Will employing AI instead of humans really help companies’ bottom lines?

These days, fewer things seemingly make tech CEOs feel more alive than speculating about replacing their employees with AI.

Taken at face value, you would think the era of robots replacing human workers was already here. Block CEO Jack Dorsey even suggested so in a February letter to shareholders (following layoffs affecting 40% of his 10,000 employees), saying that with AI, “a significantly smaller team, using the tools we’re building, can do more and do it better.”

Dorsey and other CEOs have been accused of “AI-washing” their layoffs, and for good reason. Human workers currently account for the bulk of companies’ budgets. But the jury is still out on whether AI actually is more cost effective.

AI developers like OpenAI and Anthropic have bled clash for years ($5 billion and $5.3 billion, respectively, in 2024) and don’t expect to generate profits for many more. In order to turn a profit, experts have predicted that AI tools, considered confoundingly cheap, will eventually become more expensive, ending AI’s “free-trial era,” as New York Magazine tech columnist John Herrman put it.

So, will companies—some of which already pay a pretty penny for their AI tech stacks—eventually end up paying even more for these tools than employing human workers? And how might that impact their future AI tech stacks? HR Brew spoke with several experts to find out.

Understanding disruption.

Much of AI’s usefulness and ROI will depend on the tasks it can effectively and inexpensively take over. AI systems are currently able to complete 11.7% of tasks performed by humans, the equivalent of $1.2 trillion in annual wages, according to an MIT study published in late 2025. Automation in computing and technology—where AI adoption is most visible—represented just 2.2% of those wages; tasks in administrative, financial, and professional services are also vulnerable.

But remember: The study looked at tasks, not jobs. Employers will have to approach AI disruption from that perspective, and that disruption might not be even across their businesses.

“It’s so different per each industry, and I would argue by organization, and I would even argue from business unit to business unit, depending on the type of work that we all do, depending on the tasks that we can do, either help complement it by a large language model or not,” Andrea Derler, principal researcher at Visier, told HR Brew. “So we have to break it down from, really, I think, a bird’s eye perspective to tasks that we do every day.”

Certain roles will experience more disruption, particularly those involving a lot of route tasks that can be scaled with automation. Take first-tier customer support agents, who typically address simple, frequent inquiries. Generative AI can complete these tasks and may be cheaper to employ than human workers, especially if a company expects the volume of inquiries to increase.

“It can easily deal with many more inquiries than a human being could, and you don’t have to add headcount, you can just have one chatbot dealing with a huge number of different things,” Matthias Oschinski, a senior fellow at the Center for Security and Emerging Technology, told HR Brew.

But many other jobs can’t be automated in their entirety.

For example, people analytics firm Visier has an AI agent called V that can help analysts respond to queries. For a theoretical analyst who makes $100,000 annually, taking 20 minutes to answer a query without AI would be more costly than using V, which can gather information more quickly, thereby costing the company less, Derler said. However, she noted that the analyst would still need to assist the end user, ask the right questions to the agent, and provide the information to the client.

“We’ve not seen, yet, big layoffs, in terms of people analysts being laid off because there are more tools now in analytics, because the job is much more than just doing this one query,” Derler said. “So we have to really break up tasks and jobs and roles into those tiny pieces.”

Similarly, some academic institutions have AI systems serve as virtual teaching assistants, answering students’ questions about coursework or deadlines more quickly than a professor might via email and without having to take time during office hours, said Ram Bala, an associate professor of AI and analytics at Santa Clara University’s Leavey School of Business. These virtual assistants, he said, aren’t replacing instructors, though.

“The change I’m actually seeing is that students are actually able to go into office hours to ask questions which they would not actually be able to resolve easily, using AI,” said Bala. For example, the student and instructor may spend that face time discussing career opportunities.

It’s one AI, Michael. What could it cost? $10?

There’s one major problem with generative AI: the costs. And the math ain’t mathin’. While companies are spending billions on their AI tech stacks (as much as $37 billion in 2025, according to one analysis from VC firm Menlo Ventures), AI tools are currently as cheap as they’ll ever be.

AI companies, in an effort to remain competitive with other developers, have kept their prices low to attract and retain customers, often to the detriment of their bottom lines. GitHub’s coding assistant Copilot, for example, costs as much as $21 per month, a price the platform and its parent company, Microsoft, agreed on to attract users, Erica Brescia, GitHub’s former COO, told Wired. Otherwise, she said that the tool’s value to software developers should have it priced at 100 times that amount.

Experts have speculated that AI developers are following the typical consumer tech startup playbook and will eventually raise prices. That was the case with Uber. In its early, VC-funded years, the ridesharing company subsidized its fares to compete with taxis. But between 2018 and 2021, the average cost of an Uber ride increased by 92%, according to a Rakuten Intelligence analysis. The company did not turn a profit until 2023—15 years after its founding.

There are key differences between the playbooks used by Uber and AI companies. Uber has just one main competitor, while AI companies have a lot of competition, not to mention clients that can use open-source models to develop their own tools that are cheaper to access than closed-source models. But there could be parallels in their strategies.

AI developers could take an approach where “I price the API really low, I create a lot of hype, I get people to have some FOMO that everyone else is doing this, and I need to as well,” Jason Schloetzer, an associate professor at Georgetown University’s McDonough School of Business, said. “Then later on is when the providers would pull the rug on the pricing, and then there’s this lock in where we’ve redesigned our workflows to have these things in there. They’re switching costs. I can’t easily go out and hire teams and teams of employees to come back in.”

This strategy may have some merit. This past month, OpenAI CEO Sam Altman said his company may change its pricing model so it’s metered by usage, like a utility. OpenAI currently charges individual clients $200 per month for its ChatGPT Pro plan. The business plan for small to midsize orgs costs $30 per user per month (or $25 per user per month for annual billing), while enterprises, it was rumored in early 2024, are billed around $60 per user per month, with a 150-seat minimum. But Nick Turley, head of ChatGPT, told the BG2 Pod podcast this month that the subscription model was introduced as a means to cap usage and isn’t sustainable long term. “It’s possible that in the current era, having an unlimited plan is like having an unlimited electricity plan,” he said.

Katya Laviolette, 1Password’s chief people officer, recently shared with HR Brew her concern about AI developers’ pricing strategies.

“What we’re seeing, from a top line bottom line, is the costs of these tools are very expensive, seat-based or token-based,” Laviolette said, noting that CFOs have become increasingly concerned about the level of operational expenditures dedicated to AI investments. “What I am predicting in companies is that we’re all in the curiosity, we’re all in the experimentation, the Anthropics and the OpenAIs are basically going to bring the hammer down, and they’re only doing contracts that are like one year duration because they need lock in because they need to make money.”

That will force HR leaders to be more “particular” about the tools they’re using, Laviolette said. “So you let people experiment, but then you lock down,” she added.

It’s also worth noting that these costs don’t take into consideration those associated with retaining highly skilled engineers capable of developing, quality testing, or customizing these tools for the organization, Derler noted, as well as integrating the technology into different workflows and training staff, as Schloetzer separately noted.

However, Bala said that while the economics of AI “looks a little out of whack right now,” many developers are making a lot of fixed investments as they develop the infrastructure to operate the technology, and the efficiency gains will lead to a payoff. Plus, he noted that the foundation model companies (such as OpenAI, Google, and Anthropic) have improved their products while only modestly raising prices.

“When you make all these big investments upfront with the hope that you’re going to like, make use of it, the economics does look a little terrible in the short run, but I think in the medium to long run, in my views, that the economics will sort of work out,” Bala said, adding, “and I think the bigger challenge might actually be bottlenecks in just ensuring that we are able to deploy the AI fast enough.”

Does it matter?

Generative AI is not the only type of AI out there, and employers that deem it too expensive may seek out other, more sustainable alternatives, Oschinski said. Still, employers could be willing to shell out funds for their AI tech stacks, regardless of the price hikes—that is, as long as the output is good quality.

“What I’ve gathered so far from being in the technology industry is, if it’s possible, if it’s technically possible, usually, people want to make it work if it’s valuable,” Derler said.