Lightspeed Partner: Blindly Accumulating Tokens Is Like Spamming APM in Games, AI's True Bottleneck Lies in Verification Capacity

Deep Tide TechFlow reports that on July 30, according to Fortune, Lightspeed Venture Partners partner Nnamdi Iregbulem wrote an article pointing out that the AI field is repeating the mistakes of the gaming industry—using 'Tokenmaxxing' (i.e., the belief that consuming more tokens necessarily leads to better performance) as the core metric for measuring AI efficiency. This is akin to early gamers treating actions per minute (APM) as a proxy for technical skill, which is essentially a vanity metric. Research data shows that the token consumption of AI agents is four times that of regular conversations, while multi-agent systems consume up to 15 times more. Furthermore, a 2025 analysis of AI-assisted software development found that engineers spend more than twice as much time reviewing and modifying AI-generated code (9% of programming time) compared to waiting for AI to generate code (4%). The author cites economic research papers from MIT, the University of Washington, and UCLA, stating that 'the bottleneck to growth is no longer intelligence itself but human verification bandwidth.' The article suggests that companies should prioritize building scalable verification mechanisms when deploying AI, rather than blindly pursuing token consumption. It also calls on the industry to establish metrics similar to gaming's 'effective APM' to differentiate genuinely valuable agent outputs from inefficient computational consumption.