Key Signals

• GitHub Copilot Memory is now on by default for all Pro and Pro+ users, marking a significant shift toward persistent, repository-aware AI assistance. Copilot Memory automatically discovers and stores facts about coding conventions, architectural patterns, and cross-file dependencies, then reuses that knowledge across the coding agent, code review, and Copilot CLI. Memories are scoped to individual repositories, validated against current code before use, and auto-expire after 28 days. This cross-agent knowledge sharing means insights from the coding agent immediately improve code review quality and vice versa. [1]

• Cursor expanded to JetBrains IDEs through the Agent Client Protocol, bringing agentic AI coding to IntelliJ IDEA, PyCharm, WebStorm, and other JetBrains environments. This is a major interoperability move: developers who rely on JetBrains for Java and multilanguage support can now use frontier models from OpenAI, Anthropic, Google, and Cursor with secure codebase indexing and semantic search. JetBrains Head of IDEs Aleksey Stukalov endorsed the collaboration, calling it "a win for Cursor, for JetBrains, and most importantly for developers." The ACP is emerging as a standard protocol for agent-IDE communication. [2]

• OpenAI's Codex agentic coding app launched on Windows with native sandboxing, PowerShell support, and a WinUI skill, opening the platform to over 500,000 waitlisted developers. The Mac version hit 1 million downloads in its first week and now has 1.6 million weekly active users, demonstrating strong developer demand for agent-centric coding tools. The Windows version uses OS-level restricted tokens and filesystem ACLs rather than just Linux containers, and defaults to the GPT-5.3-Codex model. OpenAI describes it as "a command center for agents" — a new form factor focused on managing agents rather than editing code directly. [3]

• Claude Code v2.1.68 defaulted Opus 4.6 to medium reasoning effort for Max and Team subscribers, signaling Anthropic's push to balance performance and speed in agentic coding. The "ultrathink" keyword was re-introduced to enable high effort for specific turns, giving developers explicit control over when to invoke deep reasoning. Older models (Opus 4 and 4.1) were removed from the first-party API with automatic migration to Opus 4.6, simplifying the model lineup. [4]

• Grok Code Fast 1 became generally available to Copilot Free users via auto model selection, continuing GitHub's strategy of diversifying the model roster behind Copilot. The model joins the pool that Copilot's "Auto" mode can select from in VS Code, Visual Studio, JetBrains IDEs, Xcode, and Eclipse. This is the latest in a series of model additions that give Copilot's routing system more options to match tasks with optimal models. [5]

AI Coding News

• freeCodeCamp published a comprehensive guide on building production-grade MCP servers for internal organizational data, covering authentication, multi-tenancy, and deployment. The tutorial goes beyond basic examples to address real enterprise needs: connecting PostgreSQL databases to AI assistants, implementing OAuth 2.0, scoping data access per user, and building RAG tools for internal documents. It demonstrates using the official MCP SDK with TypeScript, Zod for parameter validation, and Express for transport — with emphasis on tool design principles that help AI models use your tools effectively. The growing MCP ecosystem is making it easier to connect any internal system to any MCP-compatible AI client. [6]

Feature Update

• GitHub Copilot Memory is now enabled by default for Pro and Pro+ users in public preview. Previously opt-in, Copilot Memory builds persistent, repository-level understanding — storing coding conventions, architectural patterns, and cross-file dependencies that are validated against current code and auto-expire after 28 days. It works across the coding agent, code review, and Copilot CLI, with knowledge shared between agents. Enterprise admins retain full policy control, and repository owners can review and delete stored memories under Repository Settings. [1]

• Cursor is now available in JetBrains IDEs via the Agent Client Protocol. Developers install the Cursor ACP from the ACP Registry and authenticate with their existing Cursor account — it is free for all paid plan users. The integration brings Cursor's secure codebase indexing, semantic search, and multi-model agent harness into the JetBrains ecosystem. This positions ACP as a growing cross-IDE standard for agentic coding. [2]

• OpenAI Codex launched on Windows with native sandboxing, PowerShell support, and a new WinUI skill. The Mac version already has 1.6 million weekly active users since its February launch; over 500,000 Windows developers were on the waitlist. The Windows app uses OS-level restricted tokens and filesystem ACLs for sandboxing, and defaults to the GPT-5.3-Codex model with switchable options including GPT-5.2-Codex, GPT-5.1-Codex-Max, and GPT-5.1-Codex-Mini. Available for all ChatGPT plan tiers including Free. [3]

• Claude Code v2.1.68 defaults Opus 4.6 to medium reasoning effort for Max and Team subscribers. Medium effort is positioned as the sweet spot between speed and thoroughness for most tasks, changeable via /model. The "ultrathink" keyword returns for enabling high effort on specific turns. Opus 4 and 4.1 are removed from the first-party API, with automatic migration to Opus 4.6 for pinned users. [4]

• Grok Code Fast 1 is now available in Copilot Free auto model selection. The model is added to the pool that Copilot's "Auto" mode draws from in GitHub Copilot Chat across VS Code, Visual Studio, JetBrains IDEs, Xcode, and Eclipse. This expands the model diversity available to free-tier users. [5]

• GitHub Copilot CLI v0.0.422-0 adds personal hooks, plugin auto-install, and session usage metrics. Personal hooks can now be loaded from ~/.copilot/hooks alongside repo-level .github/hooks, and the new enabledPlugins config enables automatic plugin installation at startup. Session usage metrics are persisted to events.jsonl after each session. Additional improvements include /delegate support for multi-remote repositories, a config rename from .github/copilot/config.json to settings.json, and several fixes for plugin cache corruption, Windows URL handling, and markdown table rendering. [7]

• OpenAI Codex shipped two stable Rust releases (v0.109.0 and v0.108.0) plus two alpha pre-releases on the same day, indicating an accelerated release cadence. The releases (0.108.0-alpha.11, 0.108.0-alpha.12, rust-v0.108.0, and rust-v0.109.0) were published between 12:00 and 23:13 UTC on March 4. Release notes are minimal, suggesting incremental improvements to the Codex Rust implementation. [8][9]

• Gemini CLI published three releases: v0.32.1 and v0.33.0-preview.1 and a feature-rich nightly build (v0.34.0-nightly). The patch releases both cherry-pick the same fix (commit 0659ad1). The nightly build is more substantial, adding tracker CRUD tools with visualization, experimental LXC container sandbox support, iterative loop detection with model feedback, esbuild bundled npm packages, improved @file autocomplete prioritization, and MCP notification fixes. Documentation for planning workflows with Conductor examples was also added. [10][11][12]