MW Tech is now betting on AI 'agents.' Will they boost stocks or take your job?

By Jeremy C. Owens

Software companies like Salesforce are marketing their newest generative artificial-intelligence products as 'agents' and 'digital labor' - but analysts doubt agentic AI's ability to take jobs or make a lot of money anytime soon

Software companies have been promising that generative artificial intelligence would produce big sales gains for two years now, while trying to downplay any immediate negative effects on human workers.

Those big sales gains have yet to fully materialize, and the software is still struggling to live up to the hype. But that isn't stopping executives from making the same promises about the next iteration of generative AI - with one noticeable change. Now they aren't shy about framing the Next Big Thing as direct competition to people's jobs.

That Next Big Thing is AI "agents." If you run in tech circles, you've heard the phrase "2025 is the year of agentic AI" more than "Happy New Year" in recent weeks. Basically, agents are the next generation of software bots, able to perform more actual actions instead of just parroting information found on the internet.

"I think the easiest way to think of agents is they're kind of yesterday's bots on steroids," said Maribel Lopez, founder and principal analyst for Lopez Research.

While Microsoft Corp.'s $(MSFT)$ Copilot has become the standard-bearer for the first generation of generative-AI software, Salesforce Inc. $(CRM)$ has beaten Microsoft to market on agents. Agentforce 2.0 is rolling out now, and Chief Executive Marc Benioff has been doing the media rounds promising that it is a "multitrillion-dollar market."

He made that claim in an interview with the On Watch by MarketWatch podcast earlier this month, and also tested out tech's agentic-AI sales pitch: "Digital labor." And Benioff made clear that he is pitching Agentforce to customers by saying that it can save costs on certain human workers.

When asked about the monetization of Agentforce, Benioff said he has used the technology on Salesforce's help desk, and it was able to halve the number of customer-support requests that required human assistance. In his recounting, he said he was able to reduce the number of support workers and spend the savings elsewhere, and believes his customers could have similar results.

"There's just less for the support organization to do because of what we've done already with help.salesforce.com and by delivering an agentic layer for our customers. And I think that's going to be an example for other customers to do this. And that, I think, is very cool," he said.

When told it's possible not everyone would think his story was "cool," especially workers who would face new competition, Benioff did not back down.

"The labor market, as you know, is very stagnant," he said. "We have declining birth rates in a lot of countries. We don't have people to hire. There aren't people to fill a lot of these positions, and it's hard to grow your business. And now it's going to be much easier to grow your business because you're going to be able to do it with digital labor."

Salesforce is not alone in framing agents as "digital labor" or a "digital workforce." International Business Machines Corp. $(IBM)$ is also using the terms to market some of its AI solutions. And in the unofficial kickoff to the tech sector's new year, the opening keynote of the CES trade show, Nvidia Corp. $(NVDA)$ CEO Jensen Huang took 10 minutes to discuss his company's solutions for agentic AI, describing agents as "effectively a digital workforce."

Wall Street, after driving up tech valuations in hopes of big AI returns since the debut of OpenAI's ChatGPT, doesn't seem to be immediately falling for the agent hype yet, though. Doubts about Salesforce's ability to monetize agents, and even about whether these are actual "agents," have surfaced.

One of the loudest critics has been D.A. Davidson analyst Gil Luria, who has written in-depth notes about AI agents and their current financial opportunity in the software sector. When MarketWatch reached Luria this week, he had obviously had enough of the marketing behind AI agents, taking the time to roast Salesforce's ubiquitous commercials for Agentforce that feature Matthew McConaughey lamenting the lack of AI assistance at a clothing boutique and sidewalk cafe.

Luria noted that small cafes and clothing boutiques are far from Salesforce's type of customers, and that the problems the ads describe can be solved with the basic customer-relationship-management software that Salesforce has long sold.

"There's no way they would use AI for that," he said. "If [Salesforce] can't even come up with an ad to communicate what these tools are doing, [that] tells you how far they are from tools that add so much value that they're going to take off."

'That's just autopilot, that's not an agent'

Luria's core problem with the AI agents that are making it to market this year is that they are not really agents. In academic terms, agents are described as artificial-intelligence products that actually have agency to make actions based on advanced knowledge and understanding. Luria believes what Salesforce is selling is more of a middle ground between bots and actual agents that may be years away in the future.

To illustrate this point, Luria used email as an example. The current generation of AI software, like Copilot, will react to an email by generating a potential response in a reply that the user can read, edit and choose to send. He said the current generation of AI agents would prepopulate a similar response, but just send it instead of alerting the user and seeking input.

"That's just autopilot, that's not an agent," he said. "An agent is something that would open the email and would have the business logic to understand what's there, know when to forward it to you, know when to respond automatically, and execute beyond the logic that is programmed ahead of time."

Luria does not trust these "autopilots" because AI is still generating garbage far too often.

"There's only a certain amount of accuracy here. AI still makes so many mistakes that are so egregious that we are not ready to take our eyes off of that," he said. "But that's the set of products that are being sold to the market this year."

And that's a key reason why any suggestion that agents will be able to immediately replace human workers appears premature at best.

"Nobody's whacking 30% of their staff this year because they put in AI. Let's just call it - that's not happening," Lopez Research's Maribel Lopez said. "And the reason that's not happening is, let's just say they wanted to whack 30% of their staff and put in AI. They probably still wouldn't think it's accurate enough."

There are certain jobs that will be affected more than others in the near-term, though - including the types of customer-service workers that Benioff spoke of, as well as software developers. Meta Platforms Inc. $(META)$ Chief Executive Mark Zuckerberg has already said that he plans to cut midlevel coders this year, and Benioff has said that he does not plan to hire any more.

Lopez said that is because agents and other AI software are already able to make low-level coders as productive as those with years of experience.

"With AI, mediocrity is death," she said. "You have to either not know much and be cheap, or you have to know a lot and be really good at what you do. Being in the middle is not going to serve you well as an individual moving forward from a career-path standpoint."

But both Lopez and Luria pointed out that customer service and software development are both professions that have long struggled to find enough workers to satisfy demand. And, in a refrain that should be familiar to those following the "AI boom" so far, they said that widespread effects on labor are still years away.

"In 10 to 20 years, a lot of white-collar, administrative tasks will be automated using the tools that we've developed over the last few years. I believe that to be true and it's going to be truly transformative and drive a productivity boom across the global economy," Luria said. "That's 10 to 20 years, though. What we have today are some niche applications where this technology is very useful, but that's it."

'People just have to pay more for their enterprise software'

For investors, the most important question is not what jobs will be affected, but whether the money is going to start flowing. That seems to be a long-term question more than one that will be reflected in corporate income statements in 2025.

Benioff disclosed in his interview with MarketWatch that Agentforce had produced more than 3,000 signed contracts that would generate revenue, with some reaching eight-digit figures. But Salesforce would not reveal how many of those guaranteed a specific level of revenue and how many were "freemium" deals - contracts that guarantee little or no revenue, instead charging based on usage. Benioff had previously disclosed at a December launch event that Salesforce had signed more than 1,000 non-"freemium" contracts.

And the difference there is important. Salesforce has long focused on subscription pricing, but with Agentforce, it is focusing on consumption-based pricing - similar to cloud computing or data tools like Snowflake (SNOW) that charge based on how much customers use the software. Many companies have focused on bringing down their cloud-computing budgets in recent years because their bills had ballooned with greater usage.

"There will be a transition between a subscription model and a utility/consumption model - that does stand to reason. We would argue that that transition will happen over many years, if not decades, specifically because it's easier for companies to monitor subscription costs than consumption costs," Luria said.

(MORE TO FOLLOW) Dow Jones Newswires

January 17, 2025 09:15 ET (14:15 GMT)

MW Tech is now betting on AI 'agents.' Will they -2-

"And so this whole notion may make sense in the long term, but in the short term, it's going to mean customers are going to be very careful about where they deploy these tools and very deliberate to deploy them only in the highest instances, so the costs don't get out of control," he added.

Enterprise-software customers are also wary of increasing their overall software spending, a potential reason for the framing of agents as "digital labor." Benioff brought up the term in response to a question about customers' reticence to spend more on software; his argument was that agents should be looked at separately from the rest of enterprise software.

The reason for that is not to directly compete with human labor for jobs, but to compete with the budgets of other departments that are likely spending more on human paychecks, Lopez said.

"What this translates into that nobody wants to say, but is the reality of the situation, is that people just have to pay more for their enterprise software because they can't get good quality AI enterprise software without paying more for it," she said. "If [Benioff] can convince somebody that they're buying bots that are 'digital labor,' it becomes an HR thing and not just an IT thing. ... They want to tap the marketing budget in some cases, depending on what the agents are. They want to tap the HR budget in other cases, depending on what the agents are. But it's really a tech-software sale."

While doubting the near-term effectiveness and monetization of agents, both analysts have little doubt that AI success will come for Salesforce and other software companies in the future.

"It's interesting that it is Benioff that's having this conversation, because this is the same conversation he had with CFOs and CTOs 10, 15, 20 years ago," Luria said. "He was telling them [that] if you put in Salesforce to manage your customer-relationship management, instead of doing it through Excel and Outlook, you're going to save all these costs on overhead ... and you'll save money.

"And that is exactly what he ended up convincing all these CFOs and CTOs, but it took him 10, 15, 20 years to do that, right? And that's where we're at now," he added. "Yes, 'digital labor' is a real thing that will emerge over the next few decades - but convincing a CTO right now to take that on when these are entirely unproven tools is a very uphill battle."

-Jeremy C. Owens

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

January 17, 2025 09:15 ET (14:15 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.