The backwards AI pacing debate and how far business is from the frontier

The debate surrounding artificial intelligence pacing has taken center stage in Washington and Silicon Valley. Pacing, or the deliberate slowing down of frontier model development, is seen by some as unilateral disarmament in the race with China, while others view it as the responsible path for a technology that creators warn of catastrophic risks.

However, both camps have fallen victim to the "Compute-to-GDP Fallacy," which erroneously assumes that every incremental leap in AI model performance directly translates to macroeconomic output. The truth is that AI is following the same general pattern as other general-purpose technologies, which took decades to diffuse into measurable productivity.

Despite this, corporate America appears to be years behind the AI frontier, and the commercial fortunes of labs will be decided by trust and adoption, not raw capability. Pacing may cost the economy little, but racing ahead without proper alignment could cost far more. The argument that AI is being unfairly dismissed or manipulated for financial gain is not without merit.

While AI has reached a critical capability milestone, warnings of catastrophic or existential risk should not be ignored. However, focusing almost exclusively on cutting-edge models has led to mismanagement of messaging and public trust. The real hurdle lies in the data infrastructure of enterprises, with data readiness being the primary barrier to implementing AI.

Only a small percentage of companies have fully prepared data systems, and many are still uncertain about their data management capabilities. Despite the rapid development of AI models, fewer than 10% of companies report significant impacts from AI adoption, and most are still operating within the first phase of AI adoption, which involves assistance through off-the-shelf copilots.

The second phase, orchestration, covers agentic workflows that demand real investment and a human in the loop for approval. The third phase, autonomy, brings end-to-end agentic operations across interconnected systems. While the potential rewards of each phase increase, so do the risks and trust required. This is why the average Fortune 500 CEO remains in the first phase, making sizable but cautious investments in AI.