In 2007, Luke Arrigoni, an AI entrepreneur, earned $63,000 at his first job as a junior software developer. Today, he says AI tools that write better code than he did back then cost just $120 annually.

The numbers don’t sit right with him. Arrigoni, who runs Loti AI, a company that helps Hollywood stars find unauthorized deepfakes, worries that underpriced AI tools encourage companies to eliminate entry-level roles. He wants to flip the incentive structure so people’s careers don’t end before they begin. “If you make the AI systems more expensive, then you have an economic incentive to hire someone that is starting out,” he says.

AI transforming—or altogether eliminating—jobs has become a perennial anxiety. But the concern is gaining new urgency as demand for AI agents grows. Those AI systems can now make sales calls and write software code, work that was once reserved for humans.

So far, the situation isn’t dire. Hiring platform ZipRecruiter estimates that this year, summer internships in the US rebounded to roughly the same level as they were before the pandemic.

But that might change in the near future. At the Snowflake Summit in San Francisco last week, OpenAI CEO Sam Altman compared current AI tools to interns. The next-generation technology would be like a more “experienced” worker, he said. In some companies, managers have already started overseeing “a bunch of agents” the way they traditionally have “relatively junior employees,” Altman claimed.

OpenAI has talked about mitigation efforts like reskilling programs to stave off a potential jobs crisis—but it hasn’t mentioned charging higher prices for its services to slow the transition to AI work.

That’s what has Arrigoni on edge. Even after accounting for the priciest add-on features, AI coding agents cost a fraction of a junior engineer. If inexperienced workers can’t get a gig, Arrigoni believes, they might not gain the expertise needed to lead future teams—whether human or machine.

OpenAI did not respond to a request for comment.

“Less Than Human”

AI pricing has fluctuated since ChatGPT launched as a free chatbot in 2022 and triggered an AI boom. Generally, many AI companies still offer free tiers for limited use, and prices for basic tiers have declined. Top-tier plans for the newest features have grown pricier, though not to the point of generating profits for the companies offering them—or deterring adoption.

Startup executives and pricing consultants attribute low prices to intense competition among AI purveyors. “Their only way to win is mass adoption,” says Ajit Ghuman, CEO of pricing strategy company Monetizely. That means AI companies need to charge the same affordable prices as their rivals. Unless electricity or GPU shortages become major problems, or one company corners the AI market, it’s difficult to see prices rising significantly, Ghuman says.

Decagon, a San Francisco startup that sells a customer service chatbot used by retailers and tech companies, charges $1 or less per conversation—roughly half the cost of human support. In some cases, the chatbot may be more effective than a person, but Decagon believes its clients would never pay more for it. “The reason to invest in AI is efficiency,” CEO Jesse Zhang says. “You’re going to be less than human labor. That’s kind of like the point of technology.”

Zhang says his company makes money on each individual conversation after excluding certain overhead costs, but he declined to comment on the startup’s overall profitability. With $100 million raised from venture capitalists including Andreessen Horowitz and Accel, Decagon has the flexibility to prioritize growth over profitability. “Whether we could be pricing more, it’s always like a ‘what if?’” he says. “But in general we’re pretty happy right now.”

“So Cheap”

Erica Brescia, a managing director at the investment firm Redpoint Ventures, had an epiphany about AI agent pricing last month. The $250 price tag on Google’s new AI Ultra plan astounded her. “All this is so cheap,” she recalls thinking. “It’s disproportionate to the value people are getting.” She felt a price of at least double would make more sense. (That same week, Nvidia CEO Jensen Huang told Stratechery that he would hire an AI agent for $100,000 per year “in a heartbeat.” )

Previously, Brescia worked as the chief operating officer of GitHub, which helped set the bar for AI pricing. GitHub’s Copilot coding assistant started at $10 a month in 2022, months before ChatGPT’s debut. Brescia says GitHub went with a price that would attract a critical mass of users. The goal was gathering data to improve the service, and GitHub’s parent company, Microsoft, didn’t mind taking a loss on the new tool to make that happen. In reality, a price 100 times higher would now better reflect the value Copilot provides to software developers, Brescia estimates. (GitHub chief operating officer Kyle Daigle tells WIRED that the company’s goal is to support, not replace, developers and that “pricing reflects a commitment to democratizing access to powerful tools.”)

Today, Copilot tops out at $21 a month. And similar tools have followed its lead, including Zed, which has received $12.5 million in funding from Redpoint and others. In May, the company started charging a minimum of $20 a month for an AI-assisted code editor it built from the ground up.

Zed CEO Nathan Sobo expects AI companies to charge more over time because the current pricing models aren’t sustainable. But relative to humans, he wants to keep AI agents affordable so anyone can use them to augment their work, develop better software, and create new jobs. “I want as much intelligence at my disposal at as low a cost as possible,” he says. “But to me, included in that is potentially a junior engineer using this technology, ideally at as low a cost as possible.”

Decagon’s Zhang feels the same way about AI coding tools. “Would we pay more? Marginally? Yeah,” he says. But “$2,000? Probably not.” He adds “the hunger for good engineers is infinite.”

AI entrepreneurs suggest that agents could command higher prices if they were easier to set up and more reliable to use. For instance, Nandita Giri, a senior software engineer who has worked at Amazon, Meta, and Microsoft, says she would pay thousands of dollars annually for an AI personal assistant. “But strict conditions apply—you can’t get frustrated by using it,” she says.

Unfortunately, that day feels far away. As a personal project, Giri tried developing an AI agent that could prevent psychological burnout. “It just canceled all my meetings,” she says. Certainly a solution, but not the ideal one.

Now, some companies are hiring “AI architects” to help oversee agentic systems and cut down on gaffes. The question is who will occupy those roles in the future if early-career workers are cut off from opportunities today. Simon Johnson, an economist at the Massachusetts Institute of Technology, doesn’t expect companies to take into account the social cost of career disruption in making their pricing decisions. He suggests governments lower payroll taxes for entry-level roles to encourage hiring. “The right lever to pull is one that reduces costs to employers,” Johnson says.

Arrigoni is choosing a third path. At Loti AI, he has prioritized steadily hiring junior engineers and hasn’t employed AI coding tools. If the job apocalypse comes, “I don’t want to be at fault,” he says.