Claude Tag Brings Shared AI Agents to Slack

Anthropic has launched Claude Tag, a feature that lets entire teams @-mention a shared instance of Claude Code directly inside Slack. Think of it as adding an AI agent as a team member — you delegate tasks, it retains context from your Slack conversations, and keeps working while you move on.

This is a meaningful shift from individual developer tooling toward team-level AI coordination. The agent can be looped into threads, assigned work, and tracked asynchronously — no context-switching required.

Gemini 3.5 Flash Gets Computer Use

Google has added computer use capabilities to Gemini 3.5 Flash, enabling it to control browser, mobile, and desktop environments. A GitHub repo is already available for developers to test it locally or through Browserbase.

This puts Gemini in direct competition with Claude's computer use and OpenAI's Operator for agentic, environment-controlling tasks.

Figma Config: Motion, Shaders, and Vibe Coding

Figma dropped a dense set of updates at Config this week:

  • Design-to-code: Convert design layers directly into production code
  • Figma Motion: A dedicated tool for motion design workflows
  • Editable shaders: Generate and tweak GPU shaders inside Figma
  • Vibe coding for plugins: Natural language plugin development
  • Third-party agent connections: Figma Agent can now connect to external services

For startups investing in product design, the design-to-code pipeline alone could dramatically compress the gap between mockup and shipped UI.

Notion Opens Up to External Agents

Notion's new developer platform adds two major capabilities: running code-based workflows natively, and integrating external agents — including Claude Code, Cursor, and Codex — so they can operate from shared docs and task boards.

This positions Notion less as a note-taking app and more as an AI-native project operating system.

OpenAI Builds Its First Custom AI Chip

OpenAI has taped out its first in-house AI chip, codenamed Jalapeño, built in partnership with Broadcom. The chip is designed specifically for LLM inference powering ChatGPT, Codex, and the API — and future agent products.

This move mirrors Google's TPUs and Meta's MTIA strategy: owning the silicon layer reduces dependency on Nvidia and cuts long-term compute costs.

Other Notable Launches

  • GPT-5.5 Instant updated for better intent handling, constraints, and conversational feel
  • AssemblyAI Universal-3.5 Pro Realtime uses the agent's side of a call as speech-to-text context
  • Modal Auto Endpoints: Run open models in production with a single command
  • Harvey Labs: Legal foundation models with open evals and firm-owned intelligence
  • Engram: A new lab building personal models that learn from your work and update daily
  • John Jumper, lead of AlphaFold at DeepMind, is joining Anthropic