Why Everyone Is Wrong About AI (Including You) | Benedict Evans

Sep 2, 2025
Overview

Technology analyst Benedict Evans discusses AI as the biggest platform shift since the iPhone, not an existential threat. He delves into why incumbents struggle with new technologies, the importance of creating your own knowledge, and why most people still don't 'get' consumer AI, suggesting it needs to be integrated as a feature rather than a standalone product.

At a Glance
18 Insights
1h 13m Duration
12 Topics
5 Concepts

Deep Dive Analysis

AI as a Platform Shift: Not Electricity

Understanding Past Technology Platform Shifts

Incumbents' Challenges During Platform Shifts

Google's Vulnerability and AI Profit Margins

Unasked Questions: AI Data and Differentiation

The Illusion of AI Autonomy and Threats

Benedict Evans' Personal AI Use Cases

Thinking by Writing and AI's Impact on Originality

Advice for Students in the Age of AI

Analyzing Major Tech Companies in the AI Race

The Future of Autonomous Cars and EVs

Defining Personal Success and Curiosity

Platform Shift

A fundamental change in how technology works, leading to new products and services built around it for 10-15 years, similar to the internet or mobile internet. Incumbents often try to absorb it as a feature, but it can unbundle existing companies and create new ones.

Electronic Politeness

A historical example from automatic elevators, where a novel technology (like infrared sensors stopping doors) becomes so integrated and commonplace that its specific features are forgotten and it's simply 'the lift.' This illustrates how new technologies eventually become part of everyday life.

Pricing as Information System

A concept from Hayek, where prices in a market economy act as signals, informing people about what is wanted and demanded. Disrupting this signal through regulation or central planning can lead to unintended consequences and a lack of understanding of market dynamics.

Learning Loop

A theory suggesting that learning occurs through experience, reflection on that experience, and creating a 'compression' or takeaway. Consuming other people's compressions without doing the raw work can create an 'illusion of knowledge.'

AI Slop

A term referring to the vast amount of low-quality, mass-produced content generated by AI, often without genuine insight or originality. It highlights the challenge of distinguishing valuable content in an era of easy AI generation.

?
What is Benedict Evans' most controversial take on AI?

Benedict Evans believes AI is the biggest platform shift since the iPhone, but not more transformative than computing or electricity. He sees it as another significant technological change that will drive innovation for the next 10-15 years.

?
How do incumbents typically react to platform shifts?

Incumbents usually try to absorb new technologies as a feature, using them to automate existing processes. However, true platform shifts often unbundle incumbents and create entirely new companies and industries.

?
Is proprietary data a significant advantage for incumbents in the AI race?

No, because large language models (LLMs) require such an enormous amount of generalized text data that existing proprietary data from companies like Google or Meta isn't enough to create a fundamental difference. The necessary data is largely equally available to anyone.

?
How far are we from AI making itself better without human intervention?

We are not at a stage where AI can autonomously improve itself, and it's unclear when that might happen. Many claims about AI's autonomous capabilities are often misinterpretations of story-generating machines responding to specific prompts.

?
Should AI be regulated as a general technology?

Benedict Evans argues that regulating 'AI as AI' is the wrong level of abstraction, similar to regulating 'databases' or 'spreadsheets.' Regulation should focus on specific applications and their trade-offs, understanding that policies have costs and consequences.

?
Why do many people still 'not get' AI or use it infrequently?

While AI adoption appears fast due to its accessibility (free, web-based), many people look at consumer-facing LLM chatbots and don't understand how to integrate them into their daily tasks. They may struggle to form new habits or identify personal use cases beyond obvious applications like coding or brainstorming.

?
Is AI better for qualitative or quantitative analysis?

Currently, AI has zero value for quantitative analysis where numbers need to be precisely right, as it tends to produce 'roughly right' answers. Its strength lies more in qualitative tasks where a close-enough answer or idea generation is sufficient.

?
Can AI create truly original and good content?

It's a puzzling question whether AI can make things that are both different from its training data and genuinely good, as it lacks an inherent feedback loop for 'original but good' like AlphaGo had for game moves. The concept of originality itself might be a matter of perspective, depending on how closely one zooms in or out on patterns.

?
What is the 'learning how to think' skill Benedict Evans emphasizes for students?

For Evans, learning how to think means developing skills like asking the next critical question, breaking down complex problems, synthesizing vast amounts of information, evaluating credibility, and effectively explaining concepts. It's about intellectual agility rather than memorizing specific facts.

?
What is the 'bear case' for Apple in the age of AI?

The concern for Apple is that while people will continue to buy iPhones for their hardware advantages, the core experience and value capture might shift to cloud-based AI models from other companies, making the iPhone primarily a 'glowing rectangle' for running someone else's AI, similar to how Microsoft's Windows PCs became platforms for web content.

1. Cultivate Continuous Curiosity

Actively cultivate curiosity by consistently “looking for the next question” in any field or situation. This practice fosters continuous learning and discovery throughout your life.

2. Practice Critical Thinking & Synthesis

Develop “learning how to think” by practicing critical inquiry: ask the next question, break apart complex information, synthesize diverse sources, discern true meaning, evaluate credibility, and effectively explain concepts.

3. Create Your Own Knowledge Compressions

To avoid an “illusion of knowledge,” actively engage in a “learning loop” by having experiences, reflecting on them, and creating your own “compressions” or takeaways. This process ensures deeper understanding beyond just consuming others’ summaries.

4. Benchmark Ideas Against AI

Before sharing an insight, ask yourself, “Is this what ChatGPT would have said?” If the answer is yes, reconsider publishing, as this practice helps ensure your contributions are original, add value, and push beyond obvious answers.

5. Deliver Unique Insight Beyond AI

As LLMs commoditize common knowledge, prioritize providing unique and valuable insights that go beyond what AI can generate. This ensures your contributions remain relevant and valuable in a changing information landscape.

6. Employ Dual Thinking Modes

Utilize two distinct thinking modes: a discursive, free-associative one for broad exploration, and a focused, analytical one for breaking down complex topics into core components. This dual approach helps both generate ideas and achieve clear understanding.

7. Prioritize Relevance in Analysis

When analyzing data or presenting insights, consistently ask “who cares?” and “what actually matters?” This ensures your work focuses on relevant questions and delivers high-value information.

8. Understand Systemic Market Effects

Recognize that a market economy is a complex system where actions in one area will inevitably cause reactions elsewhere. Understand these systemic consequences to make informed decisions and achieve desired outcomes.

9. Recognize Policy Trade-offs

When evaluating policies or making strategic decisions, remember that “to govern is to choose,” meaning every choice has trade-offs and costs. Understand these consequences to make informed decisions.

10. Unbundle Incumbents with New Tech

When a new fundamental technology emerges, anticipate that incumbents will try to absorb it as a feature. Seek opportunities to “unbundle” existing companies and create new offerings, as this is where significant disruption and value creation often occur.

11. Beware Commoditization Trap

Be cautious when shifting from a high-margin product with unique IP to a low-margin, commoditized market. Without differentiation, you risk competing against an entire industry with little advantage, as Kodak experienced.

12. Target High-Friction AI Use Cases

Focus on identifying highly specific, high-friction, and time-consuming administrative tasks as prime targets for AI automation. Solving these “pain points” can deliver significant practical value, even if the technology is still evolving.

13. Startup Evaluation Framework

When evaluating a startup, focus on three key questions: “Could it work?”, “If it did work, what would it be?”, and “Could those people make it work?”. This framework helps assess potential and the team’s ability to execute.

14. Calibrate Through Broad Exposure

Develop “calibration” and pattern recognition by exposing yourself to a wide range of examples, such as many startups or diverse art. This helps you discern quality, understand what works, and recognize underlying patterns.

15. Seek External Context & Perspective

Recognize the insular nature of industry hubs and actively seek external context and diverse perspectives. This helps you avoid a narrow viewpoint and gain a broader understanding beyond your immediate professional circle.

16. Explore to Create Options

Actively explore diverse skills, subjects, and experiences, as you don’t know what you’ll excel at. This approach helps create future options and discover your true strengths.

17. View AI as Platform Shift

Adopt a realistic perspective on AI’s impact, viewing it as a significant but not apocalyptic platform shift, similar to the iPhone. This mindset can help you make more grounded decisions about its future influence on employment, economy, and intellectual property.

18. Embrace Unpredictable Change

Recognize that history teaches us only that “something will happen,” not what specifically. This encourages an adaptable mindset, preparing you for inevitable but unpredictable future shifts.

My sort of base case is to say this is kind of another platform shift and all the new stuff will be built around this for the next 10 or 15 years. And then there'll be something else.

Benedict Evans

The very high level threat to Google is that you have this moment of discontinuity in which everybody resets their priors that we consider as their defaults and so it's no longer just the default that you go and use Google.

Benedict Evans

Policies have trade-offs. To govern is to choose. You're making a choice when you do that, and you're choosing that has costs.

Benedict Evans

I always think the Kodak example here is kind of interesting. Because, you know, like, it's like the cliche that people say, oh, Kodak had digital cameras and they didn't get it or they ignored it or they didn't want to do it because it would destroy their business.

Benedict Evans

If the thing gets infinitely close to reasoning without ever actually reasoning, does it matter? Like at a certain point, the thing, if the thing is always right, if the thing is only wrong once in a billion years, does it matter that it's not, that it's not always right?

Benedict Evans

Is this what chat GPT would have said? And if the answer is, this is what chat GPT would have said, then I didn't publish it. Not because people can get it from chat GPT, but because anyone would have said that.

Benedict Evans

What I learned studying history at Cambridge was how to ask what the next question is, how to break this apart, how to read 100 books or 50 books in a week and find the bits that you need, how to synthesize lots of information, how to ask, well, what does that actually mean as opposed to what it looks like it means?

Benedict Evans
~$220 billion
CapEx spending by Google, Microsoft, AWS, and Meta (last year) For investment in AI infrastructure.
over $300 billion
CapEx spending by Google, Microsoft, AWS, and Meta (this year) Projected, almost tripled from a couple of years ago.
49% for $15 billion
Meta's acquisition of Scale.ai stake For AI development.
multiple tens of billions
Valuation of recent OpenAI spin-out labs For pre-product, pre-revenue labs.
~10% (± 3-4%)
Percentage of people using consumer-facing LLM chatbots daily Based on US survey data from late last year/early this year.
15-20%
Percentage of people using consumer-facing LLM chatbots weekly Based on US survey data.
20-30%
Percentage of people using consumer-facing LLM chatbots monthly or every two months Based on US survey data.
$50-60 billion
Amazon's ad revenue (last year) A significant source of Amazon's money, alongside AWS.