Codex goes mobile as AI coding agents get remote control

Codex breaks out of the terminal

The week's biggest story is that Codex is now accessible from the ChatGPT mobile app, turning your phone into a command center for AI-assisted development. Rolling out across iOS and Android, the mobile experience surfaces live project context, approvals, terminal output, diffs, and test results — keeping longer-running tasks moving when you step away from the host Mac. For enterprise teams, remote access and access tokens for automation arrived alongside the mobile launch, with remote control disabled by default for security-conscious workspaces.

Beneath the mobile headline, Codex got serious infrastructure upgrades. Codex CLI 0.131.0 introduced richer TUI controls — data-driven service-tier commands, responsive Markdown tables, and blended token usage — plus daemon-managed remote workflows and a Python SDK move to openai-codex with pinned types. A new Amazon Bedrock setup guide covers deploying Codex with OpenAI models through AWS-managed infrastructure, and fresh mobile documentation walks through setup steps, security requirements, and connection troubleshooting.

Claude Code gets session management and a smarter agent mode

Anthropic responded with its own session-management breakthrough: agent view in Claude Code provides a single dashboard of every Claude Code session — running, blocked on you, or done — accessible via the claude agents command. The companion v2.1.139 release adds a /goal command that makes Claude keep working across turns until a completion condition is met, with live elapsed time, turn count, and token usage displayed in an overlay panel. Both features are available as a research preview on Pro, Max, Team, Enterprise, and API plans.

Subsequent Claude Code releases this week hardened the agent experience. v2.1.140 fixed /goal silently hanging when hooks were disabled, and resolved background service startup failures on machines with enterprise endpoint security. v2.1.141 fixed permission prompt dismissal when switching modes mid-prompt and ensured background jobs move to Completed instead of lingering under Working. v2.1.142 upgraded fast mode to Opus 4.7 by default (with an environment variable override to pin to 4.6) and fixed MCP tool timeouts not respecting configured values. Finally, v2.1.143 added plugin dependency enforcement and hardened background session state after sleep/wake cycles.

Fast mode expands and platform tools mature

Across both platforms, performance investments accelerated. Beyond Claude Code's fast mode upgrade, Claude Platform fast mode now supports Opus 4.7 via speed: "fast" with the fast-mode-2026-02-01 beta header, delivering significantly faster output token generation at premium pricing. Anthropic also launched cache diagnostics in public beta — pass diagnostics.previous_message_id on a Messages request and the API reports a cache_miss_reason explaining where the prompt cache prefix diverged from the previous turn.

On the pricing and access layer, OpenAI SDKs — Node, Python, and Go — all added a service_tier parameter to the responses compact method. OpenAI also made the ChatGPT Apps Directory generally available by removing its beta label for enterprise and education workspaces, while rolling out a file library with storage management to Free and Go users. The week closed with the launch of personal finances in ChatGPT for Pro users in the U.S., connecting financial accounts through Plaid for AI-powered spending and investment insights.