TLDR DevOps 2026-04-03

AWS DevOps Agent is generally available, providing autonomous incident resolution, proactive prevention, and SRE task automation across AWS, multicloud, and on-prem environments with enhanced integrations, intelligence, and customization. It reduces MTTR and delivers insights to improve reliability and prevent future outages.

Amazon ECS launched managed daemon support for ECS Managed Instances, letting platform engineers independently deploy and update monitoring, logging, and tracing agents across infrastructure without requiring application teams to redeploy services. The feature guarantees daemons start before application tasks, drain last, and supports rolling deployments with automatic rollbacks. It is available today in all AWS Regions at no additional cost beyond standard compute resources.

Docker Offload is a generally available cloud service that moves the container engine to Docker's secure cloud, enabling developers in restrictive enterprise environments like VDI platforms and locked-down laptops to run Docker Desktop without workarounds. The service maintains the same workflows and commands developers are familiar with while routing containers through an encrypted, SOC 2-certified infrastructure, available as an add-on to Docker Business with multi-tenant or single-tenant deployment options for regulated industries.

Harness engineering for coding agents focuses on building “guides” (feedforward controls) and “sensors” (feedback loops) around the model to improve output quality and enable self-correction, increasing trust in AI-generated code. While these harnesses can reduce errors and automate checks, they cannot fully replace human judgment, making the developer's role essential in steering, prioritizing, and defining what “good” looks like.

Using Atlantis with GitHub Actions centralizes Terraform execution into pull requests, improving visibility, consistency, and preventing state conflicts while enforcing approvals and locking. GitHub Actions adds validation, security scanning, and policy checks, creating a secure, automated infrastructure workflow at scale.

KernelEvolve is an AI agent system that autonomously generates and optimizes low-level hardware kernels across its diverse chip infrastructure (NVIDIA, AMD, and custom MTIA silicon), achieving a 60% improvement in ads model inference throughput in hours versus the weeks human experts would require. The system treats kernel optimization as a search problem rather than one-shot code generation, exploring hundreds of implementations while maintaining a self-evolving knowledge base that gets smarter with each optimization session and can even generate code for proprietary hardware like MTIA that doesn't exist in public training datasets.

Google Cloud has announced that llm-d has been accepted as a CNCF Sandbox project, co-founded with Red Hat, IBM Research, CoreWeave, and NVIDIA. The project aims to enable any model, any accelerator, any cloud, providing an open-source foundation for distributed AI inference without vendor lock-in.

GitLab's auto-dismiss vulnerability policies reduce security noise by automatically dismissing irrelevant findings based on defined criteria like paths or CVEs, improving triage efficiency and consistency. They preserve audit visibility while enabling scalable, policy-driven vulnerability management across projects.

Join Luciq on exploring how agentic AI is redefining what it means to observe, understand, and act on app performance. Come see what modern mobile observability looks like. SIGN UP

Cloudflare released results from an independent Big 4 accounting firm examination confirming that its 1.1.1.1 DNS resolver continues to meet all privacy commitments.

AI security risks increasingly target underlying cloud infrastructure rather than prompts, exposing sensitive data through misconfigurations, supply chain flaws, and runtime vulnerabilities.