A rational conversation on where AI is actually going | Benedict Evans

Key Concepts

• Platform Shifts: The idea that AI is a fundamental technological shift comparable to the internet or mobile, characterized by a "1997-like" phase of early adoption and uncertainty.
• Lump of Labor Fallacy: The economic misconception that there is a fixed amount of work to be done; in reality, automation creates new, unforeseen jobs and increases productivity.
• Jagged Frontier: The unpredictable nature of where AI excels and where it fails, making it difficult to determine which specific tasks or roles will be automated.
• Price Elasticity: The concept that as a task becomes cheaper to perform, demand for it increases, often leading to more work rather than less.
• Commoditization of Models: The perspective that foundation models (like those from OpenAI or Anthropic) may become undifferentiated commodities, shifting value to the application layer.
• Distribution Moats: The argument that in a world where software is easy to build, the ability to distribute products to users becomes the primary competitive advantage.

---

1. The Nature of the AI Transformation

Benedict Evans argues that AI is a "big deal" on par with the internet or mobile, but warns against the "doomer" narrative of an immediate job apocalypse. He compares the current state of AI to 1997: it is exciting, but most infrastructure is still being built, and we do not yet know how it will ultimately function. He emphasizes that we are in a period of "radical uncertainty" where the most productive stance is to experiment rather than predict.

2. The "Job Apocalypse" and Economic Reality

Evans challenges the fear that AI will simply replace humans. He points to historical precedents—such as the introduction of spreadsheets in the 1970s—which did not eliminate accountants but rather changed the nature of their work and increased their numbers.

• Key Argument: Technology automates tasks, not necessarily jobs.
• Evidence: Despite decades of automation, the number of accountants and software engineers has continued to grow.
• The "Hard Part" of a Job: Evans notes that companies hire consultants (like McKinsey or BCG) not just for a slide deck (a task), but for organizational navigation, political alignment, and deep problem-solving—things AI cannot currently replicate.

3. The Value Chain: Models vs. Applications

A central thesis of the discussion is the potential commoditization of foundation models.

• The Telco Analogy: Evans compares foundation model labs to mobile network operators. While they provide the essential infrastructure, the "cool stuff" (the value) is created by developers and companies building on top of that infrastructure.
• Pricing Power: He questions whether model labs will maintain long-term pricing power. If models become commodities, the real value will accrue to the application layer—the companies that solve specific, complex problems for users.

4. Distribution as a Moat

As AI makes software development cheaper and faster, the market is becoming saturated with "wrappers." Evans argues that:

• Distribution is King: When the underlying product is a commodity, the company that can reach the most users (distribution) wins.
• Incumbent Advantage: Incumbents like Google, Meta, and Apple have a massive advantage because they already own the distribution channels (e.g., Android, iOS, social platforms).

5. Anti-AI Sentiment and Societal Impact

Evans characterizes the current anti-AI sentiment as a "big, fuzzy mess" of legitimate concerns and misunderstandings.

• Data Centers: He dismisses the "water usage" panic as a local planning issue rather than a systemic environmental crisis, noting that data centers account for a tiny fraction of total water consumption.
• Social Harms: He acknowledges that AI, like the internet, will be used for bad (e.g., deepfakes, harassment), but argues this is a constant feature of technological progress that requires management rather than total rejection.

6. Actionable Advice for the Future

Evans offers clear, pragmatic advice for those worried about their careers:

• Avoid "Head in the Sand": Refusing to engage with AI provides a sense of moral superiority but offers no professional utility.
• Submerge Yourself: Dive into the technology, understand its current capabilities, and learn how to use it to become a more effective worker.
• Focus on Skills: Instead of trying to predict the "perfect" career, focus on developing a combination of skills that make you valuable, as the specific job titles of the future do not yet exist.

Synthesis/Conclusion

The main takeaway is that while AI is a transformative force, it is not a magical replacement for human judgment. We are currently in a phase of "vibes forecasting" where we lack a theory of intelligence or a clear roadmap for the future. The most successful individuals and companies will be those who treat AI as a tool to be integrated into workflows, focus on solving real-world problems rather than just "doing the old thing but more," and maintain the humility to admit that the future is fundamentally unpredictable.