TLDR DevOps 2026-05-15

AI is accelerating software delivery, but most release strategies were not built for AI-generated changes. At Flagship: The Feature Management and Experimentation Summit hosted by Harness, leaders from Cursor, Box, United Airlines, JPMC, and more will share how modern teams are adapting feature delivery for the AI era including how to:✅ Govern and release AI-powered features with confidence

✅ Experiment safely before changes impact customers

✅ Scale feature delivery without increasing risk

Featuring sessions on feature flags, experimentation, rollout strategies, and release reliability for AI-driven applications.

Join liveon June 17 (10:00 AM PT / CET)

Register now>

Prempti is an open-source security tool that uses Falco's detection engine to intercept and evaluate AI coding agent actions in real-time, allowing developers to allow, deny, or manually approve tool calls before agents access sensitive files like SSH keys or execute potentially risky commands. The lightweight service runs without root access and includes default rules protecting against common threats like credential theft, prompt injection, and unauthorized network calls, while offering both enforcement and monitor-only modes for teams adopting AI coding assistants.

Kubernetes v1.36 introduced a major architectural overhaul for workload-aware scheduling by splitting the Workload API into a static template and a new PodGroup API that handles runtime state, enabling atomic scheduling of entire pod groups rather than individual pods. The release also added topology-aware scheduling to reduce network latency in AI/ML workloads, workload-aware preemption that treats pod groups as single units, Dynamic Resource Allocation support for shared devices across massive workloads, and native Job controller integration that automatically creates gang-scheduled workloads for indexed parallel jobs.

The Gemini CLI CI/CD extension bridges the gap between local AI-assisted development and production deployment by enabling conversational deployment, pipeline generation, infrastructure provisioning, and security scanning across Gemini CLI, Claude Code, and Antigravity environments. Using MCP servers, AI skills, and a cloud architecture knowledge base, the system automates Cloud Run deployments, Cloud Build pipelines, and Google Cloud infrastructure setup while preserving developer control through conversational approvals and least privilege access.

AWS DevOps Agent enables autonomous SRE workflows by correlating telemetry across CloudWatch, Splunk, GitHub, and Slack to investigate incidents, identify root causes, generate mitigation plans, and produce agent-ready remediation specs for coding agents like Kiro. The architecture uses Agent Spaces, webhooks, MCP integrations, and customizable skills to automate multi-cloud incident response, reducing manual troubleshooting and mean time to resolution while improving operational scalability and governance.

prove and TAP provide a lightweight way to test Unix shell tools without adding a heavy framework: each shell test prints simple ok / not ok results, and prove discovers, runs, summarizes, parallelizes, and reorders them. A small shared common.sh harness can handle temp directories, assertions, captured stdout/stderr, bailouts, and TAP formatting, making shell test suites easy to grow while keeping failures readable.

Agent-generated pull requests are increasing review load while quietly introducing more redundancy and technical debt, making human judgment and contextual review more critical despite cleaner appearing code and passing tests. Effective review focuses on blocking weakened CI, catching duplicated utilities, tracing critical logic paths, validating security boundaries around LLM workflows, and requiring scoped changes with failing pre-change tests before approval.

SQL's ORDER BY clause now supports non-selected columns, expressions, explicit null ordering, and use inside subqueries in modern SQL standards. SQL has three relevant “orders”—syntax, logical evaluation, and actual execution—so developers should trust execution plans over simplistic infographics about clause order or “filter early” advice.

Wasp's founders now see their custom language as a mistake: developers liked the high-level full-stack framework idea, but the “new language” framing created adoption friction and made IDE/tooling support much harder than expected. The real value turned out to be Wasp's compile-time understanding of the whole app, so its team is replacing the custom DSL with TypeScript while keeping the underlying framework model intact.

Amazon Route 53 Resolver endpoints now support DNS64 on inbound endpoints and IPv6 forwarding over internet gateways on outbound endpoints, enabling IPv6-only clients and hybrid networks to communicate with IPv4 services without modifying existing infrastructure.

Kubernetes v1.36 has graduated Pressure Stall Information (PSI) metrics to general availability, giving users production-ready visibility into CPU, memory, and I/O resource contention at the node, pod, and container levels.