Perplexity has expanded "Personal Computer," a tool that lets AI operate the PC itself, to Windows. According to The Verge, like the Mac version released in April, it acts as a "general-purpose digital worker" that can access local files and apps to get tasks done. What matters here is that the target is the world's most common OS. A once experimental idea has now reached the environment many people use every day.

It took about half a year to go from Mac to Windows support. Behind this is the fact that AI models are getting close to a practical level at "understanding and operating" screens and file structures. The old hurdle was building a safe way to touch local files and apps. Unlike chat AI that stays in the cloud, operating a PC carries risks around permissions and mistakes. That area is starting to be sorted out, which likely pushed the rollout to a mass-market OS.

From a long-term investor's angle, the key is a structural shift in who handles routine work. Tasks like sorting files, drafting documents, and typing across several apps may start to be done by AI on your own PC. As a change close to readers' jobs, repetitive office work is likely to feel the effect first. For now, though, the role stays supportive, and checking and judgment remain with people. The gains in efficiency and the rules to prevent mistakes need to be considered together.

The move toward agents is not only about visible apps. Google announced that it expanded "Managed Agents" in the Gemini API, adding the lightweight model 3.6 Flash, hooks (a way to insert processing at certain moments), and triggers (a way to start automatically on an event). It is about building a base that makes it easier for developers to assemble agents. Perplexity secures the user-facing entry, while Google secures the developer-facing base. The land-grab for agents is advancing from both directions at once.

The change is not limited to office tasks. According to OpenAI, researchers have begun using AI coding agents to modernize scientific computing software. Even in advanced fields like genomics, the speed of development and discovery is said to be rising. What this suggests is that AI is starting to enter knowledge work that requires expertise, not just simple tasks. Expertise itself does not become unnecessary; rather, we are reaching a stage where the gap widens depending on whether experts can use AI well.