Descriptions:
OpenAI’s Codex app brings the GPT 5.3 Codex model into a dedicated desktop coding environment for macOS, and in this video David Ondrej puts it through its paces by building a real open-source project called Open Dash — a central dashboard for sharing markdown files and skills between human teams and AI agents. The video serves as both a practical introduction to the Codex app interface and a demonstration of agentic software development practices.
Ondrej covers the full workflow: selecting reasoning effort levels (Low through Extra High), managing multiple parallel agent threads, using Plan mode to prevent premature code changes, and integrating with GitHub for one-click version control. He highlights a key context-engineering technique — wrapping instructions in XML tags to improve agent focus — and shows how to layer in iterative feedback (including screenshots of poor UI) to guide the model toward better results. The Cerebras-powered Spark variant of the model is noted for its approximately 1,000-tokens-per-second throughput for fast, lighter tasks.
The video also includes a direct comparison between GPT 5.3 Codex and Claude Opus 4.6, with Ondrej testing each on front-end design challenges. Both models are run inside the Codex app’s terminal environment, illustrating how the platform supports model-agnostic agentic workflows for developers who want to build and iterate on software without switching tools.
📺 Source: David Ondrej · Published February 16, 2026
🏷️ Format: Hands On Build
