Ihor

oh finally, checking the status of my services from my phone will be so much nicer

OpenAI launched gpt-realtime-translate. An API for live speech-to-speech translation: speak in one language, get translated audio + text on the other side. Calls, support, lessons, conferences, video rooms — all of this can now be translated directly inside your product.

first MCP that actually changed how i work, not just another connector @mobbin plugged 600k screens from real apps into claude/cursor. ask for a paywall — it pulls 43 examples from revolut, uber, duolingo and builds from what works, not from imagination

Agents SDK is turning into the execution layer around the model. OpenAI added native sandboxing so agents can inspect files, run commands, edit code, use MCP, skills, AGENTS.md, shell, and apply patch in controlled environments. Python first. TypeScript later.

AI agents don’t need better taste. They need better references. Lazyweb gives Claude Code, Codex, and Cursor access to 257k+ real app screens through MCP, so your agent can inspect a UI, find real patterns, and suggest fixes grounded in examples. Design context is moving into the IDE.

Notion Agent can now operate on your inbox. Connect Gmail through Notion AI connectors and it can search mail, draft/send replies, archive/trash messages, manage labels, and unsubscribe from senders. Write actions still ask for confirmation.

AI agents get more interesting once they can work after you close the tab. Manus Cloud Computer gives an agent a persistent Ubuntu machine in the cloud: scrapers, bots, databases, and reports running 24/7 while you sleep or take a vacation. manus.im/blog/manus-clo…

Meta Ads now has an MCP endpoint. AI agents in ChatGPT or Claude can pull reporting, inspect signal health, manage catalogs, and create/edit campaigns from natural language. Write actions still require authorization. facebook.com/business/help/…

Google quietly shipped skills for AI agents. ⭐️ 5.8K stars on GitHub. google/skills gives agents ready context for Google Cloud + Gemini API: BigQuery, Cloud Run, Firebase, GKE, Cloud SQL, AlloyDB. Install: npx skills add google/skills github.com/google/skills

This is a good Codex use case: not building a demo app, just cleaning up an existing repo. Ask Codex to upgrade it to GPT-5.5, attach OpenAI Docs as context, and let the agent do the migration pass through code, config, and docs.

Claude Code v2.1.121 is not a flashy release. It is what happens when plugins, skills, and MCP become daily workflow infrastructure: alwaysLoad for MCP servers claude plugin prune /skills search MCP auto-retry memory leak fixes Changelog: github.com/anthropics/cla…

Coding agents need process, not more vibes. ⭐ 168k on GitHub. Superpowers is a skills framework + software development methodology for Claude Code, Codex, Cursor, OpenCode, Copilot CLI, and Gemini CLI. It makes the agent clarify the task, write a spec, plan work, use git worktrees, follow TDD, run subagents, review code, and finish the branch cleanly. Not just a prompt pack. github.com/obra/superpowe…

ASO, but as agent skills. ⭐ 1.1k on GitHub. aso-skills gives Cursor, Claude Code, and other AI agents workflows for keyword research, metadata optimization, competitor analysis, Apple Search Ads, ratings, and App Store metrics. Nice for indie devs doing growth inside the IDE.

First plugin I’d install in Claude Code. 17.9K⭐ on GitHub, 73,178 installs on Claude. Claude Code Setup scans your codebase and tells you which hooks, skills, MCP servers, subagents and slash commands actually fit your project. Good starting point before building a setup by hand.