Codex update breaks Computer Use, plugins

// 1h agoPRODUCT UPDATE

Codex update breaks Computer Use, plugins

A recent OpenAI Codex desktop application update has reportedly broken the agent's core "Computer Use" and internal browser features. Users report that workspace plugins fail to initialize, stalling automation tasks.

// ANALYSIS

While OpenAI pitches Codex as a highly capable desktop agent, this update shows how fragile OS-level tool execution remains. Rushing updates without stabilizing the plugin registry can easily break developer workflows.

–The update appears to corrupt local plugin configurations and bundled marketplace manifests.
–Users report that plugins appear unavailable or fail to launch immediately after auto-updating.
–Advanced workarounds require developers to manually edit local config files or run CLI commands to rebuild the plugin cache.
–The issues contrast with OpenAI's optimistic research paper published the same day emphasizing Codex's internal adoption.

// TAGS

codexopenaicomputer-useagenttool-useclidevtool

DISCOVERED

1h ago

2026-06-25

PUBLISHED

1h ago

2026-06-25

RELEVANCE

8/ 10

AUTHOR

mark_k

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH9m ago

Anthropic launches Claude Tag in Slack

Anthropic has introduced Claude Tag, a Slack-native integration that turns the AI assistant into a persistent, shared teammate within public channels. Designed for multiplayer collaboration, the bot can remember channel context, utilize connected codebase tools, and run tasks asynchronously under an organizational identity.

UPDATE9m ago

OpenAI updates GPT-5.5 Instant default model

OpenAI rolled out a significant update to GPT-5.5 Instant, the default model for ChatGPT. The release focuses on improving conversational flow, managing complex multi-turn constraints, and enhancing local recommendations with image support.

MODEL9m ago

Mistral launches OCR 4 document model

Mistral AI has released Mistral OCR 4, a state-of-the-art document intelligence model that extracts text, tables, and structured layout data from complex PDFs and presentations. The model introduces paragraph-level bounding box extraction, block classification, and inline confidence scores across 170 languages.