TLDR DevOps 2026-06-24

An internal developer platform can be a productivity multiplier and give developers flexibility in their infrastructure choices — but first, you need to make the right choices when it comes to building blocks.

The Manning book Platform Engineering on Kubernetes takes an in-depth look at the latest best practices and open source tools from the Kubernetes community. Use it as a sort of IKEA manual to build an open developer platform, including:

• Multi-cloud infrastructure management with Crossplane
• Using Dapr for standard API distribution
• Implementing feature flags with OpenFeature

Download the full Manning ebook from Chronosphere

As of Octopus Deploy 2026.2, self hosted customers can use GitHub Connections through the Octopus Deploy App for GitHub, replacing PATs with OIDC based short lived tokens that improve security, permissions management, and reliability. For instances behind firewalls or on private networks, Octopus supports externally hosted public signing keys, allowing GitHub to validate OIDC tokens without requiring inbound internet access to the Octopus server.

Amazon ECS service auto scaling now supports 20-second high-resolution CloudWatch metrics, enabling significantly faster response to workload changes with up to 76% faster scale-out and 72% faster end-to-end provisioning. It improves performance, reduces over-provisioning costs, and simplifies scaling configuration using target tracking policies across Fargate, EC2, and managed instances.

AWS Lambda MicroVMs is a new serverless compute service that provides VM-level isolation with near-instant startup times for running untrusted user or AI-generated code. Powered by the same Firecracker technology handling over 15 trillion monthly Lambda invocations, the service is now available in four regions, with support for up to 16 vCPUs, 32 GB memory, 8-hour runtime sessions, and automatic suspend/resume capabilities that preserve application state while reducing idle costs.

Amazon CloudWatch OTLP now supports direct metric ingestion via bearer token authentication, enabling Claude Code telemetry from developer machines without collectors or IAM setup and providing PromQL based dashboards, cost attribution, productivity analytics, and alerting. Organizations can track AI coding agent usage with flexible token granularity, low ingestion costs, and centralized observability across teams, accounts, and Regions.

A step by step guide shows how to build a FastMCP based math server, containerize it, and deploy it on GKE Autopilot with Gateway API, SSL, autoscaling, and health checks. The setup enables secure, centralized, remotely accessible MCP endpoints using Streamable HTTP, allowing teams to share AI tools and integrations without running local servers.

Netflix migrated millions of batch jobs from its homegrown Compute Managed Batch (CMB) system to Kueue, an open-source Kubernetes-native job queueing system, resulting in significantly improved resource utilization across its container platform Titus. The migration, which allowed operators to switch tenants with the click of a button, brought features like preemption-based fair sharing that let teams borrow idle reserved capacity from other tenants while ensuring high-priority workloads can still preempt lower-priority ones when needed.

kubectl logs | dstl8 | claude → fix - and that pipeline is real. Dstl8 distills your log firehose into signal a model can reason over, then hands it to your own agent via MCP. Root cause in, fix out. Kubernetes, CloudWatch, OpenTelemetry.brew install control-theory/dstl8/dstl8 && dstl8 setupRead the docs | Learn More

Firecrawl, an open-source web scraping and search API, launched enhanced features including an AI-powered "Agent" endpoint that can automatically find and extract web data without requiring specific URLs, plus integrations with MCP-compatible AI clients like Claude. The platform, available both as a hosted service at firecrawl.dev and for self-hosting under AGPL-3.0 license, converts web content into LLM-ready formats like Markdown and structured JSON while respecting robots.txt directives.

AirLLM, an open-source optimization tool, enables massive language models like 70B parameter models to run on a single 4GB GPU and 405B Llama 3.1 on 8GB VRAM without traditional compression techniques like quantization or pruning. The tool achieves this by decomposing models into layers and using block-wise quantization that speeds up inference by up to 3x while maintaining accuracy, with recent updates adding support for CPU inference, multiple model architectures including Qwen2.5, and MacOS compatibility.

Cloudflare engineers spent six weeks tracking down a race condition in the hyper HTTP library that caused intermittent failures when processing larger images through their Workers Images binding, ultimately fixing it with just four lines of code that ensured the connection buffer fully flushed before shutdown. The bug, which silently truncated responses (returning only ~200 KB of an expected 3.3 MB image with a 200 status), only surfaced after a December 2025 architecture change that replaced their FL intermediary with faster Unix sockets, inadvertently exposing a timing issue where hyper would discard flush status checks and prematurely shut down connections when socket buffers filled.

OpenAI launched Patch the Planet, a Daybreak initiative partnering with Trail of Bits to not only find vulnerabilities in critical open-source software but actually help maintainers fix them using AI-assisted security research and expert human review. The program has already identified hundreds of security issues and merged dozens of patches across nine initial projects, including cURL, Python, and the Go project, with Trail of Bits dedicating its entire security research organization to validate findings, develop patches, and coordinate disclosure before anything reaches overwhelmed maintainers.

Cycle is designed for complex, highly configurable supply chain environments, with built-in AI that turns business processes into executable tests. Plays nicely with ERP, WMS, TMS, OMS, and more. Start here

A scalable GitLab and GKE CI/CD system depends on deliberate GitOps design choices, especially pull-based reconciliation, separated application and environment repositories, and promotion via merge requests rather than branches.

Microsoft is deprecating the Azure DevOps issuer for workload identity federation service connections by July 1, 2027, requiring users to migrate to the standardized Microsoft Entra issuer that already powers over 50% of these connections.