Box CEO Aaron Levie argues CEOs overestimate AI readiness because they only see prototypes and ignore implementation work

Levie’s Law of AI Psychosis:

The farther away you are from the actual work the more confident you are that humans are no longer needed

I like it

CEOs were routinely blown away by what "the kids" were doing with Word, Excel, PowerPoint in the 1990s. They had no idea where those fancy documents came from and how but they loved them.

Many never learned and fell behind.

I asked my eng team if I could ship code to prod

They told me no 💀

@levie Future generations will know it merely as “Levie’s First Law”

@buccocapital Was really hoping for a better law, damn

@buccocapital Classic “every other job but mine will be automated” syndrome.

@buccocapital This was actually already formalized as Garbled’s law

CEOs are the most delusional. Detached from reality.

@buccocapital

@buccocapital @kylascan Primest of examples lol

@buccocapital Corollary: the last mile is actually infinite…

Sold SaaS at the C-suite level for a decade Criteo, Deel, Autodesk. The demo always closed. The post-implementation audit is a completely different conversation. The happy path is a sales tool. The last mile is an engineering problem. That's why those two meetings have never once had the same people in them.

@levie This is why so many AI demos go viral but the actual products fail. CEOs see a perfect 2-minute demo and think the job is done. But the real struggle is the 90% grunt work like edge cases and security that nobody wants to talk about on stage.

This is the gap I see on the hiring side. CEOs are filtering for "AI-native" resumes and getting prototype builders. The founders winning right now are filtering for engineers who can list the 10 things that broke in their last AI rollout. Demo-fluency vs ops-fluency — the talent market hasn't caught up to the difference.

The delta between CEO demo mode and production AI is exactly where most enterprise initiatives stall. What we consistently see: the teams that close this gap fastest instrument the failure cases before they scale — map those 20 invisible steps before the first team runs the agent unsupervised. The psychosis breaks when the CEO has to own a failure mode, not just celebrate a prototype. How are you approaching that at Box — top-down mandated exploration, or letting operators define the agent scope first?

@levie the layer-of-abstraction tax. anyone two roles removed from the work has it. the operator hit the wall yesterday and knows where the cracks are. the CEO sees the demo. agents are great until someone has to clean up what the agent assumed.

@levie The last mile is where the real judgment lives: exceptions, context, permissions, review paths, ownership, risk. Model-down demos skip it because none of that exists in the prompt. Enterprise AI works when someone builds from the workflow upward. 😎

@buccocapital @davidzmorris Damn that’s so good

@levie