TLDR DevOps 2026-05-18

Most engineering teams start on GitHub Actions or Jenkins.

Then the monorepo gets real, agents 5-10-50x the commit volume... and a flaky test sparks a Friday-afternoon outage.

Shopify, Pinterest, Block, Airbnb, OpenAI and Canva all run their CI on Buildkite.

We're built for the teams that need control over what runs where.

Try the lot for 30 days, zero commitment, no credit card. There's a real engineer named Ola on standby if you get stuck.

See what's included →

incident.io launched a PagerDuty Rescue Program offering contract buyouts, AI-assisted migrations, and a 99.99% uptime SLA to help companies replace PagerDuty with its AI-driven incident management platform.

AWS Bedrock Advanced Prompt Optimization is a new tool that lets developers optimize and compare prompts across up to 5 models simultaneously using a metric-driven feedback loop with support for multimodal inputs, including images and PDFs. The service is now available in 14 regions worldwide. It charges users based on standard Bedrock model-inference token rates consumed during the optimization process.

Amazon ECS now supports linear and canary deployment strategies using weighted target groups, CloudWatch alarms, and deployment circuit breakers to enable gradual traffic shifting with automated rollback on failures. Linear deployments shift traffic in fixed increments with bake times, while canary deployments test small traffic slices first, improving deployment safety, observability, and control across microservices.

Managing Kubernetes at scale creates YAML configuration sprawl, cluster drift, and compliance challenges across multicloud environments. GitOps centralizes declarative configurations in Git repositories, enabling automated reconciliation, policy enforcement, and consistent cluster management without manual intervention.

Bun's Rust rewrite may reduce memory-safety issues, but the bigger concern is maintainability: a large AI-generated port with little human review can pass tests while still hiding untested invariants, edge cases, and future debugging traps. This is not a simple “Zig failed, Rust wins” story, but a bet on whether AI-translated, insufficiently understood production code can be safely maintained over time.

Enter: AURI. It connects directly to your AI code editor of choice through MCP, giving agents the security context they need to write safe code, so you can ship secure by default.[Try for Free] Get security fixes inline, right where you're working →

Semble is a code search library built for agents. It returns the exact code snippets they need instantly, using ~98% fewer tokens than grep+read and cutting latency on every step. Indexing and searching a full codebase end-to-end takes under a second, with ~200x faster indexing and ~10x faster queries than a code-specialized transformer, at 99% of its retrieval quality.

Dream Server is an open-source, one-command deployment platform that lets users run a complete AI stack — including LLM inference, chat, voice agents, RAG, and image generation — entirely on their own hardware without cloud dependencies or subscriptions. The Apache 2.0-licensed project supports Windows, macOS (M1+), and major Linux distributions with automatic GPU detection and model selection, while its modular architecture has already attracted contributions from over 30 developers who've built features ranging from multi-GPU support to Intel Arc compatibility.

CodeGraph is a new open-source tool for Claude Code that uses pre-indexed knowledge graphs to help AI agents navigate codebases 92% more efficiently with 71% faster exploration times, eliminating the need for token-heavy file scanning. The MIT-licensed tool runs entirely locally, automatically syncing code changes via file watchers and storing symbol relationships, call graphs, and code structure in a SQLite database that Claude Code's agents can query instantly instead of using grep and file reads.

BuildBuddy is using content-defined chunking in its remote cache so large build outputs can reuse unchanged byte chunks instead of re-uploading or re-downloading entire files after small code changes. Early results show major savings for Bazel builds—about 40% less uploaded data in a benchmark and hundreds of TiB of duplicate chunk uploads skipped in production—with support available via Bazel's experimental remote cache chunking flag.

Docker's Custom Catalogs and Profiles for managing Model Context Protocol (MCP) servers allows organizations to curate approved collections of AI tools and distribute them as portable OCI artifacts while enabling developers to create reusable, named configurations of MCP servers for different workflows. The feature, available in Docker Desktop 4.56 and 4.63, lets teams push catalogs to container registries like Docker Hub and share server setups across projects without vendor lock-in.

AWS launched full repository code review for AWS Security Agent, enabling AI-driven analysis of entire codebases to detect architectural and data flow vulnerabilities that traditional static analysis tools often miss.

Kubernetes v1.36 added a new alpha metric called `route_controller_route_sync_total` to help operators measure the impact of the watch-based route reconciliation feature introduced in v1.35.