How Great Leaders Turn AI Spend Into Real Profit | Matt Beane, UC Santa Barbara
In a Nutshell
Leaders must stop rewarding volume of AI output and instead incentivize restraint—actively killing B+ work to protect A+ standards and real skill development. Expertise erodes when novices lose access to challenge, complexity, and trusted mentor relationships, so organizations need deliberate structures like inverted apprenticeships and direct leadership engagement with tools to preserve learning. Without these shifts, deskilling will damage productivity and economic health within years, even as AI capabilities advance.
These notes were generated by AI and may contain inaccuracies.
2025 was characterized by widespread experimentation with AI across organizations, with broad licensing and efforts to generate results through trial and learning. In 2026, boards of directors are expected to demand measurable returns on AI investments from CTOs. Jensen Huang of Nvidia has stated that if paying $500,000 for an engineer, they need to be burning at least $250,000 worth of tokens.
AI predominantly generates B+ content in large volumes at low or no cost, which risks eroding recognition of A+ quality. Wise leaders reward individuals with cash, promotions, or visible recognition for stopping B+ ideas from advancing. This restraint is essential because unhealthy organizations pursue many ideas indiscriminately, resulting in excessive B+ outputs. Expertise's most valuable function is often deciding what not to pursue.
Matt Beane, Associate Professor at UC Santa Barbara in the Technology Management department and CEO/co-founder of Skillbench, wrote the article "Don't Let AI Dumb You Down" on Substack 2.5 years ago, identifying AI as a major risk to skill maintenance. The ability to produce large volumes of code or documents cheaply raises questions about whether such work should be done. Without active steps to protect learning, individuals, organizations, and future generations risk deskilling, harming the economy within a few years.
Differences in outcomes depend on individual disposition—some are inherently driven to maintain independent work and results—along with organizational incentives and social structures. When job metrics emphasize producing more lines of code or burning large quantities of tokens, efficiency questions arise: whether activity equates to meaningful progress or merely running in circles. Leaders must establish clear thresholds distinguishing B+ from A+ ideas.
Organizations must create incentives for insisting on superb ideas only. Individuals who champion A+ ideas demonstrate how innovation works through patience, persistence, and refusal to accept insufficient quality. In 2018, Beane published the first paper from his dissertation on robotic surgery, introducing the concept of shadow learning.
Shadow learning refers to building skills through non-approved or culturally inappropriate means. When new technologies disrupt work, they also disrupt on-the-job learning, which primarily occurs through interaction with experts. Across over 35 occupations studied, experts using new tools perform more work independently, reducing novice participation opportunities. Most individuals attempt traditional learning methods and struggle, while a small subset develops rule-bending approaches to maintain skill development.
In robotic surgery, effective medical residents operated on patients without senior surgeons present (inappropriate but not illegal) and watched recorded surgical videos at rates approximately 100 times higher than typical residents. Shadow learners across occupations invent deviant methods to protect skill-building opportunities when normal pathways are blocked. These practices are not recommended for adoption but serve as diagnostics for identifying what elements of work experience must be preserved.
Analysis of shadow learners revealed three consistent protective elements, termed the three Cs in The Skill Code:
Challenge
Learners must operate close to but not beyond the edge of their capability. The work requires intense focus, feels stressful, and involves performing slightly below peak due to strain. Healthy levels of challenge are necessary for building skill.
Complexity
Beyond focal tasks, learners engage with the broader work system—such as interactions with nurses, scrub techs, supply chains, hospital finances, and IT systems. They allocate time for reflection on the entire system, improving ability to handle surprises and generate new ideas. Individuals can cultivate curiosity about the broader context, while leaders can implement structures like job rotation programs to facilitate exposure.
Connection
A bond of trust and respect between individuals drives learning. Novices are motivated to improve to earn mentors' trust and respect, while seniors find meaning in developing juniors. This relationship also provides functional benefits, as trusted seniors offer future opportunities and support.
Leaders should obtain primary data through direct engagement with how work occurs. Effective leaders spend significant personal time using advanced AI tools, building things, demonstrating value, and openly reporting failures and waste to model learning. They signal that no one yet fully understands optimal AI usage.
Recent trends show firms slowing junior-level hiring while retaining senior talent, which is shortsighted. Organizations should assertively hire AI-native juniors capable of exceptional AI use despite limited professional experience and establish inverted apprenticeship dynamics. This enables bidirectional learning where seniors learn from juniors' fresh perspectives and juniors learn from seniors. Healthy organizations accept short-term productivity costs to prepare for future demands.
General-purpose technologies like AI create initial messiness with no immediate correct approaches. High standards should coexist with forgiveness toward self and others during inevitable waste and mistakes. AI may surpass humans across all tasks—including empathy, judgment, and creativity—within 3 to 40 years. Institutional systems may not adapt sufficiently quickly, risking disruptive change. The choice exists to integrate AI into solutions that create a less painful, more beneficial future for more people, rather than allowing change to unfold painfully. The goal is a future where AI's arrival benefits everyone, requiring immediate collective action.
Keep EO in your library
Save the videos and channels worth coming back to, and find them again in one place.
