AI agents lag far behind human workers. Why are tech companies laying off the humans?

AI-related layoffs ar in replete(p) swing over as tech companies adorn in unreal intelligence agents they say will take over tasks traditionally done by humans. 

But research shows the agents — autonomous software programs that use large language models to complete multi-step tasks — have a long way to go before they can reliably do those jobs.

A major AI infrastructure and software company says the agents fail to produce professionally acceptable work more than 19 times out of 20, and some analysts say the technology is overhyped and being using as an excuse to lay off workers.

Meta laid off nearly 10 per cent of its workforce last month amid a shift toward agentic AI, while Jack Dorsey's Block cut his company's staff almost in half in February, directly attributing the cuts to AI. Microsoft and Amazon have laid off thousands of workers in the last year, both mentioning shifts to AI. 

San Francisco-based Scale AI, which provides data and evaluations to governments and Fortune 500 companies, developed a Remote Labour Index benchmark to measure how well AI agents can actually perform "real-world, economically valuable remote work" from end-to-end. 

According to its research, even the best AI agents are completing tasks to "professional, client-ready" standards less than five per cent of the time. 

Testing AI agents is a relatively new field. An October 2025 study by researchers at Stanford and Carnegie Mellon universities found the agents work faster and cheaper than humans, but "produce work of inferior quality" and mask deficiencies with "data fabrication and misuse of advanced tools."

Scale's researchers assign AI agents to a broad range of tasks sourced from professional freelance platforms like Upwork — including designing leaflets or logos, editing and generating videos and developing architectural models — and compare the results to artifacts generated by human freelancers. 

The agents have the highest success rates in image generation, report writing, audio tasks and data retrieval. They're failing most commonly in complex tasks like producing architectural drawings based on certain specifications. 

The most common reason cited for failure across all projects was poor quality, described as "child-like" or amateur (about 46 per cent). More than one-third of failed submissions were incomplete, about 18 per cent were corrupt or used incorrect file formats and about 15 per cent "failed to maintain visual or logical consistency across files." 

The agents are improving, however. When Scale started its benchmark last fall, the top-scoring agent had a 2.5 per cent success rate. By March, the top score was 4.17 per cent. 

But that leaves a long road ahead, and Sehwag says agents are progressing more slowly on complex, holistic tasks. 

Those findings clash with some of the sunny messaging from tech companies.

Job boards exist exclusively for AI agents, which are sometimes marketed as full-blown human replacements.

"She outworks everyone. And she’ll never ask for a raise," boasted a controversial billboard ad from British AI company Narwhal Labs, depicting an AI agent as one-half digital and one-half young blonde woman.

Outplacement services company Challenger, Gray and Christmas found that AI led all reasons for U.S. Job cuts in March and April. In May, research and advisory firm Gartner released a survey of 350 global business executives with an annual revenue of at least $1 billion US, which found 80 per cent of those who piloted an AI or autonomous technology reduced their workforce due to automation. 

Understanding AI with Mark Daley: What is agentic AI?

Scale AI CEO Jason Droege told Semafor World Economy in April that corporate customers often ask for automation to generate savings and productivity, but he steers them away because "there’s a lot of problems that the technology’s not mature enough to solve with reliability and safety."

He said some companies are "washing" their layoffs, using AI as an excuse to reduce their workforce.

OpenAI CEO Sam Altman similarly said in February that while some AI displacement is real, some companies are "AI-washing, where people are blaming AI for layoffs that they would otherwise do."

The push for AI agents is more about selling a story than improving output, says Julie Yujie Chen, a University of Toronto associate professor who studies how digital technologies are transforming work. 

"AI requires a lot of money. It's like a cash-sucking experiment, and no companies can really predict if it will work, so they have to lay off workers to keep costs low. Sometimes using technological unemployment is just an excuse," she said. 

Companies often see their stock prices jump after announcing AI-related layoffs — Block's spiked more than 20 per cent — though the gains can be short-lived. 

"Investment in AI as an ideology, as a storytelling, has more value than keeping your workforce. That's how capitalism works," Chen said.

Meta seemed to lay this out in a memo obtained by Reuters in April. The company's CTO Andrew Bosworth told employees, upon announcing the company will start surveilling their keystrokes and mouse movements to train AI agents, that Meta is building toward a vision "where our agents primarily do the work and our role is to direct, review and help them improve."

White-collar workers are now experiencing what blue-collar workers have lived through for centuries, according to David Eliot, a PhD candidate at the University of Ottawa who researches the social and political effects of AI. He says they're effectively training machines to take their jobs.

"There's something really deeply unsettling about being involved in your own automation and your own de-skilling, about knowingly participating in the activity of reducing your value to the economy," he said.

"It's creepy."

'Artificial intelligence needs to be disarmed': Pope Leo writes

But he says a lot of small and medium-sized companies have genuinely bought into the hype and found AI "didn't work as well as they thought it would." 

An MIT report last year found that despite $30 billion to $40 billion US in enterprise investment into generative AI, 95 per cent of organizations were not seeing a financial return. 

Some AI agents, Eliot points out, have also caused high-profile failures. In December, the Financial Times reported that Amazon Web Services experienced a 13-hour interruption to its operations when an AI agent chose to "delete and then recreate" part of its environment. 

Experts say the verdict is still out on exactly how agentic AI will impact the working world in the long term — and when, if ever, it will be able to reliably fill complex human roles without strict supervision. 

"I always say, anybody who tells you they know what's going on with AI and jobs is either lying to you or trying to sell you something," Eliot said.

Digital Writer

---

---