Harvard AI Professor: The People Most at Risk From AI Aren’t Who You Think

Executive Summary

Harvard Business School professor Karim Lakhani argues that AI is changing how work and organizations are structured, but effective use still depends on human direction, judgment, and accountability. The biggest risks may not fall only on entry-level workers: Lakhani says resistant middle managers could be more exposed, while juniors who learn to use AI may move faster.

His central advice to leaders is to use AI themselves, redesign work around specific business outcomes, and keep updating how tasks are divided between people and models.

Work Styles and Learning

In a BCG consultant study, the interview cites 244 consultants working in three styles. Lakhani also describes a broader randomized experiment involving about 800 consultants:

• Cyborgs: Work with AI throughout a task, learning where both the person and model are strong. Lakhani sees this as especially relevant to R&D, sales, and marketing.
• Centaurs: Divide work between a person and AI, then review the AI’s contribution. This can work, but risks becoming outdated as model capabilities change.
• Self-automators: Hand off the task with little engagement. Lakhani says this approach produces the weakest learning outcomes and can lead people to accept plausible but incorrect work.

His practical distinction is that delegation is useful when the person still evaluates the work and learns from it. Simply passing along AI output leaves little human value or skill development.

Practical AI-Use Playbook

• Treat AI as a collaborator, not an authority. Don’t accept the first answer; question its assumptions and check its work.
• Make the model challenge you. Ask it to argue against your proposal, produce counterarguments, and explain the evidence and confidence behind its claims.
• Look for failure cases. Request further research, red flags, alternative explanations, and possible “black swans.”
• Invest in deeper work when needed. Lakhani recommends asking models to explore beyond their default, concise answers.
• Reassess task boundaries frequently. A task that a model struggled with may become feasible after a new release; capabilities can also vary unexpectedly between similar tasks.
• Retain accountability. “AI gave it to me” is not an excuse for poor-quality work: the person putting their name on the output remains responsible.
• Learn fundamentals before automating. Understand enough of a discipline to judge AI output, then use AI to take on more volume and complexity. Lakhani illustrates this with learning basic integration by hand before using a machine for many more examples.

Management and Organizational Tactics

• Require hands-on AI experience from leaders. In executive sessions, Lakhani has leaders build an application for a real business problem in 90 minutes. He says this can expose the gap between leaders’ assumptions and what current tools can do, and help them identify where workflows need to change.
• Set business goals, not “AI adoption” goals. Define desired results in areas such as customer service, R&D productivity, product launches, or operations, then work backward to determine how AI could contribute.
• Be ambitious and translate ambition into action. Lakhani advocates setting deliberately stretching goals. He gives a 10x goal with a 7x result versus a conventional 2x result as an illustration, not a forecast. Then ask repeatedly, “What else must be true?” to identify the required changes.
• Rethink the operating model. AI implementation is not just about adding tools; leaders may need to reconsider how the business creates and captures value and how work is organized.
• Address uneven adoption. Lakhani describes a “K-shaped” pattern in which a small group of heavy AI users pulls ahead while the median user stays relatively flat. He argues companies should help more employees progress beyond basic, search-like usage.
• Build in continuous learning. Models and workflows change rapidly; organizations need regular experimentation, evaluation, and updates rather than fixed assumptions about which tasks AI can handle.

Metrics and Evidence Cited

• In the consultant breakdown cited by the interviewer, 60% were described as cyborgs, 14% as centaurs, and 27% as automatizers. The figures total 101%, likely due to rounding or subtitle transcription.
• Lakhani describes a broader BCG experiment involving about 800 consultants randomly assigned to tasks with or without AI.
• In usage data from large organizations covering hundreds of thousands of people, he says users around the 95th and 99th percentiles used AI 10–20 times more than the average. He characterizes this as a widening usage gap, not proof that AI use alone caused better performance.
• Lakhani says their current organizational data did not yet show a correlation with performance. He separately cites software-industry data suggesting 10–20x performance differences between top performers and others.
• A Procter & Gamble study is described as finding that one person using AI could match the productivity of a two-person team.
• At Harvard Business School, the AI course is required for MBA students and has been revised for three years. The school assigned 900 MBA students AI “vibe coding” projects; Lakhani says roughly half of the spring classes would focus on building AI products and tools.
• Lakhani reports that some CEOs at large companies spend 20–30% of their time discussing AI; one global firm reportedly devoted 50% of management-meeting time to AI’s business impact.

Products, Education, and Examples

• Harvard’s applied AI curriculum: Students build products and business solutions, including RAG systems, evaluations, and multi-agent systems. The course shifted from a conventional final exam to project work using AI.
• Executive training: Leaders are asked to build practical tools for real operational challenges, rather than toy applications. Lakhani says this helps them understand both AI’s capabilities and the organizational changes required.
• Career development: AI may remove some routine junior tasks, but Lakhani also sees it as a possible “springboard” for beginners. Juniors need to deliberately build judgment and expertise instead of relying on copy-and-paste output.
• Company adaptation: Lakhani contrasts Amazon’s reinvention of retail with retailers that merely added websites to existing operations. He frames the strategic question for incumbents as whether they will adapt deeply, as Walmart eventually did, or fail to rethink the business.
• Business moat test: Lakhani cites OpenAI president Greg Brockman’s question: “Does your business become more valuable with the new base model, or does it disappear?” Lakhani says leaders themselves must answer this rather than delegating the strategic judgment.

Who May Be Most at Risk

Lakhani’s view is that middle managers who resist AI and organizational change may face greater risk than entry-level workers as a group. He notes that young people may be quick to experiment, while some experienced managers are insulated from AI’s capabilities.

However, he cautions that the weak market for young professionals has multiple causes and should not be attributed solely to AI. At any level, work that amounts only to copying model output or acting as a thin wrapper around a base model is vulnerable.

Sources and Presenters

• Guest: Karim R. Lakhani, Harvard Business School professor.
• Interviewer/presenter: Silicon Valley Girl host; not named in the subtitles.
• Research and examples cited: BCG, Procter & Gamble, H&M, Fabrizio Dell’Acqua, Mitch Weiss, Stephen Wolfram, Ram Koning, Greg Brockman, and Kaifu Lee.
• Sponsor mentioned: ChatLLM by Abacus AI.