Claude Code, Codex and Agentic Coding #8
Claude Code bug fixes, Codex auto-review and 90+ plugins, plus dreaming agents.
Anthropic's Claude Code endured three issues in April, all since fixed. The default reasoning level was changed from high to medium to reduce latency, then reverted after user backlash. A bug stripped older thinking from sessions idle for over an hour, fixed after two weeks. A system prompt change intended to reduce verbosity actually hurt coding quality, and was rolled back within four days. Anthropic promised better internal testing to avoid future regressions.
Meanwhile, OpenAI's Codex received major upgrades: auto-review mode, a significant speed boost for computer use, support for 90+ plugins, and a Chrome plugin for browser automation. Claude Code also added a slew of new features: push notifications via phone, a command to reduce permission prompts, session recaps, a focus mode for minimal distractions, dedicated review sessions via /ultrareview, token usage tracking, and 'dreaming' for asynchronous self-improvement. Auto mode is now live for Max users. The competition between Claude Code and Codex continues to accelerate, with both shipping dozens of fixes weekly.
- Claude Code had three bugs in April: reasoning default change reverted, idle session thinking strip fixed, prompt verbosity issue reverted; Anthropic promises better testing.
- Codex added auto-review, faster computer use, 90+ plugins, and a Chrome plugin for browser automation.
- Claude Code added /focus, /ultrareview, /usage, push notifications, dreaming, and auto mode for Max users.
Why It Matters
Coding agents are rapidly improving, making them more reliable and autonomous, reshaping developer productivity and workflows.