This past week, I co-hosted a pre-game show with Sarah Guo before Jensen Huang’s GTC keynote in San Jose. Three panels, twelve guests, covering open models, agentic AI, and physical AI. Backstage, one founder told me his biggest competitor is no longer another AI lab. It is an open-source project built by a single Austrian developer that did not exist six months ago. Minutes later, Jensen took the stage and called that project, OpenClaw, “the single most important release of software, probably ever,” comparing it to Linux and HTML. He projected a trillion dollars in chip orders through 2027, described a $35 trillion agentic AI ecosystem, and unveiled Vera Rubin, a next-generation infrastructure platform integrating seven chips across five rack-scale computers operating as a unified supercomputer. The audience was mesmerized.

But what struck me most was not what Jensen announced. It was what I observed in the hallways afterward. Every conversation was about adoption. Who is using what? How fast? How many agents? How many tokens? Founders, investors, and executives compared notes like students comparing test scores, each measuring a slightly different version of the same thing. How much AI have you deployed?

We have seen this movie before. Companies chase false currencies, and vanity metrics masquerade as fundamentals. Market share is often mistaken for winning. Growth is often mistaken for durability. Shipping features are often mistaken for building products. Now, adoption is being mistaken for advantage.

The wrong scorecards are already emerging. Percentage of code written by AI. Number of agents deployed. Tokens consumed. Features shipped per sprint. These feel like progress. They are easy to measure. They signal activity. It is true that without them, you have no chance of becoming AI-native, but they are the modern equivalent of vanity metrics because they say almost nothing about whether you are actually building something that wins.

At a recent board meeting for a portfolio company, the CTO shared that the top 5 to 10 percent of builders are now 3 to 5 times more productive than a year ago. The median builder is up perhaps 10 to 20 percent. That gap is not closing. It is widening. And the difference between the two groups is not technical fluency. Both groups use the same tools. Both have access to the same models. The difference is what they choose to build and why.

For the past two decades, the binding constraints in software were: (1) hiring engineers was hard, (2) writing code was slow, and (3) shipping products took months. Capital was allocated around these bottlenecks. Roadmaps were built around them. Competitive advantage was often a function of who could attract and retain the best engineering talent and ship the fastest. AI is dissolving those constraints in real time. Code is generated. Prototypes are instant. Iteration is nearly free. When production constraints disappear, advantages shift. When you can build anything, the question becomes what to build. When there are too many ideas, features, directions, and noise, the bottleneck is no longer execution. It is judgment.

This is the part that makes people uncomfortable because it is harder to measure and impossible to automate. Judgment is the ability to distinguish signal from noise, to say no to good ideas in favor of great ones, to hold conviction when the data is ambiguous, and to change course when new information demands it. At LinkExchange, Tellme, and Zappos, we made countless mistakes, but the decisions that mattered most were never about speed. They were about making the right choice under uncertainty. What market to enter? What customers to serve? What to build next? What to stop doing? Those choices compounded over the years and defined the outcomes.

AI compresses the distance between idea and execution. It does not compress the distance between good judgment and bad judgment. It amplifies whatever you point it at. Strong teams with clear strategies will get faster and more focused on what matters. Weak teams with vague strategies will get noisier and more distracted. Clear thinking compounds. Confused thinking unravels. The gap does not close. It widens.

Daniel Kahneman spent decades studying how humans make decisions under uncertainty, and one of his most important findings is also his most humbling. We are reliably overconfident in the quality of our own judgment. We confuse fluency with accuracy. We mistake activity for progress. We feel productive when we are merely busy. In a world where AI makes everyone more productive, the illusion of progress becomes even more seductive. We ship more, build more, deploy more, and mistake the volume for value.

The antidote is not to slow down. It is to redirect what we measure. Stop measuring activity and start measuring learning. Optimize for the feedback loops that sharpen judgment over time. Invest in people who make better decisions consistently, not just faster ones. Build systems and cultures that improve decision quality alongside accelerating decision speed. Learn from what we shipped, not just whether we shipped it.

At GTC, Wall Street’s reaction to Jensen’s vision was mixed. Investors weighed the prospect of an AI bubble alongside the trillion-dollar demand forecasts. The software sector continues to be repriced as the narrative shifts from “AI helps software companies” to “AI threatens the per-seat business model.” The entire stack, from chips to models to orchestration to pricing to procurement, is being repriced in real time. Every layer is shifting. When the ground is moving, everyone feels the urgency to adapt and adopt AI.

Urgency without judgment is just panic with a better brand. Everyone will adopt AI. Everyone will build more, ship faster, and deploy agents. That is table stakes. The founders who win will be the ones who make better choices. The real question is not how much AI you are using. The question is whether AI is improving your judgment or accelerating your mistakes. Knowing the difference is the key to your AI advantage.