Aaron Levie, Box co-founder and CEO, says students and colleges should continue teaching domain fundamentals so experts can direct AI agents, evaluate outputs, and correct errors

guess what they said about books, computers and the internet

it's the same silly argument

@levie AIs are like genes: their “expression” depends to a large extent on the environment (the prompt and context). Change the surroundings, and they produce wildly different outputs. Domain expertise matters.

@levie vibe coding gets you demos

deep understanding gets you products that survive scale

and TechBrunch gets you existential crises at 2am about both

@levie the person who doesn't know the formula can't tell if the ai made it up or made it mean something else

@akankshaluvcats @levie the gap shows up in incidents, not features. demo code passes review at 5 users and silently fails at 500 when a race condition you never thought about lights up. deep understanding is what lets you predict where the first 50 users will break the thing.

People should use AI to go deep in subjects. I believe AI is a layer on top of human knowledge that enables you to learn from the “collective”. It’s a dynamic interactive knowledge base that you should use to explore and learn. Also asking it things that are not normal for people like “what didn’t I ask that I should have asked about this topic?” Or “what are the foundational elements I should learn and how should I dive deeper into it”

@levie This is the correct take.

It's still so, so important to have a strong fundamental understanding of whatever domain you are pursuing.

AI is just the worlds greatest force multiplier on top of that pre-existing knowledge.

@levie Totally.

AI will amplify our intelligence far more than it will substitute it.

To boil it down as best I can, it's similar to like, "why bother learning to play chess, when you'll almost certainly never become a Grandmaster?"

Or why do something you'll never become the best at. Or even useful at enough to make money doing.

Playing piano or any instrument comes to mind.

ENJOY the process of learning, growing, doing something for yourself, maybe working with others, creating things. APPRECIATE the technology in its full capability.

We CANNOT have people measure their value or worth against what AI can do.

That will destroy morale and maybe society. (Except the SSRI-medical-industrial complex..)

@levie Technology is always an amplifier of human talent. Deep knowledge and experience are not replaceable just by leveraging of much more sophisticated tools. In order to solve the problem, one needs to understand the roots of the problem and why previous attempts failed.

I have to reiterate this one. We HAVE to do this right now.

"We CANNOT have people measure their value or worth against what AI can do."

Teach kids not to measure their value by how good they are at tasks. (That one caused me a lot of damage over the years and is taking time to untangle, personally.)

The evaluation problem is the one nobody talks about enough. Using AI to generate output is easy. Knowing whether the output is any good requires the domain knowledge you were tempted to skip. The novice can't tell when the agent is confidently wrong. The expert can. And in high-stakes domains that gap isn't a productivity difference. It's the difference between catching the error and shipping it. Going deep in your craft isn't just about being better at using AI. It's about being the person who can tell when AI has failed.

@levie 🎯

@scaling01 Also, he asked the question, so presumably he is willing to learn if it's relevant. I don't see the issue.

@levie AI will empower everyone, but turbocharge the experts.

@levie Same pattern in coaching. Athletes who learn bar path before they ever load it move faster three months in. Skip the fundamentals and you cannot tell when the output is wrong. AI tools amplify that gap fast.

@levie The problem is my kids are being told to go deep in something before anyone knows which somethings will still exist when they graduate.

That's the harder problem.

@levie AI does not replace fundamentals. It removes the delay before weak fundamentals hit production.

The operator with taste still has to know what good looks like, where it breaks, and when the answer is just cosplay.

@levie One-liner gut punch: the expert who steers agents will always smoke the novice who just vibes with them how many “AI-first” startups are about to learn that the hard way when their orchestrator can’t fix its own hallucinations?