AI weakens the assumption behind many career ladders: that experience reliably produces more value. When volume becomes cheap, output per minute stops measuring contribution, and even correct output can solve the wrong problem. A better measure resembles value per token, rewarding problem framing, verification and owned outcomes. Because AI assistance can help less-experienced workers most, ladders should promote demonstrated outcomes, open individual-contributor routes without direct reports, judge managers on developing people and protect apprenticeship.
Nvidia CEO Jensen Huang talked with Lex Fridman about AI and jobs. Comparing two new graduates, he said: "I would hire the one who's expert in using AI." It was career advice, but it points at a bigger problem than hiring.
I think we need a complete rethink of our career frameworks. Many ladders assume experience reliably produces more value, so they reward years served, personal output and team size. AI weakens that assumption and forces a harder question: what is value, really?
Output per minute was always a proxy
For decades, speed and volume were fair stand-ins for skill, because experienced people produced more that worked.
AI makes volume cheap: anyone can produce pages of code or analysis in minutes. Keep rewarding output per minute and we reward material someone else has to read, check and fix, creating more noise with no value gained.
Correct output, or value per token?
Correct output gets closer, because a right answer saves everyone downstream. However, a correct answer to the wrong problem is still waste.
I think the better measure looks more like value per token: how much outcome someone creates per unit of machine effort and human attention. It rewards people who frame problems well, hand the right pieces to AI, verify what comes back and own the result. Prompt counts and tool familiarity reveal almost none of it.
What if the junior wins?
When economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed 5,172 customer-support agents given an AI assistant, the least experienced gained the most in issues resolved per hour, while the most experienced gained little. The assistant narrowed the gap that experience normally creates.
Picture a two-year engineer, fluent with agents, who delivers more value than a fifteen-year veteran. A tenure-based ladder either holds that junior back or pretends nothing has changed.
Experience still matters. Seniors carry judgment the tools lack: knowing which problem is worth solving, spotting the plausible wrong answer and owning the consequence. Seniority now has to earn its place through that judgment rather than years.
In *The Empire Strikes Back* (1980), Luke Skywalker reaches Dagobah with raw power and very little judgment. Yoda seems an unlikely master, and Luke is impatient with him, but Yoda holds what Luke lacks: knowing when not to act. Luke leaves training early to rush to Cloud City and pays with his hand. That is the career ladder I want. Let the newcomer's power show up early and count, and keep the veteran for the judgment that power cannot supply on its own.
The dotcom wave, at a far larger scale
During the dotcom boom, young engineers who understood the web were in hot demand. I think we may see that again, magnitudes bigger, because AI reaches into almost every knowledge job rather than one new industry.
So I would rebuild the ladder. Promote on demonstrated outcomes, including rework and failures. Let individual contributors climb by building workflows and agent systems that improve others' work, without direct reports. Judge managers on developing people and dividing work between humans and AI, not headcount. And protect apprenticeship: if AI takes the entry-level tasks, juniors need deliberate practice explaining, debugging and checking, or we lose the next generation of judgment.
The next career ladder should stop asking how long you've been climbing and start asking what you can be trusted to deliver.
