Move over, chatbots—AI has learned to click. Claude, Anthropic’s AI assistant, has just mastered the art of using a computer, and it’s as surprising as finding your hamster using your laptop to order food.
Anthropic’s recent announcement goes beyond its upgraded AI models (Claude 3.5 Sonnet and the speedier Claude 3.5 Haiku). It’s introduced something revolutionary: computer control. Claude can now navigate computers independently, manage emails, populate spreadsheets, and—yes—even browse your boring vacation photos.
From Chat to Action Previously, AI assistants like Claude were confined to conversations and text generation. Now, Claude can visualize screen content, control cursors, click, and type—essentially mirroring human-computer interaction. Think of it as your digital assistant finally getting hands.
Anthropic’s approach sets this apart: they’ve equipped Claude with generalized computer skills, enabling it to interact with various programs and websites much like humans learning new software.
Real-World Applications Professor Ethan Mollick, author of “Co-Intelligence,” tested Claude’s abilities. “I requested a high school lesson plan for ‘The Great Gatsby,'” he recalls. “Claude took complete control—downloading the book, researching lesson plans online, creating a spreadsheet, and populating it with teaching ideas. It even aligned the plan with educational standards. I could step away and return to find the completed work.”
Anthropic’s demonstrations show even more possibilities. In one demo, Claude planned a trip by launching Chrome, searching Google, creating calendar events, and sending invitations. In another, it seamlessly completed vendor forms by cross-referencing multiple documents—a task many office workers dream of automating.
Beta Testing: Powerful but Developing The computer control feature remains in public beta, with Anthropic recommending several precautions:
- Use virtual environments with restricted permissions
- Avoid sharing sensitive data, including login credentials
- Limit internet access to trusted websites
- Maintain human oversight for critical decisions, especially financial ones
Some limitations persist. Claude still struggles with scrolling, dragging, and zooming. During one demonstration, it amusingly paused its programming task to explore Yellowstone photographs—proving that even AI can get distracted!
Enhanced Models, Better Performance Claude 3.5 Sonnet demonstrates significant improvements, particularly in coding tasks. Its counterpart, Claude 3.5 Haiku, offers performance comparable to Claude 3 Opus but with reduced costs and faster processing. Software development platform Replit already employs Claude’s new capabilities for application evaluation.
A Technological Watershed Professor Mollick emphasizes this shift’s significance: “We’re moving beyond mere conversation to true task delegation. The AI checks in with questions, drafts, or final results while you focus elsewhere.”
Major companies are taking notice. Asana, Canva, and DoorDash are already testing these capabilities, while Replit uses Claude’s computer skills for real-time application assessment.
Looking ahead As AI transitions from chat interfaces to active computer use, the question isn’t whether AI assistants will become workplace essentials, but when and how they’ll transform our workflows. Though perhaps we’ll need to remind them that scenic photos can wait until break time.
