Technology industry news, commentary and analysis, with reporting on big tech, startups, and internet culture.
The New York Times replaced a batch paywall model with real-time, causal ML-driven decisioning that uses S-learner predictors trained on an always-on RCT, convex multi-objective optimization to balance KPIs under business constraints, IPW for impact estimation, and frequent production tuning via their ML platform—resulting in improved subscription and registration outcomes.
NYT Games implemented Dark Mode by creating a design system and token libraries, scoping the rollout to the Games app, and incrementally applying palettes per game via a React Provider and hooks. The system uses React + CSS modules, CSS variables for tokens, TypeScript, Vite, and Storybook, with Playwright for visual regression testing and linting/git-hook tooling to enforce quality. The effort prioritized user-facing delivery while paying down design debt and produced measurable engagement gains.
A NYT Cooking design case study describing how the team used "provocations" (rough, high-fidelity design concepts) to guide an incremental redesign of recipe "cards". The goals were to prioritize photography, simplify visual hierarchy, and avoid scope bloat by codifying principles and shipping small conservative changes (remove card containers, change background, group recipe metadata, round photo corners). The approach improved discovery and sped up recipe selection according to early data. Author: Jayne Lee, lead product designer for NYT Cooking.
NYT engineers built the Connections Reference Dashboard to help the Connections puzzle editor manage and search historical puzzles. Backend: flask-based service that fetches from the NYT Connections API, caches puzzles to disk, and preloads future puzzles. Frontend: a responsive Vue.js application using the Composition API, axios for API calls, and UnoCSS for styling, laid out with CSS Grid/Flexbox. Features include regex search, word-frequency counts, duplicate detection, and filters to surface unpublished boards. The architecture is modular and designed for future enhancements.
Heidi Erwin describes her process for creating The New York Times' weekly "Brain Ticklers" puzzles — covering inspiration sources, drafting, playtesting and editorial review with a puzzle editor, and producing simple graphics (made in Figma) for the newsletter and print.
NYTimes product managers describe making accessibility a non-negotiable requirement by investing in people (internal groups and champions), process (training, acceptance criteria, audits), and product (an internal accessibility guide aligned to WCAG 2.2 AA and an audit tool).
The New York Times migrated its React unit tests from Enzyme to React Testing Library while upgrading from React 16 to React 18. They chose a piecemeal migration strategy, show code examples converting Enzyme tests to React Testing Library equivalents (rendering, clicking, image onload), and emphasize minimizing disruption and enabling collaboration.
NYT describes how it augmented manual editorial curation with "editorially‑driven" algorithmic programming for home page modules. Their pipeline (pooling, ranking via contextual bandits/reinforcement learning, and finishing with editorial guardrails) combines editorial signals and personalization. They built editor tools (browser extension, alerts, dashboards), features (pinning, exposure boosting, smart refreshing, exposure minimums, already‑read/seen filters) and relied on A/B testing and analytics to tune tradeoffs. The result scaled editorial workflows and increased engagement for prioritized modules.
NYT migrated its flagship React-based news site to React 18 to improve page performance (INP/Core Web Vitals) and SEO. The work required removing Enzyme and migrating tests to @testing-library/react, upgrading React-related dependencies and switching to createRoot/hydrateRoot. A major challenge was hydration mismatches with server-rendered embedded interactives injected via dangerouslySetInnerHTML; the team solved this by extracting scripts and re-attaching them client-side in controlled order. The migration yielded ~30% p75 INP improvement and roughly half the previous re-renders.
The post is a first-person account of a product design internship at The New York Times focused on internal tools and design systems. The author worked on Ink, the Times’ in-house design system, creating centralized Figma resources, writing component documentation, and designing a dashboard to track component health and usage. The piece describes collaboration within a cross-functional team and exposure to user research and newsroom workflows. It also covers continued work on Moderator, the Times’ comment moderation tool, including designing editorial workflows and features like Times Picks and Reporter Reply. The article does not provide implementation-level technical details such as languages, frameworks, or architecture diagrams.
NYTimes describes their effort to standardize experimentation using their internal ABRA platform and Stats Engine. The post focuses on metric governance (keeping metric definitions outside the experimentation platform), standard statistical reporting (daily t-tests, point estimates, CIs, p-values, MDE), special calculations (scaled impact, login-rate effects), support for off-platform analyses (a Python notebook and a bootstrapping pipeline for quantile metrics), and the cultural work (training, collaboration, governance) required to adopt standardized experimentation.
The NYTimes Growth Data team built a neutral multi-touch subscription attribution model anchored on financial subscriber reporting data. They standardized dimensions (product, channel, platform, paywall impressions), unified front-end event tracking across paywalls, and mapped those dimensions to subscriber records by resolving joins across systems. The model was developed iteratively (MVP approach) and enabled improved reporting, churn analysis, and layered audience insights.
NYTimes Games reworked its reporting by replacing ad-hoc queries with a layered aggregation data architecture, standardized per-game tables, version-controlled code (GitHub), a Slack triage workflow, reusable dashboards, and data observability/alerting; the changes increased dashboard usage and query speed and they are migrating from proprietary pipelining tech to a cloud-based solution.
