OpenAI Codex App adds Record and Replay

// 93d agoPRODUCT UPDATE

OpenAI Codex App adds Record and Replay

OpenAI has released version 26.616 of its Codex App on macOS, introducing a Record and Replay feature that converts demonstrated manual workflows into reusable automation skills. The feature requires the Computer Use permission and is currently unavailable in the European Union.

// ANALYSIS

The addition of Record & Replay marks a significant transition for Codex from a text-prompt-based assistant to a visual macro recorder powered by agentic AI, drastically lowering the barrier to automating complex desktop tasks. By utilizing computer-use capabilities to observe and replicate user actions, it makes automation accessible without requiring users to write code or describe complex multi-step processes.

* Desktop automation becomes dramatically easier, shifting from manual coding or detailed prompt engineering to simply demonstrating the desired task.

* The exclusion of the EU at launch highlights ongoing regulatory hurdles for screen-observing AI agents in Europe.

* Requiring "Computer Use" permissions emphasizes the security and privacy implications of giving AI agents direct control over the operating system.

// TAGS

openaicodex-appcomputer-usemacosautomationrecord-and-replay

DISCOVERED

93d ago

2026-06-18

PUBLISHED

94d ago

2026-06-18

RELEVANCE

8/ 10

AUTHOR

mark_k

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE1h ago

Grok Build 1.0.38 fixes transcript truncations, permissions

Grok Build version 1.0.38 resolves an issue where long agent replies in prompt-suggestion transcripts were truncated mid-sentence. The update also introduces per-deployment read_file permissions for skill and instruction files to enable finer access controls across environments.

OPEN SOURCE1h ago

Comfy-Org brings native Qwen-Image-2.1 support

Comfy-Org has released repackaged model weights and native workflow templates for Alibaba's Qwen-Image-2.1 diffusion model on Hugging Face, bringing turnkey text-to-image and image editing capabilities directly to ComfyUI. Rather than relying on custom third-party nodes or raw upstream checkpoints, creators can now drop official Comfy-Org safetensors directly into their models directory and access pre-built templates out of the box.

BENCHMARK1h ago

Brood War Bench pits frontier LLMs in StarCraft

Brood War Bench is an AI evaluation project by Ben Swerdlow testing large language models in real-time StarCraft: Brood War matches across 171 games. Codex Astra topped the leaderboard undefeated with disruptive probe tactics, while results highlighted that real-time environments severely penalize reasoning latency as lower-effort configurations often outperformed higher-effort ones.