TLDR DevOps 2026-02-20

VP: "Just have AI write all your Playwright tests, and you'll release features 10x faster."

Dev: "Sure, but who's accountable if something goes wrong?"

VP: "Still you."

Unreasonable? Yes. But when AI writes the code, who's actually accountable for quality?

This blog post from mabl talks about how most teams pair AI code assistants with Playwright and assume they're covered. But risks accumulate, and coded automation alone isn't enough; quality governance is required. Read the blog

Databricks' Predictive Optimization feature is now enabled by default for all new Unity Catalog managed tables after managing millions of production tables throughout 2025, delivering up to 22% faster queries and significantly reduced storage costs through automated maintenance. The company is expanding the feature in 2026 with Auto-TTL for automated row deletion based on time-to-live policies and a new Data Governance Hub dashboard that shows storage cost savings and optimization metrics.

Visual Studio Code v1.109 introduces multi-agent orchestration with support for Anthropic Claude and OpenAI Codex alongside GitHub Copilot, unified session management, parallel subagents, and MCP Apps. The release enhances context retention, security sandboxing, performance, and positions VS Code as a universal AI interface.

The Prometheus community is evolving its TSDB from classic primitive sample storage to native composite sample support for histograms and other types, improving efficiency, transactionality, and reliability. Ongoing work across OpenMetrics 2.0, Remote Write 2.0, and PromQL compatibility aims to enable transparent migration without breaking existing queries.

Nearly 30% of platform teams don't measure success at all, and another 24% can't determine if their metrics have improved, creating an accountability gap that threatens funding and growth, according to the State of Platform Engineering Report Volume 4. The report recommends structured measurement frameworks like DORA metrics (deployment frequency, lead time, MTTR, and change failure rate) and the SPACE framework to translate technical improvements into business value, with teams that fail to establish measurement practices by 2026 facing potential budget cuts.

An open source MCP server integrates GitHub Copilot with Azure App Service logs, enabling natural language debugging, log queries, deployment correlation, and root cause analysis directly in the IDE using existing Azure credentials. The proof of concept improves observability by combining real time tooling with embedded domain specific debugging guidance.

Docker Sandboxes launched a new shell sandbox feature that lets developers run AI agents like NanoClaw (a Claude-powered WhatsApp assistant) inside isolated microVMs with secure credential management. The shell sandbox provides a minimal Ubuntu environment with Node.js, Python, and Git pre-installed, allowing users to install any AI tool while keeping API keys protected through Docker's credential proxy system that prevents actual keys from existing inside the sandbox.

The 2025 Gartner Magic Quadrant for DevOps Platforms covered leading tools from Atlassian, Microsoft, and others. Not only did Atlassian earn a leadership spot for a third straight year, but it placed 1st in both Ability to Execute and Completeness of Vision. Get the report (free) to see why

Superpowers gives AI coding agents like Claude a structured workflow for software development, guiding them through design validation, test-driven development, and subagent-driven implementation that can run autonomously for up to two hours at a time. The framework works as a plugin for platforms like Claude Code and Cursor, automatically triggering composable "skills" that enforce workflows like breaking projects into 2-5 minute tasks, following strict red-green-refactor TDD cycles, and conducting two-stage code reviews before progressing.

Salt is a systems programming language that delivers C-level (or better) performance with compile-time, Z3-verified safety contracts and arena-based memory management—no GC or borrow checker. It uses MLIR codegen and cross-package verification to build high-performance systems with minimal runtime overhead and built-in tooling.

Choosing a programming language based solely on its surface-level syntax misunderstands what truly matters: semantics, type systems, and overall language design. Syntax affects ergonomics and readability, but it reflects deeper semantic decisions, and experienced programmers prioritize those foundations over aesthetic preferences or first-exposure bias.

Amazon CloudWatch's Alarm Mute Rules allow teams to temporarily silence up to 100 alarms during deployments or maintenance while preserving visibility. Expired rules automatically trigger pending actions if states persist, reducing alert fatigue without risking missed critical issues.

Kyverno 1.17 promotes its Common Expression Language (CEL) policy engine to stable v1 status while officially deprecating the legacy ClusterPolicy and CleanupPolicy types that used JMESPath.

Docker does not rotate container logs by default, which can cause log files to grow indefinitely and fill disk space.