Platforms, pipelines and the tooling layer. How engineering teams actually ship software.
By centralizing image construction through buildpacks, organizations can enforce consistent security policies, automate patch propagation, and generate standardized software bills of materials without relying on individual teams to maintain security expertise across hundreds of repositories.
Open-weight models have grown to handle 56% of token volume on Vercel's AI Gateway as of August, up from just 7% in December. Yet Anthropic's proprietary models still command 64% of spending, revealing a stark divide between usage and cost.
Teams deploying AI are merging code 98% faster but drowning their best engineers in review queues. Codifying feedback, capturing intent, and measuring verification work can restore balance.
A September 2026 security patch broke copy-and-paste functionality across multiple Excel versions. Microsoft has released a fix for Excel 2016 users, while others must work around the issue or roll back the update.
By exploiting zero-heavy weight distributions, Intel researchers compressed ternary model weights to 1.485 bits per weight without retraining, achieving up to 27% faster decoding on GPUs.
AWS has launched Cluster Mode for Elastic Beanstalk, a fully managed deployment option that lets teams run multiple applications on shared Amazon EKS infrastructure with a single operational baseline.
A new report on developer behavior reveals that AI coding assistants trigger addictive patterns similar to gambling, with 80% of developers describing their relationship as a dependence rather than a tool. Managers are rewarding the heaviest users despite quality concerns.
Claude Chat and Cowork merge into a single interface starting Wednesday, with new Docs and Slides features in beta. The consolidation eliminates routing decisions and lets the model handle context across tools automatically.
