How we scaled metrics aggregation by moving durable state into storage
Datadog details the evolution of its metrics aggregation platform to scale with high data volumes. The post explains transitioning from fixed partition ownership to transferable state and implementing storage-side reconciliation.
Datadog · 21 November 2026 · Read the original →
When Can You Ship on Offline Evals Alone?
Spotify discusses the decision framework for when teams building AI features can safely bypass slow live A/B tests. It examines the trade-offs of deciding whether to ship directly based on offline evaluation metrics.
Spotify · 21 November 2026 · Read the original →
De-identifying text at scale: How Airbnb protects personal data in AI training
Airbnb explains its system for identifying and scrubbing personal data from user messages, reviews, and customer support chats. This de-identification process sanitizes sensitive user information before text data is ingested into AI models for training.
Airbnb · 21 November 2026 · Read the original →
Rebuilding Guardian: How Pinterest Upgraded Its Real-Time Abuse Platform to Modern Infrastructure…
Pinterest describes rebuilding Guardian, its platform for real-time abuse detection, on modern infrastructure. The post details how the system processes real-time event data at scale to generate insights and operational actions.
Pinterest · 21 November 2026 · Read the original →
How Meta Uses AI Agents to Scale ML Innovations Across Ads Ranking Models
Meta outlines how its autonomous Ranking Engineer Agent (REA) propagates machine learning optimizations across its ads ranking fleet. The agent automates hypothesis generation, experiment design, and architectural deployment across diverse models.
Meta · 21 November 2026 · Read the original →
Building the Agent Host Protocol for sessions that continue across devices
GitHub details the creation of the Agent Host Protocol for GitHub Copilot. The protocol allows long-running AI agent tasks to maintain session state continuously across different application windows and devices.
GitHub · 21 November 2026 · Read the original →
How Rakbank built centralized observability at scale with Amazon OpenSearch Service
Rakbank implemented a multi-account telemetry pipeline ingesting 150 GB of data per day using Amazon OpenSearch Service. The architecture uses hot, warm, and cold storage tiers combined with S3 retention for long-term compliance.
AWS Architecture · 21 November 2026 · Read the original →
Half these systems exist to observe or sanitize what the other half generates.
7 stories, every Saturday
Published here every week. Follow by RSS to get it as it lands.