Anthropic announces AI agents for complex tasks, racing with OpenAI

Anthropic, the Amazon-backed AI startup founded by former OpenAI research leaders, announced Tuesday that it has achieved an artificial intelligence milestone for the company: AI agents that can use a computer to perform complex tasks such as a human would do that.

Anthropic is the company behind Claude – one of the chatbots that, like OpenAI’s ChatGPT and Google’s Gemini, have become extremely popular. Startups like Anthropic, alongside tech giants like Googling, AmazonMicrosoft and Meta are all part of a generative AI arms race to ensure they don’t fall behind in a market expected to reach $1 trillion in sales within a decade.

Anthropic’s new Computer Use capability, part of its two latest AI models, allows the technology to interpret what’s on a computer screen, select buttons, enter text, navigate websites and complete tasks perform through any software and surf the Internet in real time.

The tool can “use computers in much the same way we do,” Jared Kaplan, Anthropic’s chief science officer, told CNBC in an interview, adding that it can perform tasks involving “dozens or even hundreds of steps.”

Amazon had early access to the tool, Anthropic told CNBC, and early customers and beta testers included Asana, Canva and Notion. According to Kaplan, the company has been working on the tool since the beginning of this year.

Anthropic released the feature in public beta for developers on Tuesday. The team hopes to open up use to consumers and business customers in the coming months, or early next year, according to Kaplan.

Anthropic said future consumer applications will include booking flights, scheduling appointments, filling out forms, conducting online research and submitting expense reports.

“We want Claude to actually be able to help people with all kinds of work, and we think the chatbot setup is quite limited because you can ask a question and [get] context, but that’s where it ends,” Kaplan told CNBC.

Following the viral popularity of OpenAI’s ChatGPT, the industry quickly transitioned from text responses to AI-generated photos, videos, and voice. Now, startups and Big Tech alike are going all-in on AI agents.

Rather than just providing answers – the domain of chatbots and image generators – agents are built for productivity and to perform multi-step, complex tasks on a user’s behalf. And while the term isn’t clearly defined across the technology sector, AI agents are seen as a step beyond chatbots in that they are typically designed for specific business functions and can be scaled to large AI models. Think JARVIS, Tony Stark’s versatile AI assistant from the Marvel Universe.

Grace Isford, a partner at venture capital firm Lux Capital, told CNBC in June that there has been a “dramatic increase” in interest among technology investors in startups focused on building AI agents. They have collectively raised hundreds of millions of dollars and seen their valuations soar alongside the broader generative AI market.

Microsoft CEO Satya Nadella said during an earnings call earlier this year that he wants to offer an AI agent that can perform more tasks on behalf of a user, although there is “a lot of execution ahead.” Meta and Google executives have also touted their work in pushing AI agents to become increasingly productive.

Anthropic has become one of the hottest AI startups since it released the first version of Claude in March 2023, a product that competes directly with OpenAI’s ChatGPT in both the enterprise and consumer markets, without any consumer access or major fanfare. Supporters include Google, Salesforce and Amazon. Since January, it has introduced iOS and Android apps, a team subscription for businesses and an international expansion into Europe.

″[We’re] We’re moving to a world where these models will act much more like virtual workers than virtual assistants,” Scott White, product manager at Anthropic, told CNBC in September.

Anthropic’s announcements on Tuesday are the latest step in its long-term strategy to build those virtual employees, or agents.

Last month, Anthropic introduced Claude Enterprise, its biggest new product since the chatbot’s debut, designed for companies looking to integrate Anthropic’s AI. Beta testers and early customers of the enterprise product included GitLab, Midjourney and Menlo Ventures, according to the company.

With Claude Enterprise, customers can upload relevant documents with a much larger context window than before – the equivalent of 100 30-minute sales calls, 100,000 lines of code or 15 full financial reports, according to Anthropic. The plan also allows for “activity feeds” for super users within a company to show those newer to AI how others are using the technology, White said.

The launch of Claude Enterprise followed Anthropic’s June debut of its more powerful Claude 3.5 Sonnet, and the May rollout of its “Team” plan for smaller businesses.

In June, Anthropic also announced “Artifacts,” which allows a user to ask their Claude chatbot to generate a text document or code, for example, and then open the result in a special window.

Artifacts, or “workspaces” that allow users to “see, edit and build on Claude’s creations in real time,” White told CNBC in September, will allow Anthropic’s clients to create marketing calendars, enter sales data, create dashboards or forecast, draft code for functions, write legal documents, summarize complex contracts, automate legal tasks and more.

Shortly after Anthropic’s debut of Teams in May, Mike Krieger, co-founder and former chief technology officer of Meta-owned Instagram, joined the company as chief product officer. Under Krieger, the platform grew to 1 billion users and the engineering team grew to more than 450 people, according to a press release. OpenAI’s former security lead, Jan Leike, joined the company that same month.