Key Signals

• Cursor launches long-running autonomous agents that complete complex tasks without human intervention. The research preview enables agents to work over extended horizons, planning first before execution, resulting in larger and more complete pull requests with fewer follow-ups. This represents a significant evolution from traditional copilot-style assistance to truly autonomous code generation, available now for Ultra, Teams, and Enterprise customers. [1]

• Major AI coding tools release simultaneous updates, signaling rapid iteration in the CLI space. GitHub Copilot CLI shipped two releases in a single day (v0.0.409 and v0.0.408), while Gemini CLI released three versions and OpenAI Codex shipped two major updates, all on February 12th. The coordinated pace suggests intense competition and feature parity pressure across the ecosystem. [2][3][4][5][6]

• Spotify reports its top developers have stopped writing code manually, relying entirely on AI since December. The music streaming giant credits Claude Code and its internal "Honk" AI system for the transition, marking one of the first major tech companies to publicly claim complete AI-driven development for senior engineers. This validates the maturity of agentic coding tools for production enterprise use at scale. [7]

• OpenAI unveils Codex powered by dedicated custom silicon, establishing hardware-software vertical integration. The new chip represents what OpenAI calls "the first milestone" in its chipmaker partnership, potentially enabling faster inference and lower latency for code generation tasks. This follows the industry trend of AI companies building custom hardware to optimize their models. [8]

• GitHub Copilot CLI integrates VS Code and introduces full-screen diff mode with alt-screen support. The v0.0.409 release adds seamless IDE integration and improved visual tools for code review, including scrollable permission prompts for large diffs and a plugin marketplace for easier discovery. These UX improvements address developer friction points in AI-assisted workflows. [2]

AI Coding News

• Spotify's engineering leadership confirms complete transition to AI-driven development for top performers. TechCrunch reports that Spotify's best developers haven't manually written code since December, instead relying on Claude Code and the company's internal "Honk" AI system to generate and refactor code. The announcement represents one of the most aggressive public commitments to agentic coding by a major tech company, suggesting AI tools have reached production-grade reliability for enterprise-scale development workflows. [7]

• OpenAI partners with chipmaker to develop custom silicon for Codex code generation. The new version of Codex runs on dedicated hardware optimized for code generation workloads, which OpenAI describes as "the first milestone" in its relationship with the chip manufacturer. This vertical integration strategy mirrors similar moves by Anthropic, Google, and other AI labs seeking to optimize inference costs and latency through custom ASICs and accelerators. [8]

Feature Update

• GitHub Copilot CLI v0.0.409 ships with VS Code integration, full-screen diff mode, and plugin marketplace support. The release introduces alt-screen mode for /diff commands, a quick help overlay with grouped shortcuts accessible via "?", and theme preview capabilities for screen readers. The list_copilot_spaces tool is now included in the default GitHub MCP config, and permission prompts with long diffs are now scrollable, addressing usability issues with large code changes. [2]

• GitHub Copilot CLI v0.0.408 adds streamer mode, mouse text selection, and improved MCP server support. The update introduces /streamer-mode to hide preview model names and quota details during streaming sessions, adds mouse text selection in alt-screen mode, and makes shellId more flexible to accept numeric inputs. Background task hints now update when detached shells are killed or removed, and MCP servers now respect the cwd working directory property for better context awareness. [3]

• Cursor's long-running agents enter research preview, enabling autonomous multi-hour code generation. The new capability allows agents to plan first and execute complex tasks without human intervention, completing work previously too difficult for regular agents. Internal testing showed agents producing larger, more complete pull requests with fewer follow-up issues, and the feature is now available at cursor.com/agents for Ultra, Teams, and Enterprise plans. [1]

• OpenAI Codex v0.101.0 fixes model resolution and memory processing stability issues. The release ensures model_info preserves the requested model slug when selecting by prefix, preventing unwanted model reference rewrites. Developer messages are now excluded from phase-1 memory input to reduce noise, and memory phase processing concurrency was reduced to improve consolidation and staging stability under load. [4]

• OpenAI Codex v0.100.0 introduces experimental JavaScript REPL and websocket transport with multiple rate limit support. The major release adds a feature-gated JavaScript REPL runtime that can persist state across tool calls with optional runtime path overrides. It reintroduces app-server websocket transport with split inbound/outbound architecture and connection-aware thread resume subscriptions, adds memory management slash commands in the TUI, and promotes sandbox capabilities on both Linux and Windows with a new ReadOnlyAccess policy shape. [5]

• Gemini CLI v0.30.0-nightly.20260212.207ac6f2d hardens Plan Mode with conditional policy rules and agentic execution refinements. The nightly release shows notifications when extension commands conflict, adds dismissable "?" shortcuts help, and prioritizes conditional policy rules in Plan Mode. System prompts have been refined for agentic execution, AskUser tool usage metrics are now tracked, and Ctrl-Z suspension support enables better workflow interruption. [6]

• Gemini CLI v0.30.0-nightly.20260212.099aa9621 updates internal models to Gemini 3 and simplifies tool confirmation UI. The release caches CLI version for session consistency, fixes double rendering in shpool, and documents Plan Mode policy override validation. The F12 behavior now only opens the drawer if the browser fails, skills can be enabled in Plan Mode, and grep tool definitions have been extensively updated to limit context size and improve performance. [9]

• Gemini CLI v0.29.0-preview.2 patches headless mode detection for improved non-interactive workflows. The preview release cherry-picks a critical fix for headless mode detection, ensuring CLI behavior correctly adapts to automated and CI/CD environments where interactive prompts should be suppressed. [10]