“Let’s be fair, Fortune 500 companies are pretty late to the party on this,” said Webb, founder and CEO of the consulting firm Future Today Strategy Group. “Artificial intelligence didn’t just show up a couple of years ago.”

She said she sees “enormous amounts of capital” going into pilots “with no strategy ahead of them.” Companies, she said, then argue with their security teams and end up with orphaned projects that never do anything. She also said some executives hand the building of what becomes proprietary technology to third parties, and then get stuck. Employees get frustrated and start building their own tools, she said.

Webb argued that having AI everywhere doesn’t matter if workers don’t know how to use it, comparing it driving in a New York City cab and being handed a phone and asked to type in an address—a constant pattern for about two years now, she estimated. Drivers as so reliant on navigation tools, she said, that there’s “a certain amount of learned helplessness,” just like some leaders during this moment.

Her advice for companies wasn’t to aim to be “AI native” but more about “being flexible,” but she said companies must build mechanisms for that. She said she hasn’t yet seen large companies do so, even though many CEOs are working to understand AI.

Why gains don’t add up

Webb said part of the reason individual productivity gains don’t add up across a company is how firms define return on investment. Most apply AI to the bottom line, she said. She contrasted that with Runway, an AI video company whose strategy she considers to be a clear vision for the future that its tools help advance.

She told a story about a friend at a giant company she wouldn’t name. When the chief technology officer wouldn’t approve a secure sandbox, the friend went directly to the CEO, built his own instance, got access to a supercomputer and assembled a team. Webb said the group now plans to spend a couple hundred million dollars on AI tokens. She blamed the lack of “strong leadership and planning,” and said many companies lead with “fear and FOMO.”

Runway’s counterpoint

Michelle Kwon, Runway’s chief operating officer, described a company built the opposite way. Runway was founded about nine years ago by three NYU graduates. She said it uses AI in “essentially every part of what we do,” and its whole staff writes code.

Employees have built about 215 apps for an internal app store since earlier this year, including teams outside engineering, and their newest product came from that approach, she said. (The company announced a pilot for an autonomous ad agent the day before the panel.) One worker built it in “a handful of weeks,” she said, adding that it raised ad output by more than 1,000% from a very small base, and is now being released publicly.

Runway still buys what isn’t central to its business, she said. It won’t build its own payment system or HR compliance software.

On productivity, Kwon said: “Just because you are using AI, that is not a proxy for your productivity or you doing a good job at work.” It should be a core part of how people work, she added — as long as it is “responsible” and “doesn’t create work for other people.”

“It’s not helpful if I receive a 100-slide deck,” Kwon said. “What is someone asking me to do with that?” The same goes for chatbot output, she said, which Runway sometimes gets: people “just copy and paste the output without having thought about: how do I synthesize this?” It just “isn’t a productive use of anyone’s time.”

Peloton’s chief technology officer, Francis Shanahan, offered a comment from the crowd about the opposite problem. Employees are “inundated now because they can literally build anything,” he said, and he is finding “burnout to be more and more of an issue.” Kwon responded that Runway makes a point to “celebrate the wins” and “shout people out.” It gives the whole company the same day off, not tied to a holiday, because with summer Fridays, “who’s taking a half day? Who’s taking a full day? What’s hard to track?” When Khan joked that AI agents could keep working on those days, Kwon replied, “They can work tirelessly.”

Risk and liability

Matt Maher, founder of M7 Innovations and a consortium partner with MIT Media Lab, asked who is liable when an AI agent breaches a contract or causes other harm. Webb relayed a story about a coding project she once built that spammed 1,000 people after she put it in a live environment with a loop in it. Such failures, she said, are ordinary human errors and not AI “waking up.” Executives, she said, must “stop anthropomorphizing” AI.

Webb added that security and risk officers should say “tell me more” before they say no, and that leaders should give them some leeway. She also warned that a wave of connected devices, such as glasses and charms that listen, will arrive over the next 18 months with little planning.

Asked about boards using AI, Webb said her firm has used it for a decade, but it is useless if directors don’t understand what the data show. She compared it to her own cycling sensors. “You can’t disassociate from what’s happening,” she said, “or you are liable.”

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.