Stanford expert: The US is entering the "AI harvest period," and productivity growth is expected to double to 2.7% in 2025.

```

The Financial Times recently published a commentary article with a direct theme: The productivity “takeoff” brought by AI may finally be visible in macroeconomic statistics.

The author of the article, Erik Brynjolfsson, is the Director of the Stanford Digital Economy Lab and co-founder of Workhelix, a company focused on researching AI and organizational efficiency. He is at the forefront of academic research and has insights into the real implementation of AI in enterprises.

In this article, his core observation is: The U.S. may be moving from the “AI investment phase” to the “AI harvest phase”. He cites the latest economic data to point out that the scenario where “AI is discussed everywhere but cannot be seen in productivity data” is starting to change.

More specifically, based on updated data, he predicts that the U.S. productivity growth rate in 2025 will reach about 2.7%, almost double the annual average of 1.4% from the past decade. If this trend holds, it means that: AI is no longer just a story in PPTs, it is beginning to become a measurable efficiency gain in GDP.

The signal from macro data: Output remains but less labor is needed

Brynjolfsson first mentions a “counterintuitive” macro adjustment: The U.S. Bureau of Labor Statistics’ benchmark revision shows that growth in total payroll employment was revised downward by about 403,000 positions. At the same time, U.S. economic output has not weakened, real GDP remains strong, with a Q4 growth rate of 3.7%.

He calls this combination of “high output with less labor input” a typical feature of productivity growth and writes directly: “This decoupling — maintaining high output with significantly lower labour input — is the hallmark of productivity growth.” That is: doing the same or even more with fewer people, so productivity naturally rises.

However, the author also cautions against overexcitement, because productivity data tends to “fluctuate” and short-term readings are easily affected by statistical revisions and cyclical factors, “further periods for validation are needed.”

In this regard, MIT economist Daron Acemoglu has also publicly pointed out that the impact of AI on overall productivity depends on whether it can truly replace or augment labor in enough tasks, not just isolated improvements.

The “J Curve” explains why it’s only appearing now: first plant the tree, then get the fruit

Brynjolfsson puts AI’s diffusion path into a longer technological history framework. He mentions the economics profession’s long-standing debate over the modern “Solow Paradox”.

He summarizes this dilemma in a quote: “we have seen artificial intelligence everywhere except in the productivity statistics.” i.e.: AI is everywhere, but productivity in the statistics remains stagnant.

He explains this with the “Productivity J Curve”. Many general-purpose technologies, from steam engines to computers, do not immediately boost productivity upon adoption, but rather go through an “investment phase”.

Companies need to reorganize processes, train employees, and redo business models. Much of this investment is intangible capital, which in the short term may even depress “measurable productivity”. Only when organizational transformation is complete does the “harvest phase” begin and the efficiency shows up in the data.

Economic historian Paul David studied the “productivity lag” in the era of electrification, and found that after factories replaced steam power with electric motors, significant efficiency gains only appeared after reconfiguring plant layouts, workflows, and management. The “transform organization first, then see statistical returns” AI faces today is essentially the same logic.

The micro-level is changing: Entry-level recruiting down 16%, but “power users” are shrinking timelines

Besides macro data, the author also provides micro evidence. In a study with collaborators Bharat Chandar and Ruyu Chen, they found: In industries with “high AI exposure”, entry-level job recruiting has visibly cooled, with junior positions down by about 16%.

On the other hand, those who use AI to enhance skills are seeing employment growth. The author interprets this as: Enterprises are beginning to use AI for some “codable, standardized” junior tasks.

He also distinguishes “potential” and “realized gains”. Many companies currently only use generative AI in light scenarios such as translation and summarization, and he sharply describes this use as “glorified dictionary”.

However, among a small subset of “power users” observed by his company, AI agents are already automating end-to-end processes via interactive dialogue, such as generating complete marketing plans, compressing weeks of work into hours. He stresses that the real challenge for enterprises is not just acquiring the technology, but learning how to use it.

Externally, this phenomenon of “early adopters reaping benefits” is not surprising. Multiple McKinsey industry reports have emphasized that the value released by generative AI largely depends on process transformation and retraining, not just buying tools.

From experimentation to structural utility: Next, it’s about organizational capability and macro environment

At the end of the article, Brynjolfsson gives a strong trend assessment: “We are transitioning from an era of AI experimentation to one of structural utility.” This means AI is moving from the stage of “trying things out” to a stage of stably providing value. For enterprises, the next competitive focus will shift from “having a model” to “embedding the model into the business architecture.”

So what should companies actually do? Combining the author’s view, it can be summarized into three things: First, don’t just stay at the “glorified dictionary”, embed AI into end-to-end processes and let it participate in delivery, not just support. Second, upgrade training goals from “knowing how to use tools” to “knowing how to use AI to reinvent work methods”, using AI to boost average employee capability. Third, track returns with data and metrics to avoid initial excitement fizzling out and failing to scale.

At the same time, he also cautions that external risks may offset efficiency gains, including “geopolitical trade wars” as well as fiscal and monetary policy misjudgments. From a neutral perspective, technological progress and macro governance are two lines: the former provides possibility, the latter determines whether it can be realized smoothly.

Risk Warning and DisclaimerThe market has risks, investment needs to be cautious. This article does not constitute personal investment advice and has not considered individual users’ special investment goals, financial situation or needs. Readers should consider whether any opinions, views or conclusions in this article fit their specific situation. Invest at your own risk. ```