Software stocks sell-off "trigger" and new products launched: AI agents team up to work together, better at white-collar tasks

With a series of AI plugins triggering this week’s US stock market “SaaS doomsday” and widespread sell-off of tech stocks, American AI star company Anthropic announced the launch of the latest version of its flagship large model, Claude Opus 4.6, on Thursday.

The company also made bold claims that after AI has completely transformed the way software programming works, it’s now time to venture into a broader workplace.

Following Anthropic’s new product release, the stock prices of several financial service companies plummeted. As of press time, FaceSet dropped over 8%, hitting a new low since March 2020; Thomson Reuters fell over 7%, reaching a new low since 2021.

(FaceSet weekly chart, Source: TradingView)

One of the most notable new features in the new version is Anthropic’s release of a function called “AI Agent Teams”, which allows a single, larger task to be broken down into several segments and assigned to different agents for completion.

The company states that this feature enables users to coordinate multiple AI agents working together. One agent acts as the “team leader,” responsible for coordinating tasks, assigning work, and consolidating results. Each “AI employee” works independently, with its own context window, and can communicate with each other—for example, seeking approval from the “AI leader” for certain operations. Human users can also converse individually with each “AI employee” to provide additional instructions.

In practical applications, users can also try having “AI employees” process the same task in parallel and attempt to refute each other’s conclusions, leading to better results through “AI debate.”

Anthropic’s product director Scott White said that this new feature is akin to having a talented human team working for you. Breaking down the agents’ tasks not only allows them to coordinate in parallel but also speeds up the work.

The Opus 4.6 upgrade also introduces a context window of 1 million tokens, meaning it can handle larger codebases or more extensive documents, such as financial files and research reports.

After success in programming, Opus 4.6 is also shifting focus to more general white-collar daily tasks, such as financial analysis, research, creating spreadsheets, and PowerPoint presentations.

In the GDPval-AA test, which evaluates AI models’ performance on knowledge work tasks in finance, law, and more, Opus 4.6 scored nearly 10% higher than the industry’s second-best model (GPT-5.2).

Meanwhile, in specialized tests for financial analysis, computer usage, tool utilization, retrieval, and agent-based programming, Opus 4.6 remains an industry-leading model.

For workers, the new version of Opus also integrates Claude directly into PowerPoint. Now users can use Claude in Excel to handle complex data, then transfer to PowerPoint to generate presentations.

White said that over the past year and a half, everyone has seen the transformation in the software engineering field: “Vibe coding” has started to emerge as a concept, where people can get things done based solely on their ideas.

He added, “I believe we are now almost entering an era of ‘vibe working’.”

(Source: Caixin)