Why AI Is About More Than Productivity Gains

OpenAI recently highlighted examples of generative AI delivering productivity improvements: an office worker in Pennsylvania saving 95 minutes daily, a California teacher reclaiming six hours per week.

However, despite this, in July 2025, OpenAI announced a major economic study in collaboration with leading economists Ronnie Chatterji (Duke), Jason Furman (Harvard), and Michael Strain (American Enterprise Institute). This initiative aims to rigorously assess generative AI's true impact on productivity, employment, and the broader economy.

Indeed, the Economist noted earlier this year that OpenAI's claims have been based on carefully chosen anecdotes rather than comprehensive economic data. The publication suggested the company's internal research risks credibility by highlighting isolated successes instead of providing broader economic analysis.

OpenAI's decision to commission independent economic research into generative AI's impact is striking, given the company's previous bullish stance on its technology. Rather than reinforcing its optimistic narratives around productivity, this move appears to openly acknowledge growing concerns that AI's economic effects remain poorly understood, and that reliable answers will require deeper, more critical analysis.

As regular readers will note, I prefer to move past comfortable narratives.

I think OpenAI's decision to commission this economic study sends a clear message to executives: productivity gains do not justify an investment of this scale.

A platform of this magnitude must deliver more than just time savings. Which suggests a fundamental question:

If AI tools are as revolutionary as promised, why would we ever reference their success with productivity metrics?

When e-commerce transformed retail, it didn't just make shopping a little quicker; it reshaped consumer behaviour completely. If generative AI truly holds comparable potential, then OpenAI's emphasis on minor efficiency improvements seems distinctly out of step.

Amazon didn't streamline checkout processes; instead, it rendered entire categories of physical stores irrelevant. Traditional retail was effectively finished the moment digital commerce reached critical mass. Is conventional knowledge work now facing a similar fate? If generative AI genuinely holds the transformative power OpenAI claims, the logical implication is stark: the conventional white-collar job as we know it is already obsolete.

By emphasising isolated instances of time-saving, OpenAI risks narrating generative AI's potential to something akin to digital brochureware, a superficial improvement rather than a structural shift. MIT economist Daron Acemoglu highlights this disconnect clearly, describing these incremental productivity gains as "disappointing relative to the promises that people in the industry and tech journalism are making." He pushes further, questioning whether current deployments genuinely empower workers or merely automate roles, a route he bluntly labels "the wrong direction."

By chasing incremental efficiencies, leaders risk being blindsided by competitors who recognise that this technology isn't about marginal gains, it's about fundamentally redefining what their business could become.

Why are we even talking about productivity? Was it ever the real problem?

OpenAI's Chief Economist Ronnie Chatterji projects bold productivity boosts of "around 20 percent in tasks" (TechSurge podcast, June 2025), a staggering claim given the billions companies have already invested chasing incremental improvements before generative AI. Prior to Gen-AI, returns on such investments have remained stubbornly modest. Chatterji might view this as a breakthrough moment, but MIT economist Daron Acemoglu sharply contests his optimism, predicting a far more modest 1.1–1.6 percent GDP increase from AI over the next decade. Acemoglu doesn't stop there: he explicitly criticises the current AI deployment as overly automation-driven rather than genuinely empowering, labelling this "the wrong direction" (MIT Economics, December 2024).

Acemoglu's analysis reinforces the idea that focusing the discussion on productivity misses the broader issue entirely. Productivity itself may never have been the critical constraint limiting business performance.

If productivity was genuinely the critical bottleneck in organisational performance, the cumulative effect of decades spent implementing ERP systems, lean management, and robotic process automation should have substantially alleviated it. Yet, productivity gains from these earlier technologies have remained consistently incremental, rarely delivering the transformative breakthroughs originally promised. This pattern strongly suggests that productivity itself may not have been the underlying strategic constraint. Rather, productivity metrics might have become the default measure of success simply because they are conveniently quantifiable, rather than strategically insightful.

Economic historian Michael Roberts captures this reality well, characterising today's AI investment wave as a "dot.com bubble on steroids," highlighting a troubling mismatch between enormous investment ($332 billion) and minimal economic returns thus far ($28.7 billion) (Michael Roberts blog, July 2025).

Leaders might reconsider whether the narrative of "solving productivity" has emerged not from genuine strategic necessity, but rather as a convenient rationale for the widespread and perhaps premature adoption of generative AI.

Brochures Online or Shuttering Shops?

Stanford economist Erik Brynjolfsson also warns against superficial AI adoption, which he terms "paving the cow paths": automating existing workflows without meaningful transformation. He argues convincingly that genuine productivity gains can only emerge when organisations fundamentally rethink their structures, strategies, and processes (McKinsey, September 2024).

But there is a deeper strategic danger here. If executives assume productivity is the right measure of generative AI's potential, they risk becoming complacent about their existing business models. Unlike minor innovations, genuine technological revolutions create entirely new frameworks that replace, not merely refine, the old.

The risk for executives is not recognising that generative AI might redefine their markets entirely. And that AI could leave them with neatly optimised but fundamentally irrelevant business models.

What Should We Be Asking Instead?

If productivity wasn't our core problem before, why assume generative AI will meaningfully boost our competitiveness?

Are we fundamentally rethinking workflows, or digitising outdated processes?

The conversation continues. This week, I'll close by saying that OpenAI's decision to commission independent economic research seems to acknowledge uncertainty about AI's true economic impact. Heeding this warning, I would urge CEOs to look past comfortable productivity narratives promising efficiency gains. Daron Acemoglu's blunt assessment should ring loudly: modest gains are "disappointing relative to the promises."

If generative AI is truly revolutionary, businesses won't measure its impact in hours saved; they'll measure it by the business models and competitors it leaves behind.

References

OpenAI. (2025, July). OpenAI's new economic analysis. openai.com
Strain, M. R. (2025). Collaborating with OpenAI on AI and the economy. LinkedIn
Acemoglu, D. (2024, April). The Simple Macroeconomics of AI. MIT Economics Working Paper. economics.mit.edu
Acemoglu, D. (2024, October). What do we know about the economics of AI? MIT Economics News. economics.mit.edu
Brynjolfsson, E. & McKinsey & Company. (2024, September). The human side of generative AI: Creating a path to productivity. mckinsey.com
Roberts, M. (2025, July 27). AI: Bubbling Up. The Next Recession (blog). thenextrecession.wordpress.com
Brynjolfsson, E. (2023). GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. arXiv preprint. arxiv.org