AI Tools Daily

Windsurf ships SWE-1.5 — their first proprietary model at 950 tokens/second, now the free default. Google rebrands its entire agent developer platform and makes Claude a first-class citizen alongside Gemini. Claude Code's Agent Teams feature is doing real PR code review in production, and Anthropic's internal coverage jumped from 16% to 54%. Plus: a Claude Code bug that was quietly truncat...

Anysphere shipped Composer 2 — a coding model trained from scratch with custom CUDA kernels — signalling the end of the "IDE wrapper" era for AI coding tools. Also: Alibaba's Qwen3.6-27B beats a 397B model on coding benchmarks and runs on consumer hardware, and Claude Code's latest releases pack enterprise-grade security hardening including PID isolation and credential scrubbing....

Google flipped the switch on 50+ managed MCP servers across its entire Cloud platform — BigQuery, Maps, Kubernetes, and more — now available to any MCP-compatible coding agent. We also cover Google's A2A protocol reaching 150 organizations in production with new cryptographic identity features, Hugging Face's ml-intern agent that autonomously runs your full LLM training loop (and beat Claude Code...

Google DeepMind's Sergey Brin issues an internal memo calling for an urgent push to catch Anthropic in AI coding — and the gap data is striking. Plus: Claude Code's newest features including Auto mode and a permission shortcut that actually works, Cursor's BugBot is now self-improving across 110,000 repos, and OpenCode hits 147,000 GitHub stars growing 4.5x faster than Claude Code. Hosts: Alex & J...

Anthropic briefly pulled Claude Code from its $20 Pro plan — then reversed it after developer backlash. We break down what it signals about the future of flat-rate AI pricing. Plus: Hugging Face's new open-source ml-intern agent out-benchmarked Claude Code on scientific reasoning, Codestral 2 went Apache 2.0 (and why that matters more than the benchmarks), Google's A2A protocol hit v1.0 at...

Claude Opus 4.7's new tokenizer is quietly inflating agent costs by up to 35% — and task budgets are the fix. Plus: Google's Agent Development Kit hits stable v1.0 for Python with prompt-injection defenses, Cursor's BugBot now learns from your team's PR feedback, Roundtable MCP lets you run Claude Code, Codex, and Gemini in parallel with zero config, and the Chrome DevTools team sh...

Cursor 3.2 ships async subagents and multi-root workspaces for true cross-repo parallel coding. Claude Opus 4.6's 1M context window is now fully GA with no price premium and 128K output tokens — and the benchmark gap over previous models is staggering. Hugging Face launches ml-intern, an open-source agent that autonomously trains other AI models end-to-end. Plus: OpenCode hits 147K GitHub star...

Google rebranded Vertex AI as the Gemini Enterprise Agent Platform at Cloud Next 2026, introducing Agent Runtime, Memory Bank, and managed MCP servers for BigQuery, Maps, and more. The A2A protocol hits v1.2 with 150 orgs in production. Plus: the MCP spec now mandates OAuth 2.1 — what builders need to retrofit — and a Claude Code fix for large tool result truncation. Hosts: Alex & Jules. New e...

Claude Code v2.1.118 ships today with a game-changing MCP hooks update — tools can now fire directly from event hooks with no glue code required. Hugging Face releases ml-intern, an open-source AI agent that reads papers, trains models, and evaluates results end-to-end autonomously. Plus, Qwen-code hits 23.7k GitHub stars and v0.15.1, signaling that the open-source terminal agent race is very much...

OpenAI's Codex update brings background computer use, 90+ plugins, memory, and a built-in browser to Mac — turning it from a coding assistant into a full desktop agent. Meanwhile, GitHub ships `gh skill`, a single CLI command to install and publish agent skills across Claude Code, Cursor, Codex, Gemini CLI, and Copilot. Plus Cursor 3's Agents Window and Design Mode explained, Claude Code performan...

Cursor 3 ships a full agent-first interface with parallel cloud agents and a /best-of-n command for running the same task across multiple models simultaneously. Claude Code drops /ultrareview — a parallel multi-agent cloud code review — plus a native binary, Opus 4.7 xhigh effort, and PowerShell support for Windows. Meanwhile, Alibaba's Qwen3-Coder-Next gives local-first teams a serious open-sourc...

Claude Code shipped three versions this week — /ultrareview brings parallel cloud agent code review, Opus 4.7 xhigh lands, and the CLI switches to a native binary. Meanwhile Cursor 3's new Agents Window reshapes multi-agent orchestration in the IDE, OpenAI expands Codex into computer use and automation, and a viral Hacker News thread argues LLM function calls don't scale. Plus Gemma 4 from Google...

Claude Code v2.1.113 ships a native binary that breaks Android/Alpine setups — plus new /ultrareview multi-agent code review and Opus 4.7 xhigh effort. Cursor 3.1 adds tiled parallel agents. Google's Gemma 4 launches as Apache 2.0 open source with 256K context and tops the open model leaderboard. And Alibaba's Qwen3-Coder-Next hits 70%+ on SWE-Bench with 10x throughput for local use. Hosts: Alex &...

Claude Code dropped three versions in 48 hours — highlights include the new /ultrareview cloud-based multi-agent code review, the /less-permission-prompts skill, and Opus 4.7's new xhigh effort level. GitHub also shipped `gh skill` in CLI v2.90.0, a new command for installing and managing portable agent skills across Claude Code, Cursor, Codex, and Gemini. Plus: the ongoing reaction to Cursor 3's...

Cursor 3 ships an Agents Window that turns your IDE into an agent command center with parallel worktrees and Design Mode. OpenAI Codex adds computer use, an in-app browser, and image generation to its Mac app. Windsurf 2.0 embeds Devin cloud agents directly in the editor, Claude Code hits 30+ releases with 1M-token context for Opus 4.6, and MCP crosses 97 million installs under the Linux Foundatio...

Claude Code shipped six updates in three days — we break down what actually changed: full-screen TUI mode, 1-hour prompt caching, session recap, and PreCompact hooks. Plus: Alibaba's Qwen3.6-Plus just dropped with native tool-calling baked in and a 1M context window, and it's free on OpenRouter. We also cover the Chrome DevTools MCP that gives your AI agent live browser control, and why MCP moving...

Claude Code 2.1.108 ships a session recap feature and 1-hour prompt caching. Cursor 3 rebuilds the IDE around parallel agents with a /best-of-n command that runs tasks across multiple models simultaneously. Windsurf Wave 13 gives away its SWE-1.5 model for free. Plus: the AI coding tool stack is converging faster than anyone planned, and Claude Haiku 3 retires in four days — act now if you're on t...

Claude Code v2.1.105 dropped yesterday with worktree management fixes and new plugin hooks — Alex and Jules break down why the small changes matter for agentic workflows. Plus: the Agent Skills (SKILL.md) standard is now compatible with 12+ AI coding tools, MCP crossed 97 million installs, Cursor 3 is redesigning around agent-first workflows, and the open source model race heats up with Llama 4 Sc...

Cursor 3 rebuilds the IDE around parallel agents with a new Agents Window, Design Mode, and cloud-local handoffs. Google drops Gemma 4 — a 31B open model with multimodal support that outperforms models ten times its size under an Apache 2.0 license. Plus: Microsoft unifies Semantic Kernel and AutoGen into Agent Framework 1.0, Claude Code quietly changes its default effort level to HIGH, and Anthro...

Claude Code drops version 2.1.101 with team onboarding commands, cloud-auto-provisioned ultraplan, and voice mode. MCP hits 97 million installs and moves to the Linux Foundation as an industry standard. Cursor 3 ships parallel agent execution and an 80% Bugbot resolution rate. Plus: Haiku 3 retires April 19th (act now), a new MCP orchestration tool for routing across Claude/Cursor/Gemini/Codex, an...

Cursor rebuilt its entire IDE around AI agents with parallel execution and a new Design Mode. Google shipped Gemma 4, a fully open-source multimodal model under Apache 2.0 with native function calling for agentic workflows. Anthropic added cloud-based collaborative planning to Claude Code with Ultraplan, and MCP crossed 97 million monthly installs after being donated to the Linux Foundation's Agen...

Cursor 3 launches a full agent-fleet workspace — run parallel agents across local and cloud with seamless handoff. Google drops Gemma 4 open source under Apache 2.0, with a 31B model that hits 80% on LiveCodeBench. MCP crosses 97M installs and moves to Linux Foundation governance, effectively ending the protocol wars. Plus: Claude Code bumps default effort to high, a memory leak fix for long-runni...

Claude Code 2.1.0 ships a full Skills system, hook architecture, and permission wildcards across 1,096 commits — but Hacker News is sounding alarms about regressions on complex tasks. Plus: Godot MCP Pro drops 163 editor tools for game-dev agents, Google's Gemma 4 hits Hugging Face trending, and the MCP ecosystem crosses 97 million SDK downloads. Hosts: Alex & Jules. New episodes daily.

Anthropic launches Project Glasswing, deploying Claude Mythos Preview to 40+ partners including Apple, Microsoft, and Google for autonomous vulnerability hunting. Plus OpenAI ships GPT-5.4 with native computer-use, Claude Code gets Mantle-powered Bedrock and higher default effort, Cursor reveals real-time RL training every 5 hours, and Windsurf's pricing overhaul sparks backlash. Hosts: Alex &...

Cursor 3 launches with a new Agents Window for parallel multi-agent workflows and Design Mode. Claude Code v2.1.92 adds enterprise policy controls and a Bedrock setup wizard. Plus: the Anthropic Mythos/Capybara leak, MCP crossing 97 million installs, and the agent skills ecosystem hitting 350K. Hosts: Alex & Jules. New episodes daily.