TLDR DevOps 2026-02-13

Cloud cost optimization has become one more engineering headache. If you ever thought "there should really be an AI for this", you're in luck.

MilkStraw AI is the autonomous cost optimization agent. Connect it to your AWS account and it will find opportunities to cover on-demand usage with commitment-based discounts (in accounts managed by MilkStraw).

✅ Enjoy committed-use discounts without the actual long-term commitment.

✅ Zero access to your workloads.

✅ Zero downtime.

✅ Maximum flexibility.

See your savings estimate 👀

Markdown for Agents is a new feature that allows Cloudflare's network to automatically convert HTML content to markdown on the fly for AI systems requesting `text/markdown` via an `Accept` header. This significantly reduces token usage and processing costs for AI agents. It is available in beta for Pro, Business, Enterprise, and SSL for SaaS customers.

Entire is a newly launched company backed by a $60 million seed round that aims to build an AI-native developer platform that reimagines the software development lifecycle for agent-driven coding. Its first open-source product, the Entire CLI, introduces “Checkpoints,” a Git-integrated system that captures and versions full agent session context (prompts, reasoning, tool calls, and metadata) alongside commits to enable traceability, multi-agent coordination, and durable shared context.

Autonomous AI agents like OpenClaw can execute code, commits, and deployments at scale, accelerating delivery while amplifying risk. As agentic coding goes mainstream, governance, auditability, and clear authority models become essential to safely scale AI-driven software delivery.

AWS Control Tower adds a CloudFormation Hook Invocation Summary page that exposes logs and guidance for proactive control failures, reducing troubleshooting time. This post presents an example of an S3 public access control that shows how violations are identified, explained, and resolved before resource creation.

A bug in go-git's incomplete .gitignore implementation exposed the complexity and underspecified nature of gitignore semantics, particularly around anchoring, layering, negation, and doublestar behavior, leading to the creation of a fully compatible wildmatch-based engine. Many tools that claim to use “gitignore syntax” implement only partial or divergent subsets, and the ecosystem lacks a formal specification and shared test suite to ensure consistent behavior across implementations.

Tambo AI is an open-source generative UI toolkit for React. Connect your components—Tambo handles streaming, state management, and MCP.

Sidecar is a development tool designed to integrate the entire coding workflow into a single terminal shell, consolidating AI agent interaction, diffs, task management with TD, and workspace management. It aims to streamline planning, monitoring, and reviewing agent-written code without developers leaving their terminals.

Low-bit inference and quantization techniques are widely adopted to make large AI models, such as Kimi-K2.5 (1 trillion parameters), significantly faster and cheaper to run by reducing their memory and compute requirements. The article explores various quantization formats, including pre-MXFP methods and newer hardware-supported MXFP and NVFP standards, highlighting their crucial role in optimizing performance and energy efficiency for production systems like Dropbox Dash.

Vertex AI improved inference serving by adopting GKE Inference Gateway, which uses load-aware and content-aware routing to reduce latency and boost efficiency. In production, it delivered faster responses, better tail latency, and doubled prefix cache hit rates.

Azure NetApp Files Elastic ZRS delivers multi-AZ, zero data loss file storage with synchronous replication and automatic failover for mission-critical workloads.

Spec-driven development cannot replace coding because much of programming involves clarifying and discovering domain logic, not just transcribing it into a spec.

Scaling Kubernetes nodes from zero, while cost-effective, introduces significant startup latency, which can negatively impact user experience and CI/CD pipelines.