How OpenAI's Codex figured out how to use Adobe software

In a fascinating demonstration of artificial intelligence capabilities, Peter Gostev, the AI capability lead at Arena.ai, shared his experience with OpenAI's Codex. He recently introduced a unique assessment tool known as the Bullshit Bench, but it was his use of Codex that truly captivated my attention. Faced with the daunting task of denoising 50 images in Adobe Lightroom—a typically labor-intensive process that often requires manual intervention—Gostev opted for an innovative approach. Rather than painstakingly editing each photo by hand or struggling with batch processes, he allowed Codex to tackle the challenge. "You have to go and click into each one to denoise 50 photos. That sounds like hard work, so I just got Codex to go and work out how to do that," he explained. What’s particularly impressive is how Codex achieved this feat. It didn’t rely on an official API, a plugin, or any browser workaround. Instead, it managed to interface directly with the desktop application, showcasing its ability to operate autonomously in a manner typically reserved for human users. This incident not only highlights the advanced capabilities of AI agents but also hints at a future where they're not just tools, but can navigate and utilize software with remarkable efficiency, often surpassing human performance. Gostev's experience is a glimpse into the evolving role of AI in our daily tasks, indicating a future where technology adapts seamlessly to user needs, enhancing productivity in ways we are just beginning to imagine.