You Need to Fail at AI Before You Can Succeed with It
Accenture just eliminated 11,000 jobs in an $865 million restructuring—roles that “didn’t meet the demands of AI.” But the companies replacing workers aren’t the ones winning. The winners? Those where leaders welcome something uncomfortable: failing at AI before they try to lead it.
Listen to “Why You Need to Fail at AI Before You Can Succeed with It?” Deep Dive
As part of an $865 million restructuring program, Accenture reported that it has eliminated more than 11,000 jobs over the last three months. These roles were eliminated due to not meeting the demands of artificial intelligence (AI). Accenture has joined the ranks of UPS, Amazon, Duolingo and MSN, all of whom are in the process of or have implemented AI solutions as a way to replace their workforce.
But here’s what I have come to realize: the companies replacing workers with AI aren’t necessarily the ones that are winning. The businesses that will win and benefit from AI are those whose leaders learned to fail at AI first. That’s the only way for you and your company to succeed.
The Problem: Learning Without Failing Creates Hollow Competence
As ScheinerInc. shared previously, AI is now the universal job requirement. Yet, more than 70% of digital transformations fail because of a “lack of buy-in.” But the real issue runs deeper than this: When it comes to learning and development, the real transformation begins with people. The AI conversation should focus on the human adaptation part of the process—i.e., helping employees reimagine their roles and how they problem-solve, and what human-agent collaboration should look like.
“Allowing for failure is just as important as a way to learn. In other words, failure isn’t just acceptable, it’s essential.”
As I’m sure you’ve witnessed in your own business with anything related to innovation, leaders who never struggle through the messy trial-and-error of AI experimentation never develop the judgment to know when AI is helping versus when it’s just underdelivering on expectations, which means they can’t spot new opportunities, can’t evaluate their team’s AI work and can’t recover when these implementations fail. How can you be both judge and jury when you haven’t struggled through the AI evaluation process?
While all businesses measure success by successful outcomes, I believe that allowing for failure is just as important as a way to learn. In other words, failure isn’t just acceptable, it’s essential.
Franklin Had to Fail
In 1727, Benjamin Franklin, who was already a successful printer, did something unusual: he asked a group of younger tradesmen and craftsmen to teach him. This Junto, also known as the Leather Apron Club, wasn’t a lecture series where experts dispensed wisdom; rather, it was a workshop to which members brought real problems, struggled through solutions together and learned by doing.
The key detail that most people miss in this history lesson is that Franklin didn’t just listen to ideas about better business practices; he implemented them, failed, adjusted and came back the next week to report what happened. The Junto wasn’t about preventing failure; rather, it was about creating a safe space in which TO fail.
Who’s Your AI Sherpa?
“Driven by the concept of reverse mentorship, leaders need to be taught not by watching the AI work, but by tapping into the enthusiasm and insight of an employee who lives and breathes AI.”
Three centuries later, business leaders face the same choice Franklin did. Jeremy Utley, a leading innovation advisor and adjunct professor at Stanford, believes that it’s okay for leaders to admit what they don’t know about AI as a way to gain credibility—and then doing something about it. Utley describes these leaders’ personal AI Sherpa as “an AI-knowledgeable coworker for the purpose of AI discovery on the leader’s behalf. Sherpa shadows the leader for 2-3 days. Listen to their meetings, understand their workflows, identify the AI opportunities they can’t see. Build custom GPTs for their specific challenges. Make them fluent, enthusiastic users.”
While Utley’s concept is brilliant, here’s the critical piece most will miss:
If the Sherpa BUILDS you three custom GPTs and shows you how to use them, you learned tools, not capability. That’s curated competence. You look credible, but remain dependent.
If the Sherpa SITS WITH YOU while YOU try to build, get bad outputs, waste time, iterate and eventually succeed, you learned judgment. That’s real capability.
The distinction matters. Driven by the concept of reverse mentorship, leaders need to be taught not by watching the AI work, but by tapping into the enthusiasm and insight of an employee who lives and breathes AI—someone who can sit with them through the frustration, the failures and the eventual breakthroughs. This is what reinforces learning that is geared specifically toward leadership.
What Productive Failure Looks Like
This idea of creating a structured space for productive failure isn’t new. In his book Creativity, Inc., Ed Catmull, one of the co-founders of Pixar, discusses fostering creativity through candor. By creating an environment where candor is possible and even welcomed, directors and creative teams can give and receive direct and constructive feedback to each other, raising the game of all products in the process. One tool that Pixar uses to achieve this level of candor is called the Braintrust.
The Braintrust doesn’t prevent failure; rather, it creates a structured space in which to fail safely and iterate. The parallel to AI learning is clear: leaders need to experience why their prompts lack context, why their outputs are incorrect and why their workflows are failing.
Success in sports follows the same principle when defining a top-down and bottom-up culture and legacy. In their 100+ years of existence, the New Zealand All Blacks Rugby Team has a 75%-77% winning record, making them arguably the most successful sports franchise in history. In a team like the All Blacks, both coaches and players are expected to be leaders. Their philosophy: give the players ownership and dual-manage the All Blacks with a group of players and a group of coaches, which leads to autonomy, mastery and emotionally intelligent athletes.
The critical detail: players review their own game footage FIRST—they have to see their failures themselves before coaches intervene. Everyone leads by doing, struggling and then improving. The same could be said for today’s leaders embarking on AI initiatives— i.e., these leaders need to get immersed in AI themselves in order to help their teams become comfortable and ultimately successful with AI solutions.
Why This Matters
Those companies eliminating jobs assume AI is a straight replacement for human capability. But the businesses that will actually win are those whose leaders developed AI judgment through hands-on struggle. You can’t lead AI transformation if you don’t understand—fundamentally and experientially—what AI can and cannot do.
When implemented correctly—with leaders actually doing the work, rather than watching it be done—the AI Sherpa model creates a Learn-Fail-Adapt framework: leaders learn by attempting real problems, fail in safe conditions with guidance and then adapt their approach based on those failures. This cycle, repeated over time, builds genuine organizational capability and these leaders’ AI credibility as well.
Guiding Your AI Sherpa
If you’re considering an AI Sherpa program, ask one question: Will I be the one typing the prompts? If the Sherpa says they’ll handle it, thank them and then find someone else. Over the course of a few days of reverse mentoring, the company’s leader should:
Day 1: Attempt a real work problem with AI—and fail (Sherpa observes).
Day 2: Iterate on those failures (Sherpa helps you understand why).
Day 3: Tackle a new problem with better judgment.
Your mandate: Stop designing AI training to prevent failure; instead, start designing it to make failure safe, instructive and inevitable.:
The Choice
Benjamin Franklin understood that admitting “I don’t know” was the beginning, not the end, of learning. Pixar’s Braintrust fosters creativity through candor, and the All Blacks’ dual management of players and coaches is founded on the belief that everyone leads by doing, struggling and only then improving.
AI is no different. The leaders who will succeed aren’t those who know how to use the tools; rather, they’re the ones who know how to fail with them and how to help their teams do the same.
The question isn’t whether your organization will fail with AI; it’s whether you’ll create the conditions to fail productively by making AI failure safe, instructive and inevitable.
As always, thank you for reading and/or listening. Questions about your AI strategy? Email me here.
