Codex overtakes ChatGPT as AI splits workforce

OpenAI says Codex now handles 85% of its workers' AI output, eclipsing ChatGPT as the default tool across every department. The shift shows AI‑driven knowledge work moving from quick queries to long‑running, autonomous agents that can tackle multi‑hour tasks. This hints that future product teams will spend more time defining problems and less on manual implementation.

Ant Murphy shows why common metrics like MAU mislead product teams and offers a two‑question framework to turn any number into a decision‑making tool. By linking metrics to concrete user actions and the change they trigger, teams can cut noise and focus on outcomes that move the business.

Generic “tell a story” advice leaves creators guessing. The post breaks that down, defining a story as a sequence that elicits a specific emotion and proposing a concrete challenge‑resolution framework to structure presentations. Applying this method lets product and design teams consistently craft talks that move audiences, not just string facts together.

The ‘spaghetti table protocol’ asks a model to reorganize a complex UI element within 15 minutes. Pilot runs on GPT‑4o, Gemini and others expose consistent failures in structural reasoning and handling creative constraints, warning designers not to over‑trust current AI tools.

Astryx, Meta's internal design system now public, ships 150+ accessible, themeable React components with a CLI and StyleX‑based styling that any developer can drop into a project. Because the API, docs, and CLI are agent‑ready, AI coding assistants can scaffold UI exactly like humans, speeding up prototyping and reducing hand‑off friction.

Lenny’s 2026 sentiment survey reveals a tech workforce divided by AI. One camp feels amplified and more productive, the other feels shaken, leading to a jump in burnout from 45% to 56% and a drop in career optimism. Over half would steer newcomers away from tech despite personal optimism.

Ben Thompson drafts a mock earnings‑call script for Mark Zuckerberg, showing how Meta should frame its $10‑plus billion AI spend, Reality Labs losses, and competitive positioning. The narrative pushes a long‑term AI vision while defending current capex, giving investors a clearer story on Meta’s growth roadmap.

Fortress is an open‑source Chromium fork that rewrites fingerprint signals inside the engine, so bot detectors see a regular Chrome browser. It clears challenges from Cloudflare Turnstile, CreepJS, Sannysoft and BrowserScan without any code changes to your Playwright or Puppeteer scripts.

Polygres adds native graph traversal and vector search to any existing Postgres instance, turning relational data into a GraphRAG engine without migration. It lets LLM agents retrieve context‑aware information by combining SQL, relationships, and embeddings in a single query, speeding up retrieval‑augmented generation for teams that already rely on Postgres.

Kirki lets you design WordPress sites on a freeform infinite canvas without any third‑party plugins. The built‑in features keep code lean, improving load speed while giving designers pixel‑perfect control. Its free tier and affordable Pro plan make it a viable alternative to heavyweight page‑builders.

Meituan released LongCat‑2.0, a 1.6‑trillion‑parameter mixture‑of‑experts LLM that activates only ~48 B parameters per token. The entire training run used Meituan’s proprietary AI ASIC superpods, marking a rare shift from GPU‑dominated large‑scale model training and showing China’s push for hardware independence.

Backlog is an open‑source SQLite‑backed manager that stores tasks, plans, and context for AI coding agents outside the chat. By persisting only the needed context each sub‑agent reads, token usage drops from 500k to under 50k, enabling parallel agents and 10× cheaper workflows.