Explore what's new in AI

Gemini app lands on Mac with global shortcut and window sharing

The desktop app gives Gemini a native Mac surface, instant Option+Space access, and window sharing for code or docs, which makes it much more workflow-native.

Open Agents adds forked chats for shared agent workspaces

Forked chats let teams branch an in-progress agent conversation, which is a simple but powerful primitive for parallel work and reviews.

Visual Studio Code 1.116 adds agent debug logs and terminal tools

VS Code is making agent debugging more observable, while tightening Copilot integration and multi-tool approval flows for real work.

Lovable ships a desktop app with local MCP access and project tabs

The desktop app adds local MCP access and multi-project tabs, which makes the builder feel more like a native workstation than a thin wrapper.

Hermes Agent adds QQBot and AWS Bedrock support

Hermes Agent broadens its integration surface with a China messaging adapter and a native Bedrock provider, which makes the stack more deployable.

Warp lets Codex auto-run commands and create worktrees from tab configs

The new tab configs push more repo setup into agent startup, which reduces repetitive shell work when spinning up new coding tabs.

Anthropic shows how Claude Code session management avoids context loss

The guide turns rewinding, compaction, subagents, and fresh sessions into a workflow, which matters more as contexts get large and long-running.

Zed 0.232 makes file search order-agnostic and improves tab focus

The file finder now matches terms in any order, and focus-follows-mouse makes panel hopping less annoying during rapid code navigation.

Firecrawl packages /scrape, /search, and /interact as agent skills

By auto-wiring the right endpoint and verifying it, the skills cut the glue work needed to add search and browser actions to an agent.

Microsoft Copilot delegates emails end-to-end from Outlook

Users can now forward an email and let Copilot extract tasks, execute them, and notify when it's done, which is a real step toward autonomous inbox work.

HoloTab turns a Chrome extension into a teachable computer-use agent

The extension can record a routine, schedule it, and execute it later, which moves agentic browser automation from demo to repeatable workflow.

TanStack replaces flaky smoke tests with 147 deterministic LLM E2Es

TanStack uses a mock layer to test seven providers in about two minutes, cutting token spend and making cross-provider regressions repeatable.

Image to WebDev ranks models on screenshot-to-site generation

This arena tests a practical frontier task, turning mockups into working sites, and it shows the current leaders for visual-to-code agent work.

AiScientist uses file-backed state to run long-horizon ML research

The system keeps plans, logs, and evidence in durable workspace files, and that design lifts PaperBench and MLE-Bench Lite versus weaker baselines.

OpenCode adds a sync fence to stop stale reads in remote envs

A control-plane fence now blocks reads until remote state catches up, fixing eventual-consistency bugs that could show stale session data after writes.

OpenCode adds Qwen3.6 Plus and 3.5 Plus support in Go

OpenCode users get new Qwen models in Go, with image support and zero retention. Alibaba also had to pull them offline briefly for capacity.

Claude Doctor scans ~/.claude and writes rules for CLAUDE.md

The tool turns your own session history into guardrails, which is a practical way to stop repeat failures and encode better habits into Claude.

Claude Code 2.1.110 fixes MCP hangs, permissions, and remote control

This release targets the rough edges that slow sessions down, from MCP disconnects and retry hangs to permissions checks, remote control, and plugin installs.

Anthropic redesigns Claude Code desktop with terminal and side chats

Anthropic's desktop client now behaves more like an IDE, with an embedded terminal, in-app edits, and parallel side chats to keep agent work flowing.

Cursor adds canvases for interactive agent-generated dashboards

Cursor can now turn responses into visual canvases instead of plain text, making it easier to inspect data, dashboards, and custom UIs in one place.

Google tests Gemini Live Mode with screen sharing on desktop

The leak points to desktop screen sharing and a live overlay, which would turn Gemini from a chat box into a more hands-on desktop assistant.

Gemini expands Personal Intelligence to more users and Chrome

Google is rolling out app-aware personalization beyond paid tiers and into Chrome, which makes Gemini more context-rich but also deeper in your data.