AI Tools

Google is not just adding Gemma 4 to Android Studio. It is linking local coding, AICore prototyping, and future Gemini Nano 4 phones into one Google-controlled path.

OpenAI split Codex from the bundled ChatGPT seat, adding Codex-only seats, token billing, credits, and spend controls that make small team pilots easier to approve.

OpenAI Codex plugins are less a feature than a distribution move: Codex now has a curated install surface for reusable bundles of skills, apps, and MCP servers.

Anthropic's official plugin surface is live, and GitHub is suddenly full of guides, wrappers, and team layers built around Claude Code instead of merely using it.

Veo 3.1 Lite is less a flashy model launch than a cheaper video line item inside Gemini API and AI Studio. Google is turning AI video into normal developer spend.

A public `cli.js.map` in Claude Code release 2.1.88 exposed 4,756 source paths, turning one npm packaging mistake into a readable Anthropic roadmap leak.

A fast-rising GitHub repo turns months of Claude Code verbosity complaints into one drop-in CLAUDE.md patch that makes the assistant terser.

OpenAI didn't just add another Codex feature. It shipped a first-party Claude Code plugin, chasing developers in the terminal workflow they already use.

Anthropic says Claude Code burns 5-hour limits faster at peak times. The bigger problem is the bug-like drain reports turning usage policy into a trust issue.

March 28's agent story was a boundary story: Stanford's jai, Product Hunt browser tools, and Reddit builder chatter all pointed toward sandboxes and tighter approvals.

Fresh Product Hunt activity, a late-week HN maker post, and new browser-control docs point to the same shift: Claude Code is growing a browser layer.

GitHub now trains on Copilot interaction data from Free, Pro, and Pro+ users by default, while Business and Enterprise accounts stay outside the training pool.