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

May 31, 2026 Episode Page ↗
Overview

Benedict Evans, former Andreessen Horowitz partner, discusses AI's transformative impact, comparing it to the internet's early days. He explores job changes, value accrual in the AI stack, and the growing anti-AI sentiment, offering a rational, comforting perspective on preparing for the future.

At a Glance
5 Insights
1h 19m Duration
16 Topics
5 Concepts

Deep Dive Analysis

AI's Impact: Comparison to Internet and Mobile

The '1997 Moment' for AI Adoption

AI's Impact on Software Development

Unexpected Boom in Professional Services for AI

Distinguishing Between 'Task' and 'Job' in Automation

Historical Perspective on Job Automation and Creation

The Speed and Compounding Effect of AI Adoption

Critique of AI 'Doomers' and Enterprise Adoption Cycles

Shifting Definitions of AGI and Superintelligence

Value Accrual: Models vs. Applications

Distribution as the Ultimate Moat in AI

The Anti-AI Sentiment and Backlash

Raising Kids in an AI Future

The Question Nobody's Asking About AI

Advice for Success in the AI Future

Personal AI Use Cases

1997 Moment for AI

This analogy suggests that AI is currently in an early, exciting, and uncertain phase, similar to the internet in 1997. Most applications haven't been built, how things will work is unclear, and adoption is widely distributed, with many still not fully engaged.

Task vs. Job

This framework differentiates between specific, automatable components of work ('tasks') and the broader, more complex roles ('jobs') that often involve human judgment, politics, and strategic thinking. While AI can automate tasks, the full job may persist or evolve.

Jevons Paradox (Price Elasticity)

This economic principle, applied to technology, describes what happens when something becomes cheaper to do. Instead of doing the same amount for less money, people often do more of it for the same or even more money, leading to increased overall activity rather than reduced demand or labor.

AGI (Artificial General Intelligence)

AGI refers to human-level intelligence in machines, but its definition is constantly shifting. As AI capabilities advance, what was once considered AGI (like image recognition) is often redefined as 'just software,' making it a moving target for what constitutes true general intelligence.

Lump of Labor Fallacy

This fallacy assumes there is a fixed amount of work to be done, implying that automation will inevitably lead to permanent job losses. Historically, new technologies automate old jobs but also create new, often better, jobs and unlock greater prosperity.

?
How significant is AI compared to past technological shifts?

AI is considered as significant as the internet or mobile revolutions, but not necessarily more so. It represents a fundamental change, but its long-term impact and specific applications are still largely unknown, akin to the internet in 1997.

?
Will AI lead to a 'jobpocalypse' and mass unemployment?

Historically, new technologies automate existing jobs but also create new ones, leading to overall economic growth and new forms of employment. While there will be frictional pain and dislocation, the idea of widespread, permanent job loss due to AI is often an oversimplification of how the economy adapts.

?
Why are AI labs investing in professional services and consulting?

Companies need help integrating AI into their complex internal workflows, a process that requires significant project management, system integration, and training. AI labs are investing in professional services because organizations lack the internal capacity to implement these changes, and consultants fill this gap.

?
Where will the most value accrue in the AI stack?

It's argued that foundational AI models may become commoditized over time, similar to utilities or cloud infrastructure, leading to squeezed margins. The greater value is likely to accrue in the application layer, where companies build specific products and features on top of these models, leveraging distribution and unique use cases.

?
Is the growing anti-AI sentiment justified?

Anti-AI sentiment is a 'fuzzy mess' of various concerns, some tangible (like local electricity bills) and some less so (like exaggerated water consumption claims). While some concerns are valid (e.g., deepfakes), others are based on misinformation or a misunderstanding of how technology impacts society, similar to past backlashes against social media or databases.

?
How should one prepare for the future with AI?

The best approach is to actively engage with AI, understand its capabilities and limitations, and learn how to use it. Avoiding it or being dismissive will not be helpful, as the technology will continue to transform industries and job roles.

1. Actively Engage with AI

Do not ignore or reject AI, as this provides a false sense of moral superiority but is unhelpful. Instead, dive into AI, understand what you can do with it, and learn how it changes things to remain a valuable contributor in the evolving landscape.

2. Focus on the ‘Job,’ Not Just the ‘Task’

When considering AI’s impact, differentiate between a ’task’ (which can be automated) and the broader ‘job’ (which often involves complex human elements like politics, customer understanding, or strategic thinking). AI may automate tasks, but the core job often remains, requiring human insight.

3. Understand AI’s Limitations

Recognize that current AI models still hallucinate and are not always reliable for precise information retrieval. Use AI for tasks it excels at, like proofreading or image generation, but be aware of its weaknesses, especially when accuracy is critical.

4. Question Model Pricing Power

Do not assume that foundational AI model companies will maintain radical pricing power indefinitely. Consider that models may become commoditized, leading to value accruing further up the application stack, similar to how cloud infrastructure operates.

5. Prepare for Unpredictable Change

Accept that the future impact of AI is highly uncertain, much like the early days of the internet or mobile. Avoid rigid predictions and instead cultivate adaptability, as new technologies often create unforeseen jobs and transform industries in unexpected ways.

My most controversial opinion is that I think that AI is as big a deal as the internet or mobile, and only as big a deal as the internet or mobile.

Benedict Evans

Every time we have a new technology, it automates away a bunch of jobs, and then that automation unlocks a bunch of new jobs. And you don't know the new job because it doesn't exist yet.

Benedict Evans

Don't stick your head in the sand and say, I hate all of this stuff. That gives you a great feeling of moral superiority, and you can go on Blue Sky and shout at everybody about how evil AI is. Like, great, I'm happy for you. But that's not going to help.

Benedict Evans

AI is whatever machines can't do yet. Because once machines can do it, people say, well, that's just software.

Larry Tesla (quoted by Benedict Evans)

The model is just like the dumb thing underneath, the funny way of putting it, the dumb thing underneath that powers the feature. The model is the commodity that powers different decisions about what the feature should be and what different distribution.

Benedict Evans
0.017%
US data center water consumption Percentage of total U.S. water consumption, estimated by Livermore Lab at the end of 2024.
1,500 to 2,000 times
Mobile data consumption increase since 2010 Global increase in mobile data consumption, demonstrating exponential growth.
Approximately $1 trillion
Global mobile industry annual revenue Revenue generated by the global mobile industry.
Approximately $200 billion
Global mobile industry annual CapEx Capital expenditure by the global mobile industry, representing 15-20% of revenue.