Experts say the high failure rate in AI adoption is not a bug, but a feature: “Has anyone ever started cycling on the first try?

The panel, titled “Succeeding: How AI is Transforming the Office,” tackled head-on a widely publicized MIT study suggesting that about 95 percent of enterprise AI pilots don’t bear fruit. This statistic has fueled doubts about AI’s ability to deliver on its promise, but all three panelists – Amy Coleman, executive vice president and chief human resources officer at Microsoft; Karin Klein, founding partner of Bloomberg Beta; and Jessica Wu, co-founder and CEO of Sola, forcefully pushed back against the narrative that failure signals fundamental problems with technology.

“We’re just at the beginning,” Klein said. “Of course, there will be a ton of experiments that don’t work. But, for example, has anyone ever started riding a bike on the first try? No. We get up, we dust ourselves off, we keep experimenting, and somehow we figure it out. And it’s the same with AI.”

Klein went further, encouraging the audience to become what she calls “vibe coders,” or people who use accessible AI tools to create applications without traditional programming experience. Coleman echoed Klein’s view, saying “it’s all about experimentation.”

“We’re on this jagged frontier, which is we’re going to have victories, and then we’re going to see this trough, and then we’re going to have more victories,” she added.

The Microsoft executive, who said her own CEO challenged the leadership team to rock the code, emphasized that creating a good organizational culture matters more than the technology itself. “I think the study is really important because it reflects what a lot of people are feeling right now, which is: Is this really something that’s going to help me at work? Will it bring me more joy and take away work?” » Coleman said.

Wu provided important context in an attempt to reframe MIT’s findings. “I think the study itself says that only 5% of the AI ​​tools that people are testing are going into production. What’s really interesting is if you step back and look at what percentage of studies on implemented IT tools actually went into production before AI, it wasn’t particularly high either,” she said, noting that success rates for technology deployments at large companies historically hovered around 10%. or less.

Wu’s company, Sola, builds what it describes as “agentic process automation” tools that help businesses automate manual back-office work. She pointed out that the sheer volume of AI experimentation underway makes lower success rates inevitable. “I suspect there are a lot more tools going on, there are a lot more tools being tested, there are a lot more things being introduced,” she said. “At the same time, AI is very new. It’s going to hallucinate. You’re going to have to work with experimentation in a way that was previously [generations] I wouldn’t have done it.

The conversation went beyond defending failure rates to discuss what successful AI implementation actually requires. Coleman stressed the importance of developing “AI fluency” among staff and recommended a collaborative approach in which technical experts work alongside business users. “How do we pair someone who is really good at technology or continuous improvement, or other kinds of revolutionary methods to look at processes, and sit side by side and not do something for you, but do something with you so that they can learn how to really integrate AI into your workflow,” she said.

Coleman also argued against the idea that enthusiasm for AI diminishes the value of human labor. “The more we talk about AI, the more people think we don’t trust humans,” she said. “It’s really important that we talk about how critical humans are in all of these workflows. So it’s about talking about when I’m freed up to do what I can uniquely do as a human.”

Wu described what she sees in successful customer deployments: a combination of top-down management support and bottom-up engagement from employees who understand daily workflows. “Leadership obviously allows employees to test and build things safely, but gives people the flexibility to experiment, try new tools, encourage them to use and develop AI and help them develop their mastery,” she said. “Your companies are full of people who live and breathe the business and have been around for decades, sometimes even centuries. And so for AI to be deployed truly effectively, you need a tool that actually works alongside the people who do the work every day.”

Klein emphasized that experimentation does not require enterprise-wide deployments. “We also see startups working side by side, bringing engineers and business leaders together,” she said. “Even though we’re in a regulated industry, we can try this in our personal lives and you know, use the weekend for non-sensitive information and just start to see how this technology works, because that’s really where you’re going to get gains and breakthroughs and big ideas.”

When an audience member asked what organizational conditions needed to be in place for AI transformation to succeed, Coleman’s response revealed the cultural shift she believes is necessary. “You have to accept failure. You have to accept messiness,” she said. “We’re talking about the entry point of this transformation. You have to be OK with experimentation, and you have to be OK with these uneven ups and downs.”

She added that companies must adopt what she calls “a learning organization” in which “managers must stop evaluating tasks and start teaching learning.” Key conditions, she said, include “vulnerability and courage” as organizations navigate technology that evolves more quickly than previous transformations.

The discussion highlighted a central tension businesses face: the risk of moving too slowly in AI adoption could ultimately outweigh the risk of the experimentation itself.

You can watch the full discussion at FortuneThe Most Influential Women event below:

For this story, Fortune used generative AI to help with a first draft. An editor checked the information for accuracy before publishing it.