Anthropic Unveils New AI to Automate Computer Tasks for Developers
Anthropic, an AI startup backed by industry giants Alphabet and Amazon, has released enhanced versions of its Claude AI models, launching a new feature designed to automate complex computer tasks for developers. This “computer use” feature enables AI to perform multi-step actions by autonomously moving the mouse, selecting fields, typing inputs, and clicking in real time, all in response to high-level instructions. Jared Kaplan, Anthropic’s Chief Science Officer, described the feature as capable of handling intricate tasks that typically demand several human interactions.
Anthropic’s new capability is targeted primarily at software developers, positioning it as a pivotal step forward in the evolution of AI agents. These agents require minimal human intervention and can carry out complex, multi-step tasks rather than limiting themselves to static text generation or simple code creation. Anthropic’s new “computer use” feature takes a significant leap in functionality by orchestrating tasks across different applications and interfaces. The company demonstrated its potential by showing how the AI could code a basic website, as well as coordinate actions across multiple applications, such as Google Search and Apple Maps, to plan an elaborate sunrise outing, underscoring the versatility and adaptability of the technology.
Anthropic’s Claude family of models now offers three tiers for developers, with each tier varying in performance and price. The recent updates apply to Claude’s mid-tier model, Sonnet, and the entry-level model, Haiku. According to Kaplan, Haiku’s new version can now generate code in ways “almost comparable” to the capabilities of the previously released Sonnet model, offering developers more choices across price points without sacrificing functionality. The company also plans to upgrade its highest-tier model, Opus, before the end of the year, with additional improvements focused on more demanding applications.
While this “computer use” feature is currently limited to Claude 3.5 Sonnet, Anthropic has incorporated safeguards to prevent misuse. These protections include restrictions to prevent the AI from being used for potentially harmful purposes, such as spam, fraud, and election interference. Kaplan acknowledged that the AI’s performance is still evolving and can make occasional errors, which the company is actively working to address.
To help guide the development and future capabilities of its products, Anthropic is gathering feedback from its business customers. Mike Krieger, Instagram co-founder and Anthropic’s newly appointed chief product officer, emphasized the importance of customer input in refining and expanding the functionality of this new feature. He noted that the company’s dedicated labs team is already exploring ways to bring the “computer use” capability to individual consumers. Krieger himself sees great potential for individual users to benefit from the AI’s ability to automate complex interactions.
Anthropic’s move into AI-powered task automation marks an important moment in the industry, as the race to build sophisticated, minimally supervised AI agents continues. By equipping Claude with this new automation capability, the company demonstrates how AI can handle real-world tasks with a blend of autonomy, flexibility, and adaptability, serving as a powerful productivity tool for developers and potentially, in the future, for consumers.
You can also check our news article Salesforce Urges UK Government for Tailored AI Regulations
