Edition #34: Empower, Don't Mandate

AI adoption isn't a technology problem. It's a leadership one.

By Mykel Salomon · 2026-08-21

<h3>Why 95% of AI initiatives fail, and the human playbook of the few that don't</h3><hr><h3>🔓 Opening Reflection</h3>Companies write algorithms. Governments write laws. Someone has to write the human story.Here's something I've come to believe, and my guest this week made me believe it more: people don't resist AI because they're irrational. They resist when the story is unclear, when the purpose feels threatening, and when leaders demand adoption without building trust first.Watch how much rides on a single word. Call your AI effort cost cutting, and your people hear layoffs , and they quietly stop trusting you. Call it growth, and they start building with you. Same technology. Same tools. Opposite outcome. The framing decides almost everything that happens next.My guest this week is MJ — <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 ember-view" href="https://www.linkedin.com/in/minyang-jiang-6981a148/">Minyang Jiang</a> — Chief Strategy Officer at the fintech lender Credibly, where she leads the company's generative AI transformation. She's a Ford veteran who built a startup, spun it out, watched it fail, and rebuilt. She's a Yale Digital Ethics Fellow. And my favorite detail: she was a literature major at Harvard, now running an AI overhaul.Her core message to leaders is blunt: most of you think your AI problem is a technology problem. It's a change-management problem. The leadership work comes before the tools, not after.This edition is that work. It's not about which model to buy. It's about how to lead human beings through the most disorienting shift of their working lives, in a way that empowers them instead of frightening them.Because the companies winning with AI right now aren't the ones with the best tools.They're the ones with the best leaders.<hr><h3>🔦 Signal of the Week</h3>Two numbers explain almost everything about who's winning and who's wasting money.<ol><li>The first: 95%. That's the share of enterprise generative-AI pilots that deliver zero measurable impact on the bottom line, according to MIT's widely cited State of AI in Business 2025 study, despite $30–40 billion in investment. Only about 5% ever reach production with real value. And the researchers were emphatic about the cause. It isn't the model quality. It isn't regulation. The lead author named it plainly: it's the learning gap, for both the tools and the organizations using them.</li><li>The second number tells you where the value actually lives: 70%. BCG's research across thousands of AI engagements produced what they call the 10-20-70 rule, only about 10% of AI success comes from the algorithms, 20% from the technology and data, and a full 70% from people, processes, and culture. Companies that fund all three see roughly 3.5x higher ROI than those that pour everything into the tech. As BCG put it: the tools have commoditized, every competitor can buy the same models. What can't be bought is the organizational capability to actually use them.</li></ol>Put those two numbers together and you get the uncomfortable thesis of this whole edition:The thing quietly killing your AI ROI isn't the model. It's the 70% almost nobody budgets for. It's leadership.The 5% who win aren't smarter. They lead differently. Here's how.<hr><h3>🌍 In the Wild</h3>MJ has lived this for four years. Her playbook, section by section, with the data that backs it up.<h3>1. "Lazy leadership" — the three tells</h3>MJ has a name for the most common failure mode, and it stuck with me: lazy leadership. It shows up three ways.<ul><li>First, delegating the hardest part. A top leader says "we need to adopt this transformative technology" and then hands it to a VP or a middle manager, who's also expected to hit every other target at the same time, with no extra resources. You've taken the single hardest thing in leadership and pushed it onto someone without the bandwidth to do it.</li><li>Second, outsourcing strategy to AI. Ask what the transformation actually is, and the answer comes back: "We're going to be AI-first." That's not a strategy. The North Star should exist before the tools. When "AI-first" becomes the strategy, it's usually covering for the fact that there wasn't a real one to begin with, a hot buzzword standing in for deep thinking.</li><li>Third — and this is the one that quietly destroys trust, mandating AI you've never used yourself. Leaders show up with bold demands ("why can't we cut headcount here?") having never felt the actual constraints of the tools. As MJ put it, the antidote is almost countercultural: leaders should openly share the stories of when they experimented with AI and failed. That builds more credibility than any mandate ever will.</li></ul>The data agrees. BCG found that the CEOs seeing results now spend around eight hours a week personally building their AI fluency. The ones who don't are running on secondhand hype, and their people can tell.<h3>2. Empower vs. mandate — and the cautionary tale of Klarna</h3>Here's the distinction at the heart of the whole conversation. Mandate says: use this, or else. Empower says: I'm going to help you feel capable with this. Mandates can produce compliance in the short term, but MJ warns the long-term repercussions are real: distrust, disengagement, and people quietly sabotaging the very rollout you're forcing. And if you're frustrated by low adoption? That frustration is your problem to solve as a leader, not a stick to beat your team with.The cleanest cautionary tale is Klarna. In 2024, the fintech proudly announced its AI assistant was doing the work of some 700 customer-service agents, handling millions of chats across dozens of markets, as headcount fell more than 20%. Then the brand started breaking. Customer satisfaction dropped. By 2025, the CEO publicly admitted the company had leaned too hard on efficiency, that cost had become, in his words, "a too predominant evaluation factor", and began rehiring humans for the complex, emotional, high-stakes conversations where AI fell short.To be fair to Klarna, they didn't abandon AI, they rebalanced it, keeping automation for routine questions and returning humans to the hard ones. And that's precisely the lesson. The failure wasn't the AI. It was the cost-first framing. Research now shows 55% of companies that ran AI-driven layoffs regret it, and Gartner projects that over 40% of agentic AI projects will be cancelled by 2027. Hybrid beats wholesale replacement, almost every time.<h3>3. Cost-cutting is not a strategy</h3>MJ's sharpest strategic point: if cost-cutting is all you're doing, you don't have a strategy — you have something your competitors can copy in a quarter.And when cost is your only benchmark, watch what happens (she lived all of this): people start using AI in secret because they don't feel sanctioned to admit it — shadow IT blooms. Fear sets in — is it coming for my job? — and nobody says it out loud, but, in her words, "you can hear it in the silence." The best, most valuable ideas never make it to the table because everyone's chasing the same low-hanging efficiency fruit. And the kicker: the costs don't even disappear. MJ is on the billing, her API costs went up. Costs don't vanish with AI. They shift.The shadow-IT problem is everywhere: MIT found employees at over 90% of companies already use personal AI tools at work, while only about 40% have official licenses. The winners don't punish that — they study it and sanction it.MJ's fix is a discipline every leader can steal. Don't measure only "human capital saved." Track at least five things: Is the organization more nimble? Are people more capable with AI? Are you adapting faster? Are you shipping projects with longer-term returns? And her exercise for the cost-obsessed leader: write down every efficiency and cost-cutting idea you have — then put that list away. Now write the ideas that aren't about cutting costs. They'll come slower; they're less obvious. But those are the differentiated, revenue-growing, hard-to-copy ones. Do your efficiency work — and commit to executing at least one idea from the second list.<h3>4. The timeline nobody wants to hear</h3>If you take one thing from MJ for your next board meeting, make it this.She walked me through her actual timeline building an AI-native capability. Year one: they produced nothing. Nothing. Year two: two things — which happened to become two patents. Year three: six things. Year four, now: multiple teams running in parallel, self-sustaining, with junior, non-technical people building working agents. Her honest verdict: "It took me three years to find a human-AI hybrid model that works. If you tell me you found it in year one, I'll have a hard time believing you." A full year, she says, just to change how people think about their work.This is why quick-ROI mandates collapse. Adoption, MJ argues, is like brand trust or customer loyalty — it only becomes real once it's been stress-tested over time. You can't shortcut it, and the impatience is exactly what Gartner's 40%-cancellation forecast is measuring.<h3>5. The mechanics that actually work</h3>Here's the stealable playbook, the specific moves that separated MJ's 5% outcome from the 95%.Sequence it backwards. She did not start with a strategy. She started with culture, excitement, and training, nine-plus hours across the whole company, taking outside classes herself and bringing them back, then built an in-house lab, rotated people through it, and only then wrote the strategy, once they'd learned enough to know what the future of work actually looked like.Turn recipients into builders. Belief doesn't come from watching a demo. It comes from building something yourself. That's the conversion moment.Run a Skunkworks. Modeled on Lockheed Martin's, MJ's is a small, elite, cross-functional team handed "unsolvable problems" and almost no resources, no lunches, no t-shirts. As she tells them, "the work and the company you keep" are the reward. Crucially, confidence trickles down: managers who built it now pass it to junior people who'd never touched AI. The data backs this exactly , BCG found peer learning is the number-one driver of AI skill, with 69% of people learning most from colleagues. Build the champion network.Kill the barriers to entry. When MJ wanted to open up Claude access, the CFO wanted employees to write business cases to justify the cost. She and the VP of Product refused , that's a barrier. Instead: free access, 30 days, go play, then tell us what you tried. Many people dropped the access afterward — and that was still a win, because of what everyone learned (including discovering which of their problems were actually automation problems, not generative-AI ones). Don't gate curiosity.Change the metric. Expect roughly 90% of your small AI projects to fail — and set that expectation flat, up front. Then measure learning, not just outcomes. "Did you get better at working with AI?" is a legitimate corporate success metric. This aligns with the most counterintuitive MIT finding of all: the winning 5% design for friction — they embed AI into real, messy, high-value workflows instead of erasing the very complexity where the value (and the learning) lives.<hr><h3>💬 The Big Question</h3>Two questions, and they're for the person in charge:Are you empowering your people, or just mandating their compliance?And the one MJ would put to any CEO:If AI is truly the most important strategic work your company will do, why have you delegated it? And when did you last sit down and fail at it yourself?<hr><h3>🧠 A Small Exercise</h3>Five moves for a leader this week. Not for your team, for you.<ol><li>Do the thing you're asking of them. Spend an hour actually building something with AI. Feel the constraints yourself before you set the expectations.</li><li>Run the two lists. Write your efficiency ideas. Set them aside. Now write the ideas that grow something new. Pick one from the second list and commit to it.</li><li>Remove one barrier to entry. Find the approval, the business case, the permission slip standing between a curious employee and a tool — and take it away for 30 days.</li><li>Add one learning-based metric. Alongside your outcome metrics, start measuring whether your people are actually getting more capable. Celebrate that.</li><li>Tell one failure story. Share, out loud, a time you tried something with AI and it flopped. Watch what it does for the trust in the room.</li></ol><hr><h3>🎙️ This Week on The Human Protocol Podcast</h3><ul><li>Empower, Don't Mandate: The Human Work of AI Adoption</li><li>Guest: MJ ( <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 ember-view" href="https://www.linkedin.com/in/minyang-jiang-6981a148/">Minyang Jiang</a> ) — Chief Strategy Officer at Credibly, Yale Digital Ethics Fellow, and one of the most honest builders I've spoken with</li></ul>MJ leads generative AI transformation at a fintech lender, and refuses to sell the fantasy version. A Ford veteran and Harvard literature major who has built, failed, and rebuilt, she talks about what adoption actually takes when the people involved are, well, people.In this episode, you'll learn:<ul><li>The three tells of "lazy leadership" — and how to avoid them</li><li>Why "empower" and "mandate" lead to completely different companies</li><li>Why cost-cutting is the one AI strategy your competitors can copy overnight</li><li>The real timeline of AI transformation — and why year one may produce nothing</li><li>How to turn employees from passive recipients into confident builders</li><li>Why you should expect 90% of your AI projects to fail — and measure learning instead</li><li>Why learning requires friction — and what that means for how you lead</li></ul>🎧 Available now on <a target="_self" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 zlkRYSXVULlfDyumitBTFelWBFbIpmaxywZnES " href="https://open.spotify.com/episode/76Vuu1WE3R9L7PBy4ruvhZ?si=ggXqbvVuT1aLfO5vLZ9sSg">Spotify</a>, <a target="_self" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 zlkRYSXVULlfDyumitBTFelWBFbIpmaxywZnES " href="https://podcasts.apple.com/us/podcast/s2e33-why-your-ai-problem-is-really-a-leadership/id1843402023?i=1000784339849">Apple Podcasts</a>, and <a target="_self" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 zlkRYSXVULlfDyumitBTFelWBFbIpmaxywZnES " href="https://youtu.be/IPRwiRIGp8g?si=n82Q2WDaqVZocov0">YouTube</a>.<img class="max-w-full h-auto rounded-md my-4" src="https://media.licdn.com/dms/image/v2/D4E12AQEEGx1NjVMRTw/article-inline_image-shrink_1000_1488/B4EaAk4O2iIkAI-/0/1787325116413?e=1788998400&v=beta&t=5fBczb06UtdAyPGw0U35Bm0KJUN3OH2dYG6Y7FZ9pNc" alt="Article content"><hr><h3>🧭 Thought to End the Week</h3>I asked MJ what advice she'd give a younger person stepping into this era. Her answer wasn't about AI at all.Read, she said. Don't take the shortcut of swallowing knowledge whole. And remember that learning comes with friction , if you're learning without friction, you're not actually learning. The most important skill in life is learning how to learn, and that is the one thing AI cannot do for you.I think that's the whole edition in miniature. The leaders who win this next era won't be the ones who forced the fastest adoption or bought the flashiest tools. They'll be the ones who made it safe to try, safe to fail, and worth believing in. Who understood that you cannot mandate trust, or shortcut it, any more than you can mandate someone's loyalty or their creativity.You have to earn it. You have to build it. You have to be willing to go first, and to fail out loud.That's not the easy path. It's slower, it's more human, and it's the only one that actually works.Empower. Don't mandate.The code may change. And it will.But our humanity is the constant.See you next week.— Mykel<hr><h3>📚 Sources</h3><ol><li>MIT NANDA, The GenAI Divide: State of AI in Business 2025 — 95% of enterprise GenAI pilots deliver zero P&L impact; the cause is the organizational "learning gap," not model quality; 90%+ of firms have "shadow AI" use while only ~40% hold official licenses. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.forbes.com/sites/jaimecatmull/2025/08/22/mit-says-95-of-enterprise-ai-failsheres-what-the-5-are-doing-right/">https://www.forbes.com/sites/jaimecatmull/2025/08/22/mit-says-95-of-enterprise-ai-failsheres-what-the-5-are-doing-right/</a></li><li>MIT, via Forbes — the winning 5% "design for friction," embedding AI into high-value workflows rather than erasing the complexity where value lives. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.forbes.com/sites/jasonsnyder/2025/08/26/mit-finds-95-of-genai-pilots-fail-because-companies-avoid-friction/">https://www.forbes.com/sites/jasonsnyder/2025/08/26/mit-finds-95-of-genai-pilots-fail-because-companies-avoid-friction/</a></li><li>BCG, AI Radar 2025 / Closing the AI Impact Gap — the 10-20-70 rule (10% algorithms, 20% tech/data, 70% people, process, and culture); proportional investors see ~3.5x higher ROI; peer learning is the #1 driver of AI skill (69%). <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.bcg.com/publications/2025/closing-the-ai-impact-gap">https://www.bcg.com/publications/2025/closing-the-ai-impact-gap</a></li><li>Forbes / BCG — why the 10-20-70 principle should matter to CEOs; leaders now spending ~8 hours a week building AI fluency. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.forbes.com/sites/joemckendrick/2026/01/26/why-ais-10-20-70-principle-should-matter-to-ceos-and-everyone-else/">https://www.forbes.com/sites/joemckendrick/2026/01/26/why-ais-10-20-70-principle-should-matter-to-ceos-and-everyone-else/</a></li><li>Reuters / Bloomberg, via multiple outlets — Klarna's AI assistant handled the work of ~700 agents; CEO Sebastian Siemiatkowski later said cost had become "a too predominant" factor and rehired humans for complex support. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.mavenagi.com/resources/klarna-ai-reversal-future-customer-experience">https://www.mavenagi.com/resources/klarna-ai-reversal-future-customer-experience</a></li><li>Klarna reversal analysis — 55% of companies that ran AI-driven layoffs now regret it; Gartner predicts 40%+ of agentic AI projects cancelled by 2027. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.digitalapplied.com/blog/klarna-reverses-ai-layoffs-replacing-700-workers-backfired">https://www.digitalapplied.com/blog/klarna-reverses-ai-layoffs-replacing-700-workers-backfired</a></li><li>Ethan Mollick, Co-Intelligence — the crowd, the lab, and the strategy as the three pillars of organizational AI adoption (referenced by MJ). <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.penguinrandomhouse.com/books/741805/co-intelligence-by-ethan-mollick/">https://www.penguinrandomhouse.com/books/741805/co-intelligence-by-ethan-mollick/</a></li></ol>