About Ethan Mollick

AI as Co-Intelligence and the Jagged Frontier

Mollick conceptualizes AI not as deterministic software but as 'co-intelligence'—a non-deterministic partner with a 'jagged frontier' of inconsistent capabilities across tasks.[23][25][51] Users must shift from 'centaur' (divided) to 'cyborg' (blended) workflows, investing ~10 hours to map AI's strengths via experimentation.[23][24] This applies to individuals delegating undesirable tasks and organizations fostering rapid testing.[24][55]

Organizational Transformation and AI Adoption

AI demands rethinking structures, moving beyond ROI to foundational change via 'Leadership-Lab-Crowd' frameworks.[46][52][53] Enterprises face inertia, with adoption lagging due to IT-centric approaches; leaders must decentralize access and incentivize transparent use.[26][44] Experiments show structured AI guidance boosts startup revenue 1.9x.[37] Parallels exist between AI agents and organizational design, requiring delegation, incentives, and hand-off management.[28][29][30]

Future of Work and Productivity Impacts

AI disrupts white-collar jobs, automating tasks and leveling skills—e.g., GPT-4 boosted BCG consultants 40% in quality.[2][57] Gains are largest for low performers, threatening apprenticeships and causing 'cognitive deskilling.'[25][51] Success hinges on personal moats, augmentation over replacement, and AI-first strategies amid Coasean cost reductions.[21][41]

AI in Education and Apprenticeship Crisis

LLMs exceed humans in academic tasks, obsoleting homework and demanding pedagogical redesign.[56] AI accelerates research but erodes entry-level 'grind' essential for expertise.[25][51] Educators should pioneer ethical integration, treating AI as an 'alien person.'[55][56]

LLM Capabilities, Limitations, and Scaling

Mollick highlights LLM weaknesses in coherent fiction[9][11] and agentic judgment for on-device models,[31][32] but notes scaling via tokens boosts cybersecurity performance 3x.[33][34] Frontier models enable agents, though chatbots hinder diagnosis.[38] Outputs often seem meaningful due to human projection.[9][10]

AI Safety, Security, and Geopolitics

General-purpose LLMs like Mythos spawn security risks emergently.[16][17] Red-teaming is vital; prompt injections fail frontier models (except Gemini).[5][39] Geopolitical leads are fleeting amid Chinese advances.[6]

Enterprise and Strategic Shifts

Post-2025 GenAI impact accelerates with agentic tools.[27] Embrace AI 'weirdness' over normalization.[40] 'Bitter Lesson' favors general ML; firms must experiment to avoid obsolescence.[49]

Innovation, Experimentation, and Entrepreneurship

Moonshot thinking counters inertia.[1] AI enables one-person billion-dollar companies, structural book innovations.[44][50] Startups thrive via case studies.[37][55]