Developments that matter in production
What changed, how it works and what it means for the people running reliable systems.
25 stories · 1–12
How we reportUnderstand the new GLM-5.3 security findings, why the benchmark scores differ, and how to evaluate the model for bounded defensive work.
Background reading: AI in productionLearn how OpenAI Dots handles background research and authorized tasks, with a practical SRE maintenance example and checks for permissions and results.
Background reading: AI in productionCodex Cloud reuses prepared repositories and tools while keeping task files separate. Learn what an environment update changes and what it leaves running.
Background reading: Platform engineeringSonnet 5.5 promises faster, cheaper tasks at unchanged token prices. Check the API changes and measure retries before switching operational workflows.
Background reading: AI in productionGoogle’s legacy agents have entered a migration grace period. Compare replacements and check the queries that must survive the switch.
Background reading: ObservabilityCLM-8B scores candidate actions without generating a new answer. Its design suits routing experiments, with important limits on the choices supplied.
Background reading: AI in productionGoogle’s Storage Intelligence update connects usage findings with bulk changes. Dry runs help check the selection before objects are changed.
Background reading: Platform engineeringHow Google’s internal PageBreak project checks suspected vulnerabilities, and what that changes for engineers receiving the reports.
Background reading: AI in productionAlloyDB’s preview puts agent reads on separate compute. Here is what that changes, and how to test a burst of queries after idle time.
Background reading: Platform engineeringAWS brings agent evaluation and application telemetry into one interface. A pilot still needs to verify the evidence available across account and Region boundaries.
Background reading: ObservabilityNVIDIA introduces workload-driven GPU cluster validation. Choose the acceptance claim and thresholds before interpreting a completed run as readiness.
Background reading: Platform engineeringNodeWright brings Kubernetes-aware host changes into a declarative workflow. Check node selection, interruption limits and recovery before a fleet rollout.
Background reading: Platform engineeringDevelopments since our original reporting, with dated updates and supporting sources.
New model documentation makes retained instructions and thinking-block compatibility part of the compaction migration review.
Original report: Claude adds on-demand compaction for long-running agents