Raycast CEO Thomas Paul Mann thinks we're settling for too little from AI. While everyone's building another way to chat with ChatGPT, his productivity app is turning large language models into actual computer operators that can rename your photos, manage files, and control any app on your Mac or PC.
The AI productivity revolution just got real. Raycast CEO Thomas Paul Mann dropped a vision bomb during The Vergecast that should make every tech worker pay attention. While the industry's busy building the 47th way to chat with ChatGPT, Mann's team is turning AI into your computer's new operating system.
"If all you want is a way to talk to ChatGPT, you're practically spoiled for choice," Mann told The Verge's David Pierce. "But that's not the whole story of how we'll interact with large language models. It better not be."
Raycast isn't just another launcher app anymore. It's become a Trojan horse for what Mann calls "agentic AI" - artificial intelligence that doesn't just answer questions but actually performs tasks on your behalf. The app already handles file searches, note-taking, and app launching, but its AI integration lets it rename photos, organize documents, and theoretically control any application installed on your machine.
The implications are staggering. Instead of typing commands into a search bar, you could tell Raycast's AI to "organize my desktop photos by date and subject" or "find all my Q3 budget files and create a summary." The AI doesn't just find information - it manipulates it, moves it, and acts on it.
Mann's betting that desktop integration beats browser-based AI assistants. While companies like Google and Microsoft are cramming AI into Chrome and Edge, Raycast thinks replacing your Mac's Spotlight and PC's Start menu creates deeper, more useful access to your digital life. "It can help you create, manage, and organize files, but it can also operate inside of any app you have installed," Mann explained during the interview.
But here's where things get scary. Mann isn't naive about the risks. "It's one thing for a chatbot to make mistakes in a text chat; it's another entirely to turn a hallucinating, imperfect tool loose on your computer," he acknowledged. When ChatGPT hallucinates about historical facts, you roll your eyes. When an AI agent hallucinates while managing your files, you could lose work, corrupt data, or worse.
The technical challenges are immense. Current AI agents "by and large, don't really work," as Mann frankly admits. They struggle with complex, multi-step tasks and often misunderstand context. But Raycast's approach - working with local files rather than the entire internet - might actually make AI more reliable, not less. The scope is narrower, the data more structured, the feedback loop tighter.
This interview reveals a fundamental shift happening in AI development. The chatbot phase was just the beginning. Now we're entering the "agentic AI" era, where artificial intelligence doesn't just think and respond - it acts and executes. Apple hinted at this future with its recent AI announcements, and Microsoft is embedding similar capabilities into Windows. But Raycast is actually shipping it today.
Mann's vision extends beyond simple automation. He sees AI as the new interface layer between humans and computers. Instead of learning keyboard shortcuts, menu paths, and application-specific commands, users could simply describe what they want in natural language. It's the ultimate productivity multiplier - if it works.
The competitive landscape is heating up fast. Apple controls the Mac ecosystem and could easily build these features into macOS. Microsoft has similar advantages on Windows. But third-party developers like Raycast move faster and can integrate multiple AI models simultaneously, giving users choice and flexibility that platform owners might not.
What makes Mann's approach particularly clever is starting with power users who already understand Raycast's existing functionality. These early adopters can provide feedback on AI features without expecting perfection. It's a safer way to test dangerous capabilities than rolling them out to mainstream consumers who might accidentally delete their photo libraries.
The timing couldn't be better. As AI model capabilities improve and desktop computing evolves, there's a brief window where independent developers can establish themselves before the big tech platforms lock down these integrations. Raycast is racing to claim that territory while the opportunity exists.
Mann's vision represents the next evolution of human-computer interaction. We're moving beyond asking AI questions to letting it control our digital environments. The risks are real - AI agents could accidentally damage files or misinterpret commands. But the productivity gains could be transformative. Whether Raycast can execute this vision before Apple and Microsoft build similar features into their operating systems will determine if independent developers can stake a claim in the agentic AI future.
