19 September 2026

18 new posts across 8 engineering blogs — 7 worth your time.

One week of Engineering Reading, 7 stories, as published.

Saving another 100TB of RAM with math (and Rust)

Cloudflare shares how they optimized memory usage across their global network of thousands of servers. By leveraging mathematical optimizations and Rust, they managed to reclaim 100TB of RAM from services running on every node.

Why it matters — Even small, node-level memory optimizations yield massive resource savings when multiplied across a global, uniform server fleet.

Cloudflare · 19 September 2026 · Read the original →

Reliability

Worker Backpressure (Part 1)

This post details how Canva implemented backpressure mechanisms in their queue worker systems. It explains how they designed workers to automatically slow down ingestion when downstream dependencies fail, and safely ramp back up once those systems recover.

Why it matters — Implementing self-tuning backpressure in queue workers prevents cascading failures and system overload during downstream outages.

Canva · 19 September 2026 · Read the original →

Data & ML

The guest journey, updated in real time: extending Airbnb’s sequence recommender with Chronon

Airbnb explains how they extended their sequence recommender system using Chronon's Push Mode and Near-Real-Time (NRT) Model Transform capabilities. This architecture allows the platform to serve personalized search results instantly based on immediate guest actions, rather than waiting for offline batch processing.

Why it matters — Transitioning from batch-processed recommendations to real-time streaming features allows systems to capture and react to immediate user intent.

Airbnb · 19 September 2026 · Read the original →

Data & ML

Beyond Two Towers: Launching the 3-Tower Engagement Co-Train Model (Part 2)

Pinterest details the architecture and deployment of their 3-Tower Engagement Co-Train Model for lightweight ads ranking. The post covers how they moved beyond traditional two-tower models to improve engagement prediction accuracy.

Why it matters — Adding a third tower to traditional retrieval architectures allows models to co-train on multiple engagement objectives simultaneously.

Pinterest · 19 September 2026 · Read the original →

Data & ML

Beyond the model: Engineering AI infra with scientific judgement

The post discusses Airbnb's agent harness infrastructure, which is designed to process and analyze unstructured customer support data. It focuses on embedding scientific rigor, reproducibility, and auditability into AI agent workflows rather than relying solely on raw LLM outputs.

Why it matters — Building reliable AI agents requires wrapping LLMs in structured infrastructure that enforces scientific methodology, reproducibility, and audit trails.

Airbnb · 19 September 2026 · Read the original →

Developer Experience

Leave the Class Path in the Rearview Mirror

Netflix introduces new command-line tools designed to modernize Java development by moving away from traditional class path configurations. The tools leverage Java's native module system to create composable, agent-friendly environments that streamline build and dependency management.

Why it matters — Embracing native module systems over legacy class paths simplifies dependency management and improves startup performance for modern JVM applications.

Netflix · 19 September 2026 · Read the original →

Practice & Craft

AI Changed How Spotify Builds. What We Learned (and Fixed) About Quality at Higher Velocity

Spotify reflects on how the integration of AI tools has altered their software development lifecycle. The post shares concrete lessons and fixes they implemented to maintain high quality and reliability standards while operating at a much higher development velocity.

Why it matters — Increasing development velocity with AI tools requires updating quality assurance and testing frameworks to prevent regression and maintain system reliability.

Spotify · 19 September 2026 · Read the original →

Still chasing efficiency and stability, now with more AI.

7 stories, every Saturday

Published here every week. Follow by RSS to get it as it lands.

Subscribe by RSS All issues

← Previous issue