Stanford University’s Digital Economy Lab director, Erik Brynjolfsson, wrote in the Financial Times that the US may be seeing some productivity gains from AI.
Brynjolfsson writes from the world of productivity spreadsheets, where economists have been stuck with a modern Solow paradox for more than a decade.
Artificial intelligence is everywhere except in the productivity figures, with sceptics insisting today’s machine learning and generative AI are not in the same league as older breakthroughs.
Brynjolfsson argues the latest benchmark revisions from the US Bureau of Labor Statistics hint that the statistical haze may be clearing at last.
New data this week cuts against the story that AI has not moved the US economy at all, starting with a hefty correction to the jobs picture.
Earlier reports pointed to a year of steady labour expansion, but total payroll growth was revised down by about 403,000 jobs.
Brynjolfsson points out that this would normally indicate weaker output, yet real GDP held steady, with a 3.7 per cent growth rate in the fourth quarter. He believes that is the signature of productivity growth, with high output maintained while labour input declines.
His updated analysis puts US productivity growth at roughly 2.7 per cent for 2025, almost double the 1.4 per cent annual average of the last decade.
The idea is that general-purpose technologies, from steam engines to computers, require time and substantial investment before measurable gains appear. That investment is often intangible and poorly captured, with firms reorganising processes, retraining workers, and building new business models, while official data lags behind.
Brynjolfsson suggests the revised 2025 data could signal a shift from an investment phase to a harvest phase, in which earlier effort begins to yield measurable output.
He also points to micro-level evidence, drawing on work with Bharat Chandar and Ruyu Chen that examined the employment effects of AI.
They found entry-level hiring cooled in AI-exposed sectors, with recruitment for junior roles falling by roughly 16 per cent, even as workers using AI to augment skills saw employment grow.
This implies that companies are beginning to delegate some codified, junior tasks to software and seem perfectly happy about it.
Brynjolfsson warns that productivity metrics are volatile, and it will take more periods of sustained growth to confirm a durable trend. He also flags the usual macro troublemakers, from geopolitical trade wars to fiscal or monetary mismanagement, which could swamp efficiency gains even if AI is doing real work.
Brynjolfsson draws a line between potential and realised gains, noting many businesses still use generative AI for only a thin slice of tasks.
Some firms keep it stuck on translation and summarisation, which he calls “glorified dictionary” use, and that sort of timid deployment is not going to remake an economy.
By contrast, he says a smaller set of power users are leaning on interactive conversations with AI agents to automate end-to-end workstreams.
Brynjolfsson cites examples like producing complete marketing plans and compressing weeks of effort into hours, which sounds less like dabbling and more like a rewire.
