Write-Up: Co-Existence

Okay, I will interpret the "Part 1" section of the Scottish Mortgage report you provided as a third-party independent analyst.

---

At a Glance

Wharton professor Ethan Mollick and Scottish Mortgage deputy manager Lawrence Burns discuss the uneven nature of AI development and its investment implications. Ethan Mollick introduces the concept of the "jagged frontier," arguing that AI progresses extremely fast but unevenly—excelling in some areas while having severe shortcomings in others, which requires companies to create true competitive advantage through a unique process of adoption and integration.

AI's "Jagged Frontier": Breakthroughs and Bottlenecks Coexist

Ethan Mollick first highlights a core feature of the AI field: the unevenness of progress. He calls this the "jagged frontier," noting that AI "is terrible at things you'd expect it to be good at, and sometimes excellent at things you didn't expect."

This judgment is based on specific observations:

• Unexpected Weaknesses: AI still has trouble generating coherent text. Mollick points out that AI can write decent short pieces, but longer works tend to "lose the plot" and become confused about the target audience. Meanwhile, in areas requiring diversity, AI underperforms humans. "With the right prompt, AI can produce more diverse ideas, but its diversity is still less than that of a group of humans," Mollick emphasizes. This suggests that in investment domains requiring diverse perspectives and strategic judgment, AI is currently not a substitute for human diversity.
• Practical Advice: Addressing this weakness, Mollick suggests using "councils of AIs" to aid decision-making—i.e., using multiple models simultaneously or having the same model play different roles—to obtain a diversity of perspectives closer to that of a human group. However, he acknowledges that for reasons not yet fully understood, different AI models still exhibit high similarity.

Corporate Differentiation: The Integration Process as the New Moat

When asked why performance varies across companies when they all have access to the same frontier models at roughly similar prices, Mollick points directly to the process of adoption and integration.

He draws different conclusions for individuals versus companies:

• For Individuals: Early research with Boston Consulting Group (BCG) suggested AI is a "leveller," but more recent studies show that experts may derive greater performance gains from AI because they know how to ask better questions and identify valuable answers.
• For Companies: The differences may be even more pronounced. Mollick asserts, "The key lies in the process—how you integrate AI into your systems, how it collaborates with humans, and who makes the decisions." This means that those who better integrate AI with their own expertise, leadership support, and experimental culture can build a sustained competitive advantage, not one based solely on the model itself.

To this end, Mollick offers several practical tests that are highly valuable for evaluating a company's AI strategy implementation:

1. "What have you stopped doing as a result of AI?" If the answer is "nothing," then "something's wrong."

2. "Is the company doing anything impossible thanks to AI?" "You should have at least one effort that you believe could disrupt your entire industry," Mollick states, including projects that may not yet be mature under current technological conditions.

3. The "leadership, lab, and crowd" elements: This means having executive support, a dedicated team to address technical/organizational issues, and empowering all employees to discover AI use cases.

Shifting Work Bottlenecks: From Writing Code to Defining Problems

Mollick believes that AI has already reshaped and is continuing to reshape roles and work bottlenecks within organizations. A powerful example is programming. "Now, the best engineers I know rarely write code themselves," Mollick says. Their core value has shifted to directing AI agents by specifying requirements and designing tests. The work bottleneck is shifting from "writing clean code" to "defining requirements" and "designing tests."

This shift imposes new demands on organizational capabilities. Mollick invokes the concept of "dynamic capability" from business research, arguing that organizations that can continuously adjust their capabilities in response to environmental changes will adapt better. But he emphasizes that adaptation is only half the task. Since "nobody knows anything for sure," companies must experiment to help invent this future, not merely react to it.

Mentioned Positions

This section does not materially discuss any individual stocks with specific position changes or in-depth business analysis. Mollick's views are primarily focused on macro trends and industry-level applications.

Judgments Worth Remembering

1. "Jagged Frontier"

• Speaker: Ethan Mollick
• Judgment: AI progress is not uniform; it excels where you least expect it and fails where you least expect it. Mollick calls this the "jagged frontier" and notes, "AI is inconsistent in writing… long-form pieces easily lose the plot."

2. "What have you stopped doing because of AI?"

• Speaker: Ethan Mollick
• Judgment: This is the key litmus test for whether a company has truly embraced AI. If the answer is nothing, "something's wrong." This implies that genuine transformation necessarily involves abandoning old processes.

3. AI Leveller Effect Reversed: Expert Advantage Will Widen

• Speaker: Ethan Mollick
• Judgment: Early studies suggested AI is a "leveller," but recent research finds that experts, by virtue of their knowledge, ask better questions and thus gain more from AI. This means AI adoption may widen rather than narrow the performance gap between different skill levels.

4. The Best Engineers No Longer Write Code

• Speaker: Ethan Mollick
• Judgment: "Now, the best engineers I know rarely write code themselves." Their value has shifted to directing AI agents by specifying requirements and designing tests, indicating that the work bottleneck is moving from technical execution to requirement definition and validation.

5. Companies Diverge Due to "Adoption Process," Not "The Model Itself"

• Speaker: Ethan Mollick
• Judgment: Given that nearly all companies can access the same frontier models, differences will depend on "how AI is integrated into systems, how it collaborates with humans, and who makes decisions." This is Mollick's strong rebuttal to the view that "AI cannot constitute a moat."

6. Historical Analogies May Not Repeat: AI Could Differ from the Railroad Bubble

• Speaker: Ethan Mollick
• Judgment: Responding to Burns's analogy of railroad/fiber-optic bubbles, Mollick warns, "I don't think that's a given," because current AI systems already demonstrate practical uses and companies are reaping returns. This judgment challenges the widespread expectation of a bubble burst.

7. The Possibility That "AI Has Had No Real Impact in the Past Five Years" Has Disappeared

• Speaker: Ethan Mollick
• Judgment: This represents a significant inflection point in confidence. Mollick believes AI's influence could be so profound that "this might look like the Second Industrial Revolution (Industrial Revolution II)." This is an extreme, transformative judgment.