Workers are likely to chat more with AI agents than with their human colleagues in the future, Slack chief marketing officer Ryan Gavin predicts.

The big picture: AI agents are joining the workplace whether workers want them or not.

Catch up quick: For the past year, companies have been pushing autonomous AI agents — chatbots with the ability not just to answer questions, but to act on users' behalf.

• Salesforce, which owns Slack, launched Agentforce in September with bots that serve as sales reps, service agents, personal shoppers and sales coaches.
• AI agent startup Sierra, founded by former Salesforce co-CEO Bret Taylor and former Google executive Clay Bavor, raised $175 million last year at a $4.5 billion valuation.
• According to The Information, OpenAI expects to triple its revenue in 2025, with nearly a third of that projected revenue coming from sales of its agent tools to SoftBank.

What they're saying: Gavin says AI agents will soon become as integral to the workplace as human co-workers, transforming the way teams collaborate.

• "I think that right now people are underestimating just how much the world of work is about to change," Gavin told Axios. "In just three or four or five years, I could be talking to agents as much, if not more than I'm talking to my human colleagues today."

Salesforce CEO Marc Benioff also believes in the human/bot workplace of the future.

• Benioff regularly refers to agents as the "limitless workforce." At Davos last month, he told Axios' Ina Fried that today's CEOs will likely be the last to "manage a workforce of only human beings."

Between the lines: Gavin sees AI agents as augmenting, not replacing, human workers — offering services that would otherwise be too costly.

• "What if every single employee had a human resources agent that sat right alongside them in Slack?" Gavin told Axios.
• Humans are expensive and hard to train. They ask for raises and advocate for better workplace conditions. Sometimes, they quit.

The other side: Humans like to interact with other humans.

• Whether they're working from home or back in the office, knowledge workers are lonely.
• "When you start to only interact with machines, verbal and nonverbal social behavior starts to atrophy," Thomas Plante, professor of psychology at Santa Clara University, told Axios.

Zoom in: Some professions have already been working alongside AI agents for years — though they haven't always called them that.

• Developers, for example, have always had help with coding, says Thomas Dohmke, CEO of GitHub. "We just had different names for them: workflows, pipelines, those kinds of things, right?"
• "We will have more complex agents that do more complex tasks," Dohmke told Axios. "But I don't think that it's a step change. I think it's a gradual evolution of something we've been on ever since the term DevOps was invented."

Yes, but: OpenAI launched its first two agents (Operator and Deep Research) over the past two months, and both are underwhelming right now, though they show promise.

• "The biggest challenges with AI agents in the workplace stem from their fundamental differences with traditional software — they are not deterministic so they do not work the same way every time," Tatyana Mamut, co-founder and CEO of Wayfound, a platform designed to "manage" agents, told Axios via email.
• Mamut said she and her team have been working alongside 25 AI agents for a year. "It feels a lot like managing a team of interns who are book-smart but still make frustrating mistakes on the job once in a while," she said.
• "AI agents are inconsistent and sometimes misremember things," she told Axios.

Mamut says this inconsistency is a feature, not a bug.

• "Just like AlphaGo's move 37, when an AI created a completely new way to play the game of Go and started to play the game better than any human, AI agents have the potential to achieve superhuman results through their creativity," Mamut said.

Most of the fear surrounding agents is the prospect that they might replace many human workers.

• "A lot of jobs of today will be gone tomorrow, as all of this starts to unfold," Plante says. "People are always worried about their jobs, but the reality is that throughout history, jobs come and go based on technological developments."
• "The world changes and we sort of adapt or die."

Anthropic said yesterday its latest model — Claude 3.7 Sonnet — integrates advanced reasoning capabilities for the first time.

Why it matters: While OpenAI and others have released separate models focused on longer reasoning times, Anthropic adds advanced reasoning into standard models.

How it works: Anthropic says the new model is capable of near-instant responses or extended responses that show step-by-step reasoning with users able to choose which type of response they're looking for.

• Anthropic's API users can control how long Claude spends reasoning before answering. That's important because longer reasoning time increases the cost of a query.
• Claude 3.7 Sonnet also has enhanced computer use abilities and improved coding skills.
• Anthropic said the new model will be available on all its Claude.ai plans, via its API and through cloud providers Google and Amazon.

• Elon Musk is reportedly using an AI tool to sort through responses to the email he sent government workers last week requiring them to list five accomplishments from the past week — or be fired. (NBC News)
• Anthropic is said to be finalizing a $3.5 billion financing round that would value Claude's maker at $61.5 billion. (WSJ)
• Chegg is suing Google over its AI-generated search summaries, while the online education firm also says it's exploring strategic alternatives for the company. (CNBC)

Here's what happens when you give AI a Rorschach test.