AI is the modern Solow paradox

In 1987, the economist Robert Solow made an observation that would define a decade of technology debate: "You can see the computer age everywhere except in the productivity statistics." Computers were spreading rapidly through offices and factories, yet measured output per worker was stubbornly flat. 

The paradox was eventually resolved not by questioning the technology, but by understanding the lag: productivity gains from general-purpose technologies only materialise after organisations invest in the complementary changes - new processes, new skills, new ways of working - that allow the technology to fully take hold.

That resolution gave rise to what researchers call the productivity J-curve. Before the gains arrive, performance can actually dip, because organisations absorb the costs and disruption of adoption before they capture the benefits. The curve goes down before it goes up. The firms that won from computing were not necessarily the first to buy the hardware — they were the ones that redesigned their operations around it.

We are living through an almost identical moment with AI.

Recent firm-level survey evidence shows that around 69% to 70% of firms are already using AI, yet more than 80% report no meaningful effect so far on employment or labour productivity (productivity here measured as sales per employee). That’s why the phrase “modern Solow paradox” fits: AI is suddenly everywhere, but organisations aren’t quite seeing it yet in hard performance numbers.

The recently revised NBER paper explains what is actually happening underneath those flat top-line metrics. In Denmark, across highly exposed occupations, employers have adopted AI chatbot initiatives widely, workers report time savings and quality gains, and new AI-related tasks are spreading. But two years after ChatGPT’s launch, the paper still finds minimal effects on earnings and recorded hours. In other words, the real news is not “AI does nothing.” It’s that AI mainly reshapes work first and only later, if at all, shows up in conventional productivity statistics. That is very close to the logic behind Solow’s original paradox and the later “productivity J-curve” argument: general-purpose technologies require complementary investment and organisational redesign before measured gains appear.

That leads to the uncomfortable conclusion many organisations would rather avoid: AI delivers very little if you simply lay it on top of existing work. The NBER evidence shows the strongest reported benefits where firms combine encouragement, tooling, and training; in those settings, 93% of workers had used chatbots, 28% used them daily, and 19% reported saving more than an hour a day. But those gains were mostly absorbed into different work: the majority of employees reallocated saved time to other tasks, and many took on new responsibilities in content generation, oversight, compliance, and AI integration. A recent Accenture-backed UK study points in the same direction: workers are becoming more efficient on individual tasks, but only a minority of organisations have redesigned core processes around AI, so local gains are not becoming enterprise gains.

That also explains the disconnect between future expectations and present reality. Executives still expect AI to raise productivity and reduce employment over the next three years. So belief in the destination remains strong even while evidence on the journey is thin. The market language around “proof-of-concept fatigue” is really a sign that buyers are moving from experimentation to operational accountability. Practically, it shows AI is now being forced to clear the same hurdle as any other enterprise investment: measurable ROI, not heroics and theatrics.

So who is succeeding?

The evidence suggests it is not the firms with the loudest AI narrative, but the firms with the best operating discipline. In the cross-country firm data, adoption is higher among younger and more productive firms, and larger or higher-paying firms expect bigger gains. Success, in other words, seems to come where AI is paired with managerial intent, process change, and role redesign. This is apparent in reported successes: Klarna reduced customer service time by over 80% using AI. AstraZeneca achieved 50% shorter development timelines and 70% time savings on documentation. Rolls-Royce increased machine utilization by 30% via AI planning.

This is exactly where Synyega can help. Synyega is an independent adviser that does not partner with software vendors, which matters because AI economics get distorted very quickly when tool choice, licensing advice, and consumption decisions are driven by seller incentives. Our ITAM services are built around reducing risk, cost, technical debt, and hybrid licensing complexity; our FinOps for AI offer is explicitly about visibility, ownership, forecasting discipline, and commercial control before AI spend becomes embedded; and our converged managed services focus on giving clients complete visibility and a governed, optimised software and cloud estate. That is the practical answer to the modern Solow paradox: not more tech or tools, but better governance, cleaner commercial foundations, stronger entitlement control, and process redesign that turns AI activity into measurable business value.

References:

Anders Humlum and Emilie Vestergaard, "Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI," NBER Working Paper 33777 (2025), https://doi.org/10.3386/w33777.