The Jagged Frontier of Deployment | Karim Lakhani | ADIA Lab Symposium 2026

Value of the Information & Strength of the Argument

The value of the information is high because it is based on causal evidence from randomized controlled trials, which is rare in AI impact research. The argumentation is solid: it combines quantitative performance metrics with qualitative insights (e.g., identity threat, junk food reaction) to provide a nuanced view. The speaker transparently distinguishes between high-certainty RCT results and lower-certainty correlational findings, strengthening credibility. The discussion of the J-curve and organizational absorption adds practical relevance for decision-makers.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is exemplary, relying on peer-reviewed or forthcoming studies with large sample sizes and randomized assignment. Sources include field experiments at BCG and Procter & Gamble, as well as the book ‘Competing in the Age of AI’ (2020). The title precisely reflects the talk’s focus on deployment challenges and the jagged frontier concept. The presentation includes no promotional content; the only source linked is the symposium website. The talk’s methodology and transparency about data quality enhance its reliability.

The title accurately reflects the content: the talk focuses on the 'jagged frontier' of AI deployment, presenting field experimental evidence and organizational implications, perfectly matching the title.

8/10

The talk presents rigorous field experimental evidence from large-scale RCTs involving more than 1,500 professionals at BCG and P&G, combined with established economic and organizational theories. The speaker is a Harvard professor and co-author of widely recognized research, though some findings are preliminary or correlational.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:40Host introduces Karim Lakhani, highlighting his research on AI and the 'jagged frontier'.
• 02:30Lakhani opens with the question: is AI a difference of degree or a difference of kind compared to the internet?
• 05:00Discusses AI as a general purpose technology, creating supply and demand shocks.
• 08:00Introduces the BCG field experiment design: 758 consultants assigned to strategy and analytical tasks with and without GPT-4.
• 12:00Presents results: 12% more tasks completed, 25% faster, 40% quality improvement on strategy; performance degradation on analytical tasks (85% vs 60-70% correct).
• 15:00Highlights the 'junk food' reaction from a consultant and the identity threat experienced by top performers.
• 18:00Describes three phenotypes of AI usage: centaurs, cyborgs, and self-automators.
• 22:00Begins P&G study on teams: individuals with AI match teams without AI in average quality.
• 28:00Teams with AI are 3 times more likely to produce top 10% ideas; expertise convergence between R&D and commercial staff.
• 35:00Discusses the J-curve effect: firms see a three-year performance dip after AI adoption before improvement, citing Christina McLaren's research.

• ADIA Lab Symposium 2026 — Official page of the symposium where this keynote was presented.