Companies already have AI, but their workers do not know what to do with it
Uber put about 30,000 employees on AI at once and burned through its tokens in one quarter, but Andy Berman of RunLayer argues, as relayed by Lon Harris, that companies have yet to connect AI to the data that would make it valuable.
This Week in Startups · Is AI the new cigarettes? Our VC panel digs into Anthropic’s S-1 | E2356The bottleneck holding back artificial intelligence is not computer chips but workers. That was the argument from Jason Calacanis, responding to D.A. Wallach, who had argued that the problem is price, not supply, because customers spend only about 0.1% on AI.
"I think it's a meat sack constrained, DA," Calacanis said. "The meat sacks at organizations are not able to deploy it." By meat sacks, he means human employees.
What Calacanis means is that the technology is ready, but the people inside large organizations cannot deploy it effectively. In his view, growth depends on which companies put all of their workers on the tools, and whether those workers learn to do more than shuttle inputs and outputs back and forth.
His example was Uber, the ride-hailing company. Calacanis said Uber moved quickly and put about 30,000 employees on the technology at once, then burned through all of its allotted AI work, measured in tokens, in the first quarter. The question, he said, is when workers will learn to direct the tools to produce business results, instead of just carrying material between the AI and the job.
Lon Harris relayed a similar view from Andy Berman of RunLayer. In Berman's view, as relayed by Harris, companies have yet to attach AI to the data and systems that would make it more valuable, and the reason is human hesitation, lack of understanding, and security concerns. Calacanis said he thinks it is both, putting it this way: "The meat sack doesn't know what to do."
Dave McClure thinks the shift is early and will still happen. He likened companies talking about AI to "teenagers in high school talking about sex." His point was that many firms claim to use the technology, few use its full potential, and most will catch up over time. Most people, he said, still use it for simple prompts, while only a few use agents, software that can carry out several steps on its own.
Calacanis described a worker who summarizes an article, then asks what to do next, not realizing that summarizing was the demo four years ago.
Questions people ask
What does it mean to attach AI to a company's data and systems?
It means letting the AI read internal files, customer records, and software tools, with permission, so it can act on real work instead of just answering general questions. A connected assistant can draft a refund email from the order history, file the ticket, and update the inventory. Without that connection, workers are stuck copying and pasting between the AI and their job.
Why are big companies worried about letting AI see their data?
They worry that sensitive emails, customer information, or trade secrets could leak, be used for training, or be exposed by a mistake or hack. Regulated industries like health care and finance face fines if private data escapes. So many firms start with pilots and strict controls until they trust the security and legal setup.
What is the difference between a prompt and an AI agent?
A prompt is a single request, like asking for a summary or a draft, where you get one answer back. An agent is software that can pursue a goal through several steps, such as checking a calendar, booking travel, and notifying colleagues, with some supervision. Agents require more access to data and clearer rules, which is one reason adoption is slower.