Cuban Says AI Has Not Taken Office Jobs Because It Is Too Hard to Install
If the technology were truly ready to replace workers, companies would not need thousands of outside engineers to install it, he says.
All-In with Chamath, Jason, Sacks & Friedberg · Mark Cuban on the AI Bubble: Who Actually Gets Wiped Out?Two years after predictions that artificial intelligence would wipe out half of white-collar work, employment is still growing and companies are still hiring. The reason, says longtime tech investor Mark Cuban, is that using AI for yourself is easy and installing it for a whole company is hard.
On your own laptop, the demo works. You type a prompt, the chatbot writes the memo, codes the project, or powers through the homework, and it feels like productivity jumps a hundredfold. Inside a business, that same trick falls apart. The AI has to reach into private files, follow permissions and approvals, do the same job reliably every day, and fit work that was never organized for a machine.
Getting AI to work inside a company is messy. Chief executives "have no clue what is going on," Cuban says. The harness is the extra plumbing and guardrails that connect the chatbot to company systems and keep it on track. Cuban says that explaining the need for a harness to them has no chance of working.
His evidence is who does the installing. Cuban points to Microsoft hiring about 6,000 people, and to Anthropic and OpenAI saying they will deploy their people into these companies. These outside engineers, what he calls forward-deployed engineers, are AI maker staff who work on site at the customer to custom fit the software.
If the technology could truly think and act on its own, Cuban argues, you would simply ask it what to do and how to set it up. The fact that you need human helpers to make it useful at work tells you AI is hard.
Questions people ask
What does a forward-deployed engineer actually do?
They are software engineers employed by the AI maker who work on site at the customer for weeks or months. They connect the AI to internal data, build custom tools around it, and fix failures, which is essentially high-end tech consulting billed as part of an AI deal.
Why is business use so much harder than using a chatbot myself?
A prompt works when one person asks for one answer. A company needs the same answer to be correct every time, for many users, with access to private records and limits on who can see what. Building those connections, checks, and permissions is the hard part.