Daily Edition · Page 16

THE DAILY INDEX

Technology, AI, software, security, crypto and business

GENERAL

TLDR DevOps

DIGITALOCEAN INFERENCE ROUTER, NOW CACHE-AWARE: WHY THE CHEAPEST MODEL ISN'T ALWAYS THE BEST DEAL

DigitalOcean has added cache-aware routing to its Inference Router, letting the system account for the cost of breaking a warm prompt cache when deciding whether to switch models mid-session. The update pairs with two new controls: an explicit model affinity header that ties requests to a session, and a routing budget parameter that caps how much extra cost a model switch can incur.

AWS GLUE 6.0 NOW AVAILABLE WITH 30% LOWER PRICE AND FULL APACHE ICEBERG V3 SUPPORT

AWS Glue 6.0 is now generally available in all AWS regions where Glue operates, with pricing 30% lower than previous versions. The release is built on Apache Spark 4.1, Python 3.13, and Scala 2.13, and includes full support for the Apache Iceberg v3 specification via Iceberg 1.11.0. A new VARIANT data type with shredding support lets teams store and query JSON, logs, and event data without flattening schemas. No API changes are required to use the new version.

DOCKER VERIFIED PUBLISHER APPS ARE NOW SELF-SERVE

Docker has made its Verified Publisher program self-serve, so software vendors can now apply directly through Docker Hub instead of going through a sales team. Approved publishers get a verified badge, prioritized search ranking, and access to analytics that identify which companies are pulling their images. The badge covers all content types on Docker Hub, including MCP servers, models, sandboxes, and agents. Current verified publishers include Google, Microsoft, AWS, Datadog, and Grafana Labs.

WILD AI-RELATED RELIABILITY INCIDENTS ARE COMING

AI agents acting as autonomous first responders may successfully resolve many routine incidents, but their ability to take operational actions also creates new failure modes when they encounter situations they cannot handle. The most dangerous scenario is an agent attempting remediation in an unexpected way, making an already complex outage worse before humans enter the loop and forcing responders to diagnose the combined behavior of the software system and the agent itself.

THERE CONTINUE TO BE REASONS FOR SOFTWARE TO BE SLOW

LLMs can make performance work cheaper to implement, but implementation cost is often not the main reason optimizations do not ship: teams may tolerate slower software, cut engineering budgets, deprioritize performance, or face larger costs in testing, migration, correctness, and long-term maintenance. AI-generated optimizations can also overfit benchmarks or introduce reliability problems, so faster code generation does not necessarily translate into faster production software.

TLDR IndexPage 16