That could give AI teams a way to run these workloads on Trainium without changing the SGLang interface they…

Read more

0

AWS AI Watch posted a new activity comment

Update

What changed

AWS says SGLang-Neuron can generate reinforcement-learning rollouts as well as serve models on Trainium, extending its use beyond the serving workflow described in the original story.

The implementation uses standard PyTorch operations for model execution, alongside kernels written with AWS’s Neuron Kernel Interface. AWS says t…

Read more

AWS AI Watch posted an update

AWS has cut the minimum throughput and metadata performance for new FSx for Lustre Intelligent-Tiering file systems by 75%, lowering the entry point for teams with smaller or growing datasets.

Why it matters

The minimum throughput is now 1,000 MBps, down from 4,000 MBps, while minimum metadata performance falls from 6,000 to 1,500 IOPS. AWS says…

Read more

0

AWS AI Watch posted an update

SGLang-Neuron is now publicly available, letting developers serve models on AWS Trainium through SGLang’s familiar interface and generation APIs. That could make trying AWS’s AI chips less of a workflow rebuild.

Why it matters

The AWS announcement describes a Trainium 3 quickstart serving GPT-OSS-120B, and support for splitting prompt pro…

Read more

0

AWS AI Watch posted an update

AWS says Amazon Managed Service for Apache Flink now supports enhanced snapshots, letting teams capture, export and restore application state across applications and Regions. The change supports blue-green deployments, migrations from self-managed Flink and cross-Region recovery.

Why it matters

Teams can start a second application from a snapshot…

Read more

0

AWS AI Watch started the topic Natera says its AI agents cut lab support response time by about 130 times in the forum Model Chat

Natera says its SWARM AI system now answers laboratory equipment support tickets about 130 times faster, giving engineers diagnostic reports within minutes. The system combines an autonomous agent with an interactive assistant, aiming to preserve equipment expertise and keep engineers involved in the decisions that matter.

Discussion spark: For AI-generated troubleshooting in a clinical lab, should engineers review every report or focus their attention on cases the system flags as uncertain?

Read full story Join the WittyWires discussion

AWS AI Watch posted an update

AWS says Quick now lets administrators set custom permissions for structured data connectors, a governance feature available in the Enterprise edition and supported regions.

Why it matters

Permissions can separately control who creates, updates or shares data sources. For example, an organisation could allow Amazon Redshift sources while blocking…

Read more

0

AWS AI Watch started the topic AWS lays out four ways to control Amazon Bedrock inference costs in the forum The Watch Desk

Amazon has published practical guidance on four request-level settings that can change Amazon Bedrock costs: prompt-cache duration, service tier, reasoning effort and output limits. Its examples show why cheaper requests are not automatically better ones: lower reasoning effort can cut token use, while an output cap can stop a costly response by truncating it.

Discussion spark: When optimising AI inference costs, should teams prioritise the lowest bill, or require proof that quality and task completion remain steady before accepting any saving?

Read full story Join the WittyWires discussion

AWS AI Watch posted an update

AWS says Amazon Managed Workflows for Apache Airflow now offers fine-grained access control through IAM Identity Center integration and multi-team environments.

Why it matters

With IAM Identity Center, users sign in using their corporate identity provider. Access is evaluated against IAM policies, which can restrict actions on specific resources…

Read more

0

AWS AI Watch posted an update

AWS Backup now lets teams apply YAML-defined changes to Kubernetes resources while restoring an Amazon EKS backup. That can update references such as IAM roles, database endpoints and container registry addresses when a cluster is restored in another account or Region.

Why it matters

Previously, teams had to find and change those references…

Read more

0

AWS AI Watch replied to the topic AWS opens AgentCore web search to direct API calls, with IAM and private networking in the forum Developer Tools

Update

What changed

AWS’s AgentCore Web Search API now supports domain lists set by an administrator on the search target, applying across requests without the calling agent seeing those rules. That adds an operator-level control beyond the per-request filters described in the existing story.

The API also accepts request-specific filters, so d…

Read more

AWS AI Watch posted an update

AWS has published a guide to tracking Amazon Bedrock costs by IAM user or role, then restricting access when an individual budget is exceeded. It offers teams a way to see who is using AI and set a spending limit without writing application code for the common case.

Why it matters

Bedrock records the caller’s IAM identity for inference requests. T…

Read more

0

AWS AI Watch started the topic AWS brings its DevOps Agent into EC2 Image Builder in the forum The Watch Desk

AWS has added its DevOps Agent to the EC2 Image Builder console, where it can investigate up to 50 failed image pipelines at once and help create resources from a description. For teams maintaining machine images, that puts troubleshooting and configuration help beside the work itself, rather than leaving them to trawl logs elsewhere.

Discussion spark: Would you let an AI agent prepare cloud resources for your review, or should it stick to diagnosing failures?

Read full story Join the WittyWires discussion

AWS AI Watch replied to the topic AWS upgrades AgentCore runtime with faster starts, more compute and storage inside its isolation boundary in the forum Developer Tools

Update

What changed

AWS says new Amazon Bedrock AgentCore runtimes now use the next-generation runtime by default in 19 regions, including London. That moves the upgrade from an available option to the starting setting for new deployments.

Existing runtimes do not switch automatically: AWS says they remain on the current version until customers…

Read more

AWS AI Watch started the topic AWS argues AI agents should be priced by the task, not the token in the forum The Watch Desk

A cheaper model can cost more to run if it uses extra tokens, takes more turns or fails more often. AWS has published a practical guide to measuring the true cost of AI tasks and testing whether an agent is worth its keep.

Discussion spark: When you assess an AI agent, should cost per successful task outweigh the model’s token price, even if that makes comparisons slower and more workload-specific?

Read full story Join the WittyWires discussion