AI Tools Daily

Apple's WWDC keynote landed — a rebuilt Siri with its own app and Apple Foundation Models powered by a Google Gemini collaboration. Plus Cursor 3.7's Context Usage Report (and why your context BUDGET, not just the window, is what slows agents down) and Windsurf becoming Devin Desktop with the Agent Client Protocol. We break down ACP vs MCP, the Claude Code "ultracode" rename, and the MiniMax M3 weights watch. Hosts: Alex & Jules. New episodes daily.

MiniMax releases M3, an open-weight model with a 1M-token context window — and we teach how sparse attention makes that affordable. Plus Claude Code 2.1.166 adds fallback models and strips user authority from agent-to-agent messages (the confused deputy problem, explained), and Cursor 3.7's Context Usage Report finally shows where your tokens go. Hosts: Alex & Jules. New episodes daily.

MiniMax launches M3, an "open-weight" frontier coding model with a 1M-token context — but the weights aren't out yet and the benchmarks are self-reported. Plus: GitHub Copilot CLI's new adversarial /rubber-duck command and scheduled prompts, Claude Code's security-hardening release wave, and Cursor's new tiered Teams pricing. Hosts: Alex & Jules. New episodes daily.

Windsurf got an over-the-air update and opened as Devin Desktop — Cognition unified its cloud and local agents under one IDE with ACP support for Claude, Codex, and more. Plus: what Claude Opus 4.8's Dynamic Workflows actually unlocks for multi-agent builders, Claude Code's new auto-loading plugin system, and GitHub Copilot's $100 Max plan. Hosts: Alex & Jules. New episodes daily.

Microsoft Build dropped a wave of GitHub Copilot CLI updates yesterday: a Rubber Duck adversarial review agent, /every and /after prompt scheduling, local voice input, Gemini model support, and a GA SDK for building Copilot extensions. Plus a breaking change in Claude Code's changelog that silently kills workflow hooks — and a community knowledge graph tool claiming a 120x token reduction over standard codebase exploration. Hosts: Alex & Jules. New episodes daily.

GitHub Copilot's billing flips to AI Credits today — no more fallback model when you run out — and we break down exactly what changes and what to check. Plus: Claude Code now auto-loads plugins from .claude/skills directories (no marketplace needed), auto mode lands on Bedrock/Vertex/Foundry, Windsurf's SWE-1.6 is free for three months with meaningful agent UX improvements, and Hugging Face's SmolLM3-3B brings reasoning to the 3B parameter class. Hosts: Alex & Jules. New episodes daily.

Claude Code dropped two major updates overnight: Opus 4.8 with high-effort defaults and a new dynamic workflows system that orchestrates tens to hundreds of background agents in parallel. We also cover Zyphra's open-source MoE diffusion model hitting 7.7x inference speedup, and LiteLLM shipping a Kubernetes-based self-hosted agent platform for production teams. Hosts: Alex & Jules. New episodes daily.

Anthropic ships self-hosted sandboxes and MCP tunnels for Claude Managed Agents — letting agents run tool execution on your own infrastructure and reach private networks without exposing endpoints. Claude Code v2.1.152 adds `/code-review --fix` to auto-apply review findings to your working tree. Cursor launches Composer 2.5, a frontier-level in-house model matching Opus 4.7 benchmarks at one-tenth the API cost, paired with Build in Parallel for dependency-aware async execution. Plus: Kimi K2.6's 300-agent swarm beats GPT-5.5 in a live coding challenge. Hosts: Alex & Jules. New episodes daily.

Google officially retired Gemini CLI at I/O and gave free, Pro, and Ultra users until June 18 to move to Antigravity CLI — but the switch from open-source to closed-source has developers frustrated. We cover that, plus a Claude Code `/code-review` command that now posts inline GitHub PR comments, a critical macOS crash fix in Claude Code, and GLM-5.1, a new open-weight coding model with an MIT license and 200K context window. Hosts: Alex & Jules. New episodes daily.

Anthropic acquired Stainless — the SDK generator used by OpenAI, Google, and Cloudflare — and is winding it down for everyone but Claude. We break down what that means for developers on both sides. Plus: Google's new Managed Agents API turns one API call into a full sandboxed Linux agent loop, xAI enters the coding CLI race with Grok Build, and the inevitable convergence of every AI coding tool into the same feature set. Hosts: Alex & Jules. New episodes daily.