Podcast Episode
The parallel is particularly striking given the historical trajectory of productivity growth. US Bureau of Labor Statistics data shows productivity grew 2.7 percent annually from 1947 to 1973, before personal computers became widespread. That figure dropped to 2.1 percent between 1990 and 2001 after PCs went mainstream, and declined further to just 1.5 percent from 2007 to 2019. While recent quarters have shown improvement, with non-farm business sector labour productivity rising 4.9 percent in the third quarter of 2025, Gownder argues this falls far short of the transformative gains AI proponents have promised.
The finding is supported by survey data showing that only 15 percent of AI decision makers reported an EBITDA lift for their organisation in the past 12 months. Fewer than one third can tie the value of AI to profit and loss changes. The pressure to demonstrate returns is mounting, with 61 percent of senior business leaders reporting they feel more pressure to prove ROI on AI investments compared to a year ago.
The study recruited 16 experienced developers from large open source repositories averaging over 22,000 stars and more than 1 million lines of code. These developers were asked to work on 246 real issues from their repositories, with tasks randomly assigned to either allow or disallow the use of AI tools. When permitted, developers primarily used Cursor Pro with Claude 3.5 and 3.7 Sonnet, frontier models at the time of the study.
The results were striking. Developers using AI tools completed tasks 19 percent slower than those working without assistance. Even more revealing was the perception gap, developers expected AI to speed them up by 24 percent, and even after experiencing the slowdown, they still believed AI had made them 20 percent faster.
The study identified several contributing factors to this paradox. Developers accepted less than 44 percent of AI generated code suggestions. Seventy five percent reported they read every line of AI output, and 56 percent made major modifications to clean up AI generated code. Working on large, mature codebases with intricate dependencies and coding standards proved particularly challenging for AI tools lacking deep contextual understanding.
The workslop problem illustrates how AI can create work rather than eliminate it, as employees must invest significant time reviewing, correcting, and salvaging AI generated materials that initially appear acceptable but ultimately require substantial human intervention.
The firm estimates that generative AI will now account for 50 percent of automation driven job losses, up from 29 percent in its 2023 forecast. However, Gownder cautioned that many recent layoffs attributed to AI are actually financial decisions masquerading as AI job losses, with companies cutting staff in anticipation of automation that doesn't yet exist.
Forrester's 2026 predictions suggest over half of layoffs attributed to AI will be quietly reversed as companies confront the operational realities of replacing human workers prematurely. This creates significant disruption for workers and organisations alike, all in pursuit of productivity gains that remain elusive.
Like other general purpose technologies, the full effects of AI won't be realised immediately. Some researchers suggest we may be in the dip of the J curve, diverting effort toward business adaptation while unaccounted for productivity benefits remain hidden in the data.
The question facing the industry is whether the current productivity paradox represents a temporary implementation lag before transformative gains materialise, or whether the technology simply won't deliver on its most ambitious promises. With billions already invested and millions of jobs potentially at stake, the answer has profound implications for workers, businesses, and economies worldwide.
The AI Productivity Paradox: Two Years of Hype Without the Results
January 19, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Two years into the generative AI boom, a striking disconnect has emerged between the technology's promises and its measurable impact on economic productivity. Despite billions invested in AI infrastructure and widespread adoption across industries, the anticipated productivity gains remain conspicuously absent from official statistics, according to a Forrester analyst who says the data simply doesn't support the hype.
The Modern Solow Paradox
JP Gownder, vice president and principal analyst at Forrester, told The Register this week that productivity gains from artificial intelligence are not appearing in the data. The observation echoes what economists call the Solow Paradox, named after Nobel Prize winning economist Robert Solow who famously remarked in 1987 that the effects of the PC revolution can be seen everywhere except in the productivity statistics.The parallel is particularly striking given the historical trajectory of productivity growth. US Bureau of Labor Statistics data shows productivity grew 2.7 percent annually from 1947 to 1973, before personal computers became widespread. That figure dropped to 2.1 percent between 1990 and 2001 after PCs went mainstream, and declined further to just 1.5 percent from 2007 to 2019. While recent quarters have shown improvement, with non-farm business sector labour productivity rising 4.9 percent in the third quarter of 2025, Gownder argues this falls far short of the transformative gains AI proponents have promised.
Enterprise Projects Failing to Deliver ROI
The gap between expectation and reality appears even more pronounced at the enterprise level. An MIT Media Lab study found that 95 percent of enterprise generative AI projects fail to deliver measurable profit and loss impact. A lot of generative AI stuff isn't really working, Gownder stated bluntly. So no actual ROI.The finding is supported by survey data showing that only 15 percent of AI decision makers reported an EBITDA lift for their organisation in the past 12 months. Fewer than one third can tie the value of AI to profit and loss changes. The pressure to demonstrate returns is mounting, with 61 percent of senior business leaders reporting they feel more pressure to prove ROI on AI investments compared to a year ago.
AI Coding Tools: The Perception Gap
Research on AI coding tools, widely promoted as one of the technology's most promising applications, has revealed a significant disconnect between perception and reality. A randomised controlled trial conducted by the nonprofit research organisation METR examined how AI tools at the February to June 2025 frontier affected the productivity of experienced open source developers.The study recruited 16 experienced developers from large open source repositories averaging over 22,000 stars and more than 1 million lines of code. These developers were asked to work on 246 real issues from their repositories, with tasks randomly assigned to either allow or disallow the use of AI tools. When permitted, developers primarily used Cursor Pro with Claude 3.5 and 3.7 Sonnet, frontier models at the time of the study.
The results were striking. Developers using AI tools completed tasks 19 percent slower than those working without assistance. Even more revealing was the perception gap, developers expected AI to speed them up by 24 percent, and even after experiencing the slowdown, they still believed AI had made them 20 percent faster.
The study identified several contributing factors to this paradox. Developers accepted less than 44 percent of AI generated code suggestions. Seventy five percent reported they read every line of AI output, and 56 percent made major modifications to clean up AI generated code. Working on large, mature codebases with intricate dependencies and coding standards proved particularly challenging for AI tools lacking deep contextual understanding.
The Workslop Problem
A new phenomenon has emerged that actively undermines productivity gains. Dubbed workslop, this refers to AI generated content that appears polished on the surface but lacks substance. Research from BetterUp and Stanford found that workers spend nearly two hours addressing each workslop incident. Seventy four percent of respondents reported negative consequences from low quality AI outputs.The workslop problem illustrates how AI can create work rather than eliminate it, as employees must invest significant time reviewing, correcting, and salvaging AI generated materials that initially appear acceptable but ultimately require substantial human intervention.
Jobs at Risk Despite Underwhelming Performance
Perhaps the most troubling aspect of the current situation is that despite disappointing productivity data, job losses are still forecast to be substantial. Forrester's latest forecast predicts AI and automation will eliminate 6.1 percent of US jobs by 2030, approximately 10.4 million positions. These jobs are lost structurally, like they're gone for good, because they've been replaced, Gownder explained. That's not an insignificant hit to the economy.The firm estimates that generative AI will now account for 50 percent of automation driven job losses, up from 29 percent in its 2023 forecast. However, Gownder cautioned that many recent layoffs attributed to AI are actually financial decisions masquerading as AI job losses, with companies cutting staff in anticipation of automation that doesn't yet exist.
Forrester's 2026 predictions suggest over half of layoffs attributed to AI will be quietly reversed as companies confront the operational realities of replacing human workers prematurely. This creates significant disruption for workers and organisations alike, all in pursuit of productivity gains that remain elusive.
The Implementation Lag Question
Experts suggest the current situation mirrors historical patterns with transformative technologies. Unlocking productivity value from generative AI requires massive complementary investments, including reorganising workflows, retraining staff, and redesigning software systems. Implementation lags have likely been the biggest contributor to the paradox, as AI capabilities have not yet diffused widely.Like other general purpose technologies, the full effects of AI won't be realised immediately. Some researchers suggest we may be in the dip of the J curve, diverting effort toward business adaptation while unaccounted for productivity benefits remain hidden in the data.
Market Correction Ahead
Forrester predicts a significant shift in 2026 as the gap between hype and reality becomes impossible to ignore. The firm forecasts that enterprises will delay 25 percent of AI spending into 2027, and that enterprise ROI concerns will exceed the tensile strength of vendor hyperbole. AI will inevitably lose its sheen, trading its tiara for a hard hat, as the industry moves from hype to hard business outcomes.The question facing the industry is whether the current productivity paradox represents a temporary implementation lag before transformative gains materialise, or whether the technology simply won't deliver on its most ambitious promises. With billions already invested and millions of jobs potentially at stake, the answer has profound implications for workers, businesses, and economies worldwide.
Published January 19, 2026 at 3:09am
