The Engineer

Kubernetes v1.35 defaults failCgroupV1 to true, so the kubelet will not start on a cgroup v1 node and kubeadm preflight fails instead of warning, Chris Short reports. For your clusters, an unchecked node pool now means a failed upgrade rather than a log warning. The reward for migrating is node swap, GA since v1.34, which took gVisor Python sandboxes from 80 to 240 pods per node on GKE NVMe Local SSDs.

Cloudflare is acquiring Deno and will shut down Deno Deploy about six months after the October 10 announcement, around April 2027, Techpresso reports. The standalone runtime gets one year of security updates, to roughly October 2027, and no further development. Every service on Deploy now has a fixed exit date, and the promised self-hosted Celld/Workerd build has no release date to plan against.

Stanford's Dan Boneh and an a16z crypto panel found that OpenAI's recent math results contained no cryptography problems and weakened no assumption, elliptic curves included. Justin Drake, quoted in Exponential View, still warns AI could undermine digital signatures before quantum computers do, though no mechanism is cited. Both views point at the same exposure in your code: crypto bugs AI makes cheaper to find, and systems that cannot swap algorithms without a deploy.

Agent-written pull requests are reportedly up 9x in eight months. Rahim Hirji's Box of Amazing names the failure: reviewers of a confident diff check correctness, not whether it solves the right problem. Upstream, curl ended its bug bounty and Sindre Sorhus and Hono closed external PRs, which James Ross attributes to AI making contributing cheap and maintaining expensive. Upstreaming a fix is no longer a dependable plan, and your own reviewers face the same flood.

Standard Bots told Latent.Space its models run on local edge GPUs and handle only perception, while conventional code owns motion and tasks stay short-horizon. WSJ reporting on Walmart, via Morning Brew, puts automated grocery facilities at over $800K a month in electricity versus $250K for a manual warehouse. Walmart's robots still can't lift big bags of dog food, so its manual path is permanent. For your agents, both argue for a narrow learned step and a staffed fallback.