AI Coding Tools Week 14: New Critic Agent & Major Interface Overhaul

By Scott Havird · · Weekly Digest

14 releases this week! New AI critic for code review, Cursor 3.0 launches, and multi-registry architecture lands in Gemini CLI. +17% release velocity.

AI Coding Tools Week 14: New Critic Agent & Major Interface Overhaul

TL;DR 📋


  • New AI critic agent automatically reviews your code plans and catches errors before you ship
  • Cursor 3.0 drops with completely redesigned interface

    • This Week's Releases

    Claude Code (5 releases)


    Claude's been busy with policy updates and interactive features. The standout addition is a forceRemoteSettingsRefresh policy that blocks CLI startup until settings are fetched - useful for enterprise environments that need fail-closed behavior.

    Undefined Tool (6 releases)


    Major week for this mystery tool with 6 rapid-fire releases. The headline feature: a new Critic agent that automatically reviews your plans and complex implementations using a complementary model. It's in experimental mode for Claude models, but early reports suggest it catches logic errors you'd miss. Also shipped: built-in CLI skills and HTTPS redirect support for MCP OAuth flows.

    Cursor (1 release)


    Big news: Cursor 3.0 landed with a completely new interface. No detailed changelog yet, but expect the usual Cursor polish with better code completion workflows.

    Gemini CLI (1 release)


    Version 0.36.0 introduces multi-registry architecture and tool filtering for subagents. This looks like foundational work for more complex agent orchestration.

    OpenAI Codex CLI (1 release)


    Windows users get better sandbox security with OS-level egress rules instead of just environment variables. Plus new app-server client capabilities.

    Highlight Deep Dive 🔍

    The Critic agent from the undefined tool is potentially game-changing for solo developers. Instead of just generating code, it automatically reviews your implementation plans using a separate model to catch logical flaws, edge cases, and architectural issues. Think of it as having a senior developer constantly peer-reviewing your work. While still experimental, this could significantly reduce bugs that make it to production - especially for complex features where you're juggling multiple moving parts.

    Quick Stats 📊


  • Total releases: 14 (up from 12 last week)
  • Biggest mover: Undefined tool with 6 releases
  • Velocity trend: +17% week-over-week
  • Most active: Claude ecosystem with continuous policy and MCP improvements

    • The release pace is accelerating as tools mature beyond basic code completion into sophisticated development workflows. The Critic agent and multi-registry architectures suggest we're moving toward AI tools that can handle entire development lifecycles, not just individual coding tasks.

    Tools covered: github-copilot, claude-code, cursor, gemini-cli, openai-codex

    View all articles on Havoptic