Divya van Mahajan

How to Make AI Work at Work

ai-daily-brief-podcast

How to Make AI Work at Work

Overview

This episode of The AI Daily Brief (recorded May 25, 2025, a Memorial Day weekend “long read” episode) features host NLW (of Superintelligent) reading and providing commentary on a post by Professor Ethan Mollick titled “Making AI Work: Leadership, Lab, and Crowd.” The central thesis is that while AI demonstrably boosts individual worker performance, organizations are failing to capture those gains at the organizational level — and that fixing this requires a deliberate combination of leadership vision, centralized experimentation (“the Lab”), and bottom-up employee innovation (“the Crowd”). NLW broadly agrees with Mollick’s framework but argues it underestimates the pace of change, particularly the emergence of AI agents as a transformative forcing function.

Source video URL: (not provided)

Prerequisites

  • Basic familiarity with generative AI tools (ChatGPT, GitHub Copilot, enterprise co-pilots)
  • General understanding of enterprise change management and organizational design
  • Awareness of the distinction between AI chatbots/assistants and AI agents (autonomous, task-executing systems)
  • Familiarity with terms like “vibe coding,” “deep research,” and “AI upskilling” is helpful but not required

Main Points

1. Four Key Facts About AI Adoption

  • Workers report significant personal productivity gains: A Danish study found workers believed AI halved working time for 41% of their tasks; a U.S. survey found workers reported tripling their productivity.
  • AI usage at work is widespread and growing: The same Danish study found 65% of marketers, 64% of journalists, and 30% of lawyers had used AI at work; U.S. usage grew from ~30% in December 2024 to ~40% by April 2025. KPMG’s Pulse Survey found daily AI usage in enterprises jumped from 22% to 58% between Q4 2024 and Q1 2025.
  • Transformational capabilities are available but underutilized: Deep research tools compress hours of analytical work into minutes; agents capable of executing real tasks are emerging; model quality is increasing rapidly.
  • Organizational performance gains remain elusive: Companies report only small-to-moderate AI-driven gains; there is no significant impact yet on wages or hours worked as of end of 2024.

2. The Core Tension: Individual Gains vs. Organizational Gains

  • Individual AI productivity gains do not automatically translate into organizational performance improvements.
  • Capturing organizational gains requires organizational innovation: rethinking incentives, processes, and the fundamental nature of work.
  • The “muscles” for organizational innovation have atrophied; companies have historically outsourced this to consultants or enterprise software vendors. Those generalized approaches do not apply here because no one — including major AI companies — has a definitive playbook for any specific organization.

3. Leadership: Vision and Urgency

  • Leadership must move beyond signaling urgency (as seen in viral memos from CEOs of Shopify and Duolingo) to painting a vivid, concrete picture of what the AI-powered future looks like for the organization.
  • Workers need answers to specific questions: Will efficiency gains mean layoffs or growth? How will AI use be rewarded or penalized?
  • Leaders do not need certainty, but they must share a directional goal.
  • AI does not replace whole jobs yet, but it does replace specific tasks within jobs — leaders need to anticipate and redesign workflows accordingly.
  • Example: Cross-functional software engineering teams working alongside subject matter experts and marketers can “vibe code” projects in days that previously took months.
  • NLW frames the central leadership question as: “Efficiency AI” (do the same with less) vs. “Opportunity AI” (grow into what was previously impossible). He argues organizations defaulting to efficiency AI will be outcompeted by those pursuing opportunity AI.

4. The Crowd: Bottom-Up Discovery and Hidden AI Use

  • Employees with deep domain expertise are best positioned to identify where AI is genuinely useful through trial and error — outsiders cannot replicate this judgment.
  • A persistent problem: official AI tool adoption often caps around 20% of workers, yet 40%+ report using AI at work — because many are hiding their AI use (“secret cyborgs”).
  • Workers hide AI use to avoid being penalized, having tools removed, or having their work perceived as less legitimate.
  • Leadership remedy: provide clear zones of permitted experimentation, biased toward allowing AI use wherever ethically and legally feasible, rather than issuing vague ethics talks or blanket restrictive policies.

5. The Lab: Centralized Experimentation

  • A dedicated internal “Lab” — distinct from abstract strategy or analysis functions — should focus on building things.
  • The Lab should be ambidextrous: exploring future possibilities (months away) while also deploying a steady stream of practical tools and workflows.
  • Lab members should include subject matter experts, technologists, and non-technologists; enthusiastic crowd members often make ideal recruits.
  • Key Lab activities:
    • Rapidly distribute prompts and solutions surfaced by the crowd.
    • Build organization-specific AI benchmarks to track what AI can and cannot do for your particular context.
    • Build prototypes for AI-agent-driven business processes that do not yet fully work — so that when models improve, deployable prototypes are ready immediately.
    • Create provocations: demos and visceral experiences that help people who have not engaged with AI grasp its transformative potential and overcome inertia.

6. Rethinking the Nature of Work

  • Mollick’s conclusion: organizations were built around human intelligence because that was all that existed. AI changes this fundamental constraint.
  • When bottlenecks shift (research from execution to knowing what to research; coding from writing to knowing what to build; content from production to knowing what matters), organizations must restructure around the new bottlenecks.
  • The challenge is not implementing AI as a technology; it is transforming how work gets done while the technology itself keeps evolving.
  • Successful companies build feedback loops between Leadership, Lab, and Crowd that allow them to learn faster than competitors.

7. NLW’s Commentary: Where He Diverges

  • Agents as a paradigm shift: NLW’s primary critique is that Mollick’s framework, while accurate, reflects a 2024 mindset. The emergence of credible AI agents in early-to-mid 2025 has fundamentally shifted organizational thinking from “how do we capture individual efficiency gains?” to “how do agents remake our entire organizational structure?”
  • Speed of change: Statistics cited in the piece (e.g., 20% official tool adoption cap) are already out of date given the KPMG data showing 58% daily usage. The pace of change is faster than the piece implies.
  • Upskilling is being deprioritized prematurely: When agents became plausible, employee upskilling was “kicked back down” to a secondary priority; NLW sees this as an overcorrection, and believes upskilling must evolve from prompting skills toward agent management skills.
  • The playbook question: NLW partially disagrees with Mollick’s claim that no external playbooks exist. He reframes it: there is no playbook with answers, but there are emerging playbooks that help organizations ask the right questions.
  • Urgency level: NLW argues organizations still nudging slowly are not merely behind — they are dangerously behind — because some competitors are already pursuing complete organizational reorganization, not pilots.

Key Concepts

  • Leadership, Lab, and Crowd: Mollick’s three-part framework for AI organizational transformation — executive vision, centralized experimentation, and bottom-up employee innovation working in concert.
  • Secret cyborgs: Employees who use AI tools privately to boost their own performance but hide that use from employers to avoid penalties or stigma.
  • Organizational innovation: The redesign of incentives, processes, workflows, and structures to convert individual AI gains into firm-level performance improvements.
  • Ambidextrous lab: An internal AI unit that simultaneously explores future capabilities and exploits current ones by releasing practical tools continuously.
  • AI benchmarks (organizational): Custom evaluation frameworks built by a company to measure how well AI performs on tasks specific to that organization’s context.
  • Efficiency AI vs. Opportunity AI: NLW’s distinction between organizations that use AI primarily to cut costs (do the same with less) vs. those that use AI to pursue growth and previously impossible capabilities.
  • Provocations: Demos or hands-on experiences designed to make AI’s transformative potential viscerally real for skeptical or disengaged employees.
  • Vibe coding / vibe work: Informal term for rapid, AI-assisted software or project development by cross-functional teams without heavy traditional engineering coordination.
  • Agent readiness: Organizational preparedness to design, deploy, and manage AI agents that can autonomously execute business processes.
  • Feedback loops (Leadership–Lab–Crowd): Continuous learning cycles connecting executive strategy, experimental prototyping, and employee discovery to accelerate organizational AI adoption.

Summary

Ethan Mollick’s “Making AI Work” identifies a central paradox in enterprise AI adoption: workers are experiencing large individual productivity gains, yet organizations are capturing little of that value. His explanation is that individual gains do not automatically become organizational gains — doing so requires deliberate organizational innovation through three interlocking mechanisms: Leadership (setting a vivid, credible vision and answering workers’ core questions about their futures), the Lab (a dedicated internal unit that builds, tests, benchmarks, and distributes AI tools and workflows), and the Crowd (the employees who discover effective AI use through domain expertise and trial and error, but who currently often hide that use due to fear of penalties). NLW reads and endorses this framework as broadly correct and actionable, placing organizations that follow it in the top quartile of performers, but argues it is already somewhat behind the moment: the emergence of AI agents in 2025 has triggered a more radical rethinking of organizational structure than the individual-productivity-gain framing suggests, the pace of adoption is accelerating faster than Mollick’s statistics reflect, and organizations still moving cautiously are not merely slow — they are at genuine competitive risk from peers who are pursuing wholesale organizational reinvention rather than incremental pilots.