OpenAI's Codex: AI Mastering Adobe Lightroom Without API

Peter Gostev, head of AI capabilities at Arena.ai, recently shared an experience that highlights the potential of artificial intelligence agents. In addition to his work, he created what he calls the defecation test, officially known as BullshitBench. However, it is another one of his experiments that particularly caught attention.

When faced with the tedious task of denoising 50 images in Adobe Lightroom, he decided to enlist Codex, OpenAI's coding service, to find an automated solution. Gostev explained that the task typically involved clicking on each image individually to apply the denoising, a laborious and repetitive process. Rather than tackling it manually, he asked Codex to handle it. To his surprise, the AI managed to accomplish this task without relying on an official API, a plugin, or a browser workaround.

What makes this experience particularly noteworthy is Codex's ability to interact directly with the Adobe Lightroom desktop application. This direct interaction, without explicit support for such functionality, illustrates the direction AI agents are taking. They are no longer just assisting users but are navigating and operating autonomously within complex software environments.

Gostev, who possesses advanced technical skills, acknowledged that what he achieved with Codex is something few people could replicate without similar expertise. Yet, this paves the way for a new era where AI agents could not only assist but surpass human capabilities in executing complex and repetitive tasks.