OpenAI Codex June 29: Codex Remote Now Generally Available, ChatGPT Mobile QR Pairing, CLI v0.142.0 Ships Plugin Marketplace and Multi-Agent Delegation

OpenAI Jalapeño: First Custom LLM Inference Chip Built with Broadcom - Better Perf-per-Watt Than Nvidia for Codex Remote

• On June 24, TechCrunch confirmed OpenAI unveiled Jalapeño, its first custom-built LLM inference accelerator developed with Broadcom, claiming better performance-per-watt than current Nvidia hardware for serving large language models powering services like Codex Remote.

Editor’s read: Better inference efficiency means lower marginal compute cost for OpenAI - which has historically flowed through to lower pricing or higher usage limits. Worth watching for Codex Remote pricing updates.

OpenAI Codex Remote Now Generally Available: ChatGPT Mobile Controls Mac or Windows via QR Code, DigitalOcean Plugin, Inline Review Notifications

• On June 25, OpenAI announced Codex Remote is generally available for all users.
• The ChatGPT mobile app can now pair with a Mac or Windows host via QR code scan to drive Codex remotely, with inline review comments surfaced as push notifications on your phone.
• A DigitalOcean plugin was also available at launch per the official Codex changelog.

Editor’s read: The mobile-to-desktop QR pairing is the biggest UX unlock yet for Codex Remote - review and approve agent tasks from your phone without switching contexts. Set it up before your next long-running agent session.

Codex CLI v0.142.0: Usage Credits Tracking, Plugin Marketplace, Multi-Agent Delegation, and Indexed Web Search

• Released June 22, Codex CLI v0.142.0 is the largest feature release of the week: introduces in-UI usage credits tracking, a plugin marketplace for extending Codex capabilities, multi-agent task delegation allowing Codex to spawn and coordinate sub-agents, and indexed web search as a native built-in tool.

Editor’s read: Multi-agent delegation and indexed web search together close a significant gap with Claude Code's agentic modes. The plugin marketplace is early but is the vector for ecosystem expansion - watch which plugins ship in the first month.

Codex CLI v0.142.2 Now Current Stable: MCP Tool Search Default, macOS Proxy Support, Plugin Dark Mode, Bedrock Fix

• v0.142.1 (June 25) added comprehensive Windows system proxy support covering PAC, WPAD, static, and bypass configurations.
• v0.142.2 (June 25) set MCP tool search as the default search mode, added macOS proxy support, introduced plugin dark mode, and fixed a Bedrock integration bug.
• v0.142.3 and v0.142.4 are maintenance patches with no user-facing changes.

Codex CLI codex-zsh v0.1.0: Initial ZSH Shell Integration Ships

• Released June 25, codex-zsh v0.1.0 is the initial release of native ZSH shell integration for Codex CLI, allowing Codex to read and interact directly with your ZSH shell environment, history, and configuration.

Editor’s read: ZSH integration closes a gap that bash-only users have not noticed but ZSH users have felt acutely. If you run ZSH with oh-my-zsh or custom prompts, install codex-zsh v0.1.0 now and test shell context reading.

OpenAI Codex CLI v0.143.0 Alpha Track: 29 Pre-Release Builds This Week (alpha.3 Through alpha.29, June 23-28)

• The v0.143.0 alpha track has been extremely active June 23-28, shipping alpha.3 through alpha.29 - 27 pre-release builds in six days.
• These builds are not recommended for production use; follow the alpha releases for early preview of upcoming v0.143.0 stable features.

Best AI Coding Tools Pricing June 2026: OpenAI Codex Free vs Paid vs Claude Code, Cursor, GitHub Copilot Compared

• Developers Digest published June 28 a side-by-side pricing reality check for Codex, Claude Code, Cursor, and GitHub Copilot.
• covering what each free tier actually provides, usage limits, and real monthly cost at professional usage levels.

Matt Pocock: Audit Your Codex AI Coding Instructions or Burn Tokens for Nothing

• On June 24, TypeScript educator Matt Pocock posted on X that many teams running Codex and similar AI coding agents have instructions in their configs that are effectively no-ops - the model reads them but takes no action - burning tokens systematically on every task.
• He recommends a systematic instruction audit.

Editor’s read: No-op instructions are a hidden tax on every Codex session. Run a quick audit: for each custom instruction, can you verify it actually changes Codex behavior in a test? If not, remove it.

Karpathy: Enterprise AI Coding Tools Like Codex Now Write Majority of Code - Developers Shift into Management Roles

• On June 24, Andrej Karpathy posted on X that enterprise AI coding has crossed a new threshold.
• tools including Codex are deeply integrated into multiplayer workflows, writing the majority of code, with human developers increasingly operating as managers and reviewers rather than primary authors.

Agentic PR Problem: Codex Agents Creating Pull Requests Too Large or Opaque for Human Review - Setting Oversight Gates

• A June 28 essay by Jon Udell identifies the growing workflow breakdown where AI coding agents including Codex generate PRs too large, complex, or opaque for meaningful human review, and proposes concrete team policies for defining PR size limits and required review criteria before autonomous agent commits.

Editor’s read: As multi-agent delegation in v0.142.0 makes Codex capable of more autonomous work, the unreviewable PR risk scales with it. Define your max PR size and review checklist before delegating longer tasks.

Theo Gives Away 44-Minute AI App Idea Breakdown - Underbuilt Opportunities for Codex-Powered Development

• On June 24, developer educator Theo (t3.gg) released a 44-minute video cataloguing AI app ideas he considers underbuilt and available for developers to ship.
• framing them as rapid-development targets well suited to Codex's new plugin and multi-agent capabilities.