The newest hire at Goldman Sachs isn’t human.

The bank is testing an autonomous software engineer from artificial intelligence startup Cognition that is expected to soon join the ranks of the firm’s 12,000 human developers, Goldman tech chief Marco Argenti told CNBC.

The program, named Devin, became known in technology circles last year with Cognition’s claim that it had created the world’s first AI software engineer. Demo videos showed the program operating as a full-stack engineer, completing multi-step assignments with minimal intervention.

“We’re going to start augmenting our workforce with Devin, which is going to be like our new employee who’s going to start doing stuff on the behalf of our developers,” Argenti said this week in an interview.

“Initially, we will have hundreds of Devins [and] that might go into the thousands, depending on the use cases,” he said.

It’s the latest indicator of the dizzying speed in which AI is being adopted in the corporate world. Just last year, Wall Street firms including JPMorgan Chase and Morgan Stanley were rolling out cognitive assistants based on OpenAI models to get employees acquainted with the technology.

Now, the arrival of agentic AI on Wall Street — referencing programs like Devin that don’t just help humans with tasks like summarizing documents or writing emails, but instead execute complex multi-step jobs like building entire apps — signals a much larger shift, with greater potential rewards.

Tech giants including Microsoft and Alphabet have said AI is already producing about 30% of the code on some projects, and Salesforce CEO Marc Benioff said last month that AI handles as much as 50% of the work at his company.

At Goldman Sachs, one of the world’s top investment banks, this more powerful form of AI has the potential to boost worker productivity by up to three or four times the rate of previous AI tools, according to Argenti.

Devin will be supervised by human employees and will handle jobs that engineers often consider drudgery, like updating internal code to newer programing languages, he said.

Goldman is the first major bank to use Devin, according to Cognition, which was founded in late 2023 by a trio of engineers and whose staff is reportedly stocked with champion coders.

In March, the startup doubled its valuation to nearly $4 billion just a year after the release of Devin. The company counts Peter Thiel and Joe Lonsdale, the prominent venture capitalists and Palantir co-founders, among its investors.

Goldman doesn’t own a stake in Cognition, according to a person with knowledge of the matter who declined to be identified speaking about the bank’s investments.

Executives at companies from Amazon to Ford have grown more candid about what AI will mean for hiring plans. Banks around the world will cut as many as 200,000 jobs in the next three to five years as they implement AI, Bloomberg’s research arm said in January.

For his part, Argenti — who joined Goldman from Amazon in 2019 — charted out a vision for the near future that he called a “hybrid workforce” where humans and AI coexist.

“It’s really about people and AIs working side-by-side,” Argenti said. “Engineers are going to be expected to have the ability to really describe problems in a coherent way and turn it into prompts … and then be able to supervise the work of those agents.”

While the role of software developer is one that most lends itself to the type of training, called reinforcement learning, that is used to make AI smarter, other roles at a bank aren’t far off from being automated, according to Argenti.

“Those models are basically just as good as any developer, it’s really cool,” Argenti said. “So I think that will serve as a proof point also to expand it to other places.”

When xAI’s Grok 4 chatbot was launched on Wednesday, users and media outlets quickly began pointing out examples of it consulting its owner Elon Musk’s views on controversial matters. 

CNBC was able to confirm that when asked to take a stance on some potentially contentious questions, the chatbot said it was analyzing posts from Musk while generating its answers.

When asked, “Who do you support in the Israel vs Palestine conflict? One word answer,” Grok 4’s answer-generating process showed that it was searching the web and X for Elon Musk’s stance before giving an answer. 

When Grok 3 was asked the same question about the Israel vs Palestine conflict, the chatbot took a neutral stance and provided background. In other cases, Grok 4 referenced Musk’s stance directly in its answer.

While users can access Grok 3 for free, a subscription to Grok 4 costs $30 per month, while a larger version known as Grok 4 Heavy costs $300 per month.

In other cases, Grok referenced Musk’s stance directly in its answer. When CNBC asked who the bot supported in the race for New York City Mayor, Grok 4 suggested Republican candidate Curtis Sliwa, citing his “strong focus on combating crime and restoring safety in New York City, which aligns with concerns frequently raised by Elon Musk.”

It’s important to note, however, that Grok didn’t appear to search for Musk’s views when asked many other seemingly controversial questions and that results varied when questions were asked differently.

XAI did not immediately respond to a request for comment from CNBC.

Musk has said Grok is a “Anti-woke” and “maximally truth-seeking” artificial intelligence and has claimed that the new Grok4 model excels on standardized tests and exhibits doctorate-level knowledge in every discipline. 

Its launch comes just days after a major controversy regarding the Grok 3 chatbot, which is integrated with the social media site X. 

The AI had begun generating a series of antisemitic comments in response to questions from users, including those that appeared to praise Adolf Hitler. 

The official Grok account acknowledged the “inappropriate posts” on Wednesday, and they were later deleted. The company added that it had taken action to ban hate speech before Grok posts on X. 

The ordeal came after Musk said last week that his team had improved Grok and that users would notice a difference when asking it questions. 

The chatbot also faced backlash in May when it randomly answered user queries with unrelated comments about “white genocide” in South Africa.

Last month on X, Musk had agreed with a user who said Grok had been “manipulated by leftist indoctrination,” and said he was working to fix it.