Platforms, pipelines and the tooling layer. How engineering teams actually ship software.
As Kubernetes deployments expand to edge environments and AI workloads strain memory limits, the ecosystem is shifting toward Linux cgroup v2 and adopting new operational practices to handle distributed, resource-constrained scenarios.
Ryan Dahl's Deno team is joining Cloudflare to combine its open-source Celld project with Workerd, ending Deno's standalone runtime and competing with Workers directly.
Cloudflare has launched profiling capabilities for Workers and Durable Objects, allowing developers to capture interactive flamegraphs and identify resource bottlenecks in production workloads.
Peter Farkas argues that releasing model weights without source code or training data does not constitute true open source, and warns that loose terminology in AI could weaken the definition across all software.
Harness has acquired Augment Code's Cosmos software factory and related tools to integrate autonomous coding agents with its delivery platform. The most significant capability—learning from failed deployments—remains unshipped.
Independent testing of Claude Sonnet 5.5 against the newer Opus 5.5 reveals the cheaper model outperforms on two of three complex coding tasks, despite requiring more tokens per operation.
Agent-friendly APIs demand a different approach to scaling than traditional web services. Applying decades-old microservices patterns—bulkheads, statelessness, and smart endpoints—alongside a unified database backend can handle bursty AI traffic while preserving conversation context across replicas.
A new open source proxy layer from Percona will let applications built for single-instance Redis move to clustered Valkey without code rewrites. The tool is heading to public repositories this month.
