Businesses are still waiting for a true AI productivity boom

Companies are being offered a golden ticket: invest in AI and your company’s productivity will skyrocket. Costs will go down. Workers will produce more. It sounds too good to be true.

That’s how it feels at the moment.

On the other hand, there are others like software engineer Iren Azra Zou. She says that thanks to Anthropic’s Claude Code, she can now complete software engineering tasks in a day that previously took her a week. “This saves us an incredible amount of time,” said Zou, who works at trucking logistics startup Double Nickel.

Productivity gains for businesses and the economy are less obvious. More code doesn’t necessarily mean better products or features, but while it can mean significantly more spending, AI’s productivity gains haven’t translated as neatly into areas other than coding.

For people like Sarthak Gupta, a data scientist at Amazon, AI is actually creating more jobs, at least in the short term. He is working longer hours as part of what he calls the “automation phase.”

“That upfront investment is real,” Mr. Gupta said. He builds pipelines, integrates AI tools, and onboards existing workflows to new systems. “The bigger liberation is not the velocity of a single task, but the same pipeline continuing to deliver results month after month, quarter after quarter, every time the work is repeated,” said Gupta.

Gupta and Zuo’s experience shows that the AI ​​productivity disconnect is deepening. Although some employees are completing tasks faster, researchers say advances in AI have not consistently translated into increased productivity, revenue or profits for companies. And companies facing pressure to show that large-scale AI spending is worthwhile are racing to prove that individual efficiency can be scaled.

Improved productivity

Leaders passionately discuss AI and productivity. Analysis from business intelligence platform Business Insider AlphaSense compared how often the word “AI” (or its synonyms) appears near the word “productivity” in the earnings calls of major companies. In total, these words appeared 637 times during the second quarter earnings call, an increase of approximately 25% year-over-year.

Labor productivity, which measures the efficiency of workers across the economy, has soared in recent years. U.S. productivity growth has outpaced pre-pandemic trends since 2020 and has further strengthened since the end of 2022, around the time ChatGPT was released.

But these advances aren’t necessarily due to AI. About 90% of companies actively using AI reported that AI had no impact on productivity over the past three years, according to a February National Bureau of Economic Research research report based on a survey of about 6,000 executives.

Researchers instead point to other explanations for the recent productivity spike, including remote work, increased job changes in 2021 and 2022, and changes in the composition of the workforce.

“So far, the impact of AI on productivity appears to be small, with no significant change in aggregate productivity growth,” Mark Zandi, chief economist at Moody’s, told Business Insider.

A new Wharton paper by Jessica Wachter and Jonathan Wachter argues that high-tech companies are spending as if they expect such a productivity boom to materialize, but if it doesn’t, “the current buildup would be the largest misallocation of capital in history.” They warn that some big tech companies could be at risk of bankruptcy if they don’t improve productivity quickly.

McKinsey senior partner Alexander Skalevsky said the “artificial intelligence paradox” persists because many companies don’t understand how to scale AI across their operations.

While workers often report increases in individual productivity, and companies often see promising results in pilot projects, it’s much more difficult to translate those individual gains into company-wide improvements. Suharewski said part of the challenge is getting employees to adopt the technology and learn how to use it effectively.

For example, Uber Chief Operating Officer Andrew McDonald said last month that there is no direct correlation between increased use of AI and “convenience features for consumers.”

His comment started the calculation of token max. Tokens, which are units of text data that are processed by AI models, and the way they are priced, the workers who consume them, do not necessarily make companies more productive, but instead end up racking up sometimes huge bills. Enrique Dans, a professor of technology and innovation at IE University specializing in business in Spain, said this was a perfect example of the indicators being wrong.

“When a metric becomes a goal, it’s not a good metric anymore,” Dans told Business Insider, adding, “It’s not about measuring people’s productivity according to the number of tokens they spend. That’s ridiculous. The metric should be, ‘What did you accomplish? What did you accomplish?'”

While many workers are still figuring out how to use AI effectively, Michael Feroli, JPMorgan’s chief U.S. economist, said the skills needed to use large-scale language models may require less training than previous technologies, and productivity gains from AI could materialize faster than in past technology cycles.

“In some cases it could be sooner,” he said, adding: “It could be years, not decades.”

Unpleasant AI middle

AI’s greatest evangelists have predicted a utopia of abundant productivity, booming GDP, universal basic income sustaining human life, and forgetting the four-day weekend and working as we know it.

Elon Musk predicted that in 2023, “There will come a time when we won’t need a job.You can have a job if you want personal satisfaction, but AI will do everything for you.”

It is clear that we are not there yet and probably never will be. Instead, we are stuck in the middle of an unpleasant AI. GDP remains strong (but not abnormal), labor force participation among workers in their prime years is on track, and this article was written by a human.

This is not to say that AI is not impacting the labor market. More and more companies are citing AI when cutting staff or slowing hiring. Some say the job cuts are to finance AI investments or to anticipate productivity gains that haven’t been fully realized.

“AI productivity gains will happen slowly over time,” Zandi said. He doesn’t think we’ll see a major AI boost in economic data until at least the late 2020s or early 2030s. “I don’t think there will be mass layoffs or job losses. There will be mass job losses in certain industries, but there will be job gains in other industries. The net should be a reasonably well-coordinated labor market.”

Companies are also making headcount decisions for reasons less related to AI, such as overemployment during the pandemic and economic uncertainty related to inflation, tariffs, and the Iran war.

Software engineering roles could be an early sign of how these dynamics will impact other white-collar roles, especially since so many coding tasks can be offloaded to AI tools. So, even though tech companies are laying off employees everywhere, why hasn’t there been a significant drop in job openings for software engineers recently?

One theory is that AI is causing companies to generate far more code than ever before, and this output still needs to be managed and fixed by software engineers.

“Someone has to understand what is being built, maintain it, fix any security issues that arise, and upgrade the underlying systems,” Aaron Levy, CEO of cloud storage company Box, said in a June 1 post. “That’s what it’s all about.”

Similarly, Silicon Valley investor David Sachs said coding has become a “breakthrough use case for AI,” citing increased activity on developer platform GitHub.

“The fact that the demand for software engineers is increasing, not decreasing, should cast doubt on the whole ‘AI will cause mass unemployment’ narrative,” he wrote in X.

AI Spreadsheet Moment

Companies want to demonstrate that AI is worth investing in. Workers want to prove their worth. The continuing bottleneck is that organizations are building new infrastructure on the fly, but so far business rules have not been rewritten.

It’s a moment that reminds me of when spreadsheets were first introduced to the workplace. When Excel’s predecessor, Lotus 1-2-3, was released in 1983, it suddenly and fundamentally changed how accountants and bookkeepers worked.

“AI has not yet caused the job collapse that some predicted. Just like spreadsheets and email before AI, this technology will ultimately make workers more productive,” the outplacement firm Challenger said in a June report.

Spreadsheets didn’t become the backbone of the world’s financial system overnight, but it’s now hard to imagine a workplace without Excel. AI could become another fundamental tool in the workplace. We just haven’t made the leap from new software to a procedural backbone yet.

“AI is not a mature tool that you can unpack, plug in, and start redefining processes,” IE University’s Dans said. “This is probably going to happen soon, but we’re not there yet.”