AI Slowdown: Why Businesses Need a Pause to Harness Existing Technology

The artificial intelligence industry has long been in a relentless pursuit of developing more advanced models, pushing the accelerator in a competitive race for technological dominance. Now, a notion once considered almost unthinkable is gaining traction: slowing down the pace of development. The progress of frontier AI labs is closely tied to the performance of the US stock market and underpins substantial infrastructure investments. However, for businesses still grappling with how to effectively leverage AI's elusive productivity gains, a slowdown could be beneficial. It would provide them with the necessary time to bridge the significant gap between AI's rapidly advancing capabilities and their practical application.

Calls for a more measured approach to frontier AI development are growing louder. Anthropic's chief executive, Dario Amodei, has advocated for this, expressing concerns that capabilities are advancing faster than our ability to comprehend, align, and control them. His proposal suggests deliberately decelerating improvement to allow safety research and independent evaluation to catch up.

The practicality, desirability, and feasibility of such a slowdown remain subjects of considerable debate. However, for many businesses, this discussion is secondary, as they are far from fully utilizing the AI capabilities they already possess. According to the Stanford 2026 AI Index, a significant 88 percent of surveyed organizations reported using AI in at least one business function last year. Yet, the breadth of adoption does not equate to depth of integration. The research indicates that the use of AI agents remains in the single digits across nearly all business functions, with relatively few companies having fundamentally redesigned their workflows around AI.

Research from the Federal Reserve Bank of St. Louis found that generative AI accounted for only 5.7 percent of total working hours by August of last year. This data suggests that companies already possess powerful AI tools but are failing to maximize their potential. They do not require another substantial leap in model capability to realize the productivity benefits that are currently within reach. A randomized study by METR even found that experienced open-source developers working on familiar repositories took 19 percent longer when allowed to use early-2025 AI tools, highlighting that the bottleneck is increasingly organizational.

Companies face the challenge of connecting AI models to proprietary data, integrating AI into existing systems, and redesigning workflows rather than merely inserting chatbots. They also need to establish appropriate governance, security, and human oversight, while employees must learn to delegate tasks effectively to AI. Managers, in turn, need to re-evaluate roles, performance metrics, and decision-making authority. These organizational changes occur at a significantly slower pace compared to the rapid advancements in AI labs.

OpenAI's enterprise research indicates that the most intensive AI adopters utilize substantially more capabilities across a wider range of tasks than the average employee, reporting greater productivity benefits. Seventy-five percent of surveyed workers also reported being able to perform tasks using AI that were previously beyond their capabilities. This underscores that the AI productivity frontier within most companies lags far behind the technological frontier.

This pattern is not unusual. General-purpose technologies historically precede organizations' ability to fully exploit them. Electricity predated factories being redesigned around distributed electric motors, and computers arrived long before companies reorganized processes around digital information. AI is likely to follow a similar trajectory, further strengthening the argument for a development pause. Such a pause would not necessarily stall the AI economy, which is vital for public equity markets, but would instead grant companies much-needed time to extract value from the remarkable capabilities already available. This involves moving beyond experimentation to workflow redesign, shifting from individual productivity tools to end-to-end processes, connecting AI to enterprise data and systems, and strategically deploying agents where they offer genuine value.

Companies must also build governance frameworks that permit experimentation without introducing unacceptable risks and, crucially, develop the organizational capacity to effectively utilize AI. While executives should continue to monitor advancements at the frontier, as better models will undoubtedly create new opportunities and a substantial slowdown is not guaranteed, they should not conflate a slower technological frontier with diminished business prospects. Even if frontier development were to halt dramatically, businesses would still have a vast array of economically valuable opportunities to pursue across customer service, software development, finance, procurement, marketing, and numerous other functions. Therefore, a slowdown in AI's cutting edge does not imply a halt in business opportunity. The AI race, particularly for enterprise users, needs to transition from solely building superior AI to constructing more robust businesses around it.