Anthropic releases AI agents that can do complex tasks independently

Anthropic AI on Tuesday launched a pair of updated AI models coupled with a new capability to independently perform computer tasks like a human would and save users’ keystrokes.

The Amazon and Alphabet-backed AI startup, together with other tech giants like Meta and Microsoft make up a group of tech firms in the generative AI arms race as they seek to outdo each other and stay ahead of competition.

Anthropic’s new model performs various tasks

A high-flying competitor to ChatGPT maker OpenAI, Anthropic revealed that it has updated its model Claude to perform a wide range of tasks on a computer search as web searches, open applications as well as input text using the mouse and keyboard.

“The new model with the ‘computer use’ function has the ability to tell AI where to take the mouse and click as well as what and how to type to complete the complicated tasks.”

Anthropic’s chief science officer Jared Kaplan.

According to a Reuters article, the ability to do tailored complicated tasks lies with the software developers acting as AI agents who can perform tailored, complicated tasks, interfering minimally to carry out multi-step actions.

Researchers have called agents AI development frontiers for better technology beyond chatbots, because they go beyond simple prose or computer code and not specific actions.

Anthropic showcased its abilities in a use case scenario using Google Search and Apple Maps to plan a sunrise outing among other programs as well as coding a basic website.

At different prices based on their performance, Anthropic offers software developers three versions of its family of AI models known as Claude.

Sonnet, the second version, and the lowest-priced version, Haiku will receive the first updates this week.

Kaplan said the new update of Haiku, the 3.5 version can generate computer code with speeds closer to the earlier version of Sonnet which was launched in June.

Anthropic to make features available to all in Q1 of 2025

Anthropic CEO Dario Amodei told Reuters in an interview in June that they were targeting to introduce the best version of Opus by the close of the year.

Amazon had early access to the model, according to Anthropic, while first-mover customers and other testers included Canva and Notion. Kaplan added that the company had been on this development since the start of the year.

Despite rolling out updates this week, the company said it hopes to make all the features be accessible to all customers during this quarter or in the first quarter of 2025.

According to Anthropic, booking flights, filling out forms, conducting online research, and filing expense reports will be part of future consumer applications. Kaplan also said that the AI startup’s target is to enable Claude to assist people solve different kinds of tasks.

“What would you do if you got rid of a bunch of hours of copy and pasting or whatever you end up doing? I would say go and play more guitar.”

Mike Krieger, Anthropic chief product officer.

Since the first release of Claude, that directly challenges OpenAI on the market in March 2023, Anthropic has become one of the most sought-after AI startups. The two AI companies are in direct competition in both enterprise and consumer markets.

Tuesday’s announcement is Anthropic’s latest move to build more virtual collaborators or agents, which is in line with its long-term strategy.

In a move to foster development, the startup in September launched Claude Enterprise, reportedly their best product since the debut of its chatbot, which is designed for organizations looking to integrate Anthropic’s AI.

For that specific technology, GitLab, Midjourney, and Menlo, are the early customers and beta testers, according to the company.