The AI That Uses Your Computer
Anthropic just deployed Claude Sonnet 4.6, and the announcement deserves attention. According to the company, this model "approaches Opus-level intelligence" — their most powerful model — while significantly improving "computer use" capabilities: spreadsheet navigation, web form filling, graphical interface interaction.
Sonnet 4.6 now replaces Sonnet 4.5 as the default model for free and Pro Claude users. It's a quiet but significant change: Anthropic believes this model is ready for the general public.
What Is "Computer Use"?
Computer use is an AI's ability to interact with a computer as a human would: move the mouse, click buttons, type text, navigate between applications.
Unlike traditional API integrations, this approach doesn't require specific connectors for each application. The AI "sees" the screen and acts on it. Theoretically, it can use any existing software.
Anthropic introduced this feature with Claude 3.5 Sonnet, but early versions were slow and error-prone. Sonnet 4.6 represents significant maturation of this technology.
Concrete Improvements
According to Anthropic, Sonnet 4.6 particularly excels at:
Spreadsheet navigation: Ability to browse, filter, and manipulate data in Excel or Google Sheets reliably.
Web forms: Automatic filling of complex forms, including those with dropdown menus, checkboxes, and conditional fields.
Multi-step workflows: Executing tasks requiring multiple applications, like "extract data from a PDF, put it in a spreadsheet, then send an email with the summary."
Coding gains are also notable. Anthropic positions this model as a premier development assistant, rivaling the latest GPT and Gemini iterations.
Why It Matters
Improved computer use has major implications for task automation:
Accessibility: Companies no longer need to develop expensive API integrations. AI can use existing interfaces.
Legacy systems: Old software without APIs can finally be automated. For companies trapped with obsolete systems, it's a potential liberation.
Democratization: Tasks previously reserved for developers (web scraping, RPA automation) become accessible to anyone who can describe what they want to accomplish.
Emerging Risks
This increased power comes with concerns:
Security: An AI capable of clicking buttons and filling forms can also, theoretically, perform unauthorized actions if manipulated.
Liability: Who's responsible if an AI agent makes a mistake — deletes the wrong file, sends an email to the wrong recipient, validates an incorrect transaction?
Dependency: As companies delegate tasks to AI agents, they potentially lose understanding of their own processes.
The Agent Race
Sonnet 4.6 fits into intense competition. OpenAI develops its own agent capabilities with GPT-4o and successors. Google pushes Gemini toward agentic AI. Microsoft integrates Copilot ever deeper into Windows and Office.
Anthropic has an advantage: their focus on safety and alignment gives them credibility others lack. For companies concerned about deploying AI agents responsibly, Claude remains a defensible choice.
Opus vs Sonnet: Positioning
The "approaches Opus-level intelligence" mention is revealing. Anthropic positions Sonnet as the everyday work model — fast, efficient, economical — while Opus remains reserved for tasks requiring the most sophisticated reasoning.
This stratification is smart. It lets Anthropic capture the mass market with Sonnet while maintaining a premium offering for demanding use cases.
Verdict
Claude Sonnet 4.6 represents an important step in AI agent evolution. Computer use improvements are tangible and open significant automation possibilities.
For existing Claude users, the update is seamless and welcome. For those still evaluating options, Sonnet 4.6 deserves a place on the shortlist.
The question is no longer whether AI can use our computers — they already can. The question now is: are we ready to let them?