May 31, 2026 · 1h 20m
Technology analyst Benedict Evans compares the AI boom to the 1997 internet
A rational conversation on where AI is actually going | Benedict Evans
Understanding the historical patterns of technology adoption helps businesses cut through current AI hype and identify where long-term economic value will actually accumulate.
- 1AI integration will follow historical patterns of slow enterprise adoption and professional services growth.
- 2Distribution remains the ultimate competitive moat as underlying AI infrastructure faces rapid commoditization.
- 3The technology is transitioning from performing simple tasks to managing entire job functions.
Don't miss
Benedict Evans explains why distribution, not technology, remains the ultimate moat for software companies in the AI era.
The brief
Technology analyst Benedict Evans argues that today's artificial intelligence boom closely mirrors the early days of the internet in 1997, marking the start of a long, predictable cycle of adoption.
While hype suggests immediate disruption, Evans points out that integrating general-purpose technologies into the broader economy always requires slow, systematic changes in business workflow.
The real value in the AI era will not belong to the underlying infrastructure, which faces rapid commoditization, but to the companies that control distribution and customer relationships.
The next phase of AI development will shift from executing simple, isolated tasks to managing entire job functions, reshaping professional services and corporate operations.
What was said on this episode
29 statements · 9 positive · 11 negative · 4 mixed · 5 neutral
AI will be as consequential as the Internet or mobile, but not larger.
“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.”
Listen at 0:01
New technologies repeatedly automate existing jobs and unlock new ones.
“We've had that process over and over again.”
Listen at 0:15
It is not possible to predict exactly which work technology will automate.
“You can't predict which things are going to be exposed.”
Listen at 0:31
People should experiment with AI to understand its practical uses.
“What helps is you diving into this and coming out, understanding what you can do with.”
Listen at 1:07
AI is currently at an exciting early stage comparable to the Internet in 1997.
“Like, it's very exciting.”
Listen at 3:26
Among teenagers, roughly 15–20% use AI daily and another 20% weekly.
“It's still like kind of 15, 20% of people are daily active users and another 20% are weekly active users.”
Listen at 8:58
Redesigning enterprise workflows for AI requires a dedicated multi-person project.
“That's a project. That's a project that needs like five or 10 people to sit down and spend a month or two working it out”
Listen at 11:30
Claude Code can produce code and features but cannot decide which product to build.
“Claude, co can write you the code, but what code do you want? It can make you the features. Sure, but what features do you want?”
Listen at 15:19
After technological transitions, societies ultimately become wealthier despite disruption.
“But when you come through on the other side, we're all richer”
Listen at 19:49
Enterprise software and related jobs will change radically over several years.
“Yes, that whole estate will look radically different and all those jobs will have changed.”
Listen at 22:05
The world has approximately 5.5–6 billion mobile smartphones.
“there's about 5 and a half, 6 billion mobile smartphones in the world”
Listen at 31:06
Mobile telecommunications is a low-growth, low-margin commodity utility.
“because it's an ex growth, low margin commodity utility”
Listen at 34:15
AI model companies currently look like undifferentiated commodity infrastructure providers.
“they're undifferentiated commodity infrastructure providers.”
Listen at 40:20
U.S. data centers consume approximately 0.017% of national water use.
“it came out at about 0.017% of US water consumption.”
Listen at 49:21
U.S. data-center water consumption is about 0.017% of total U.S. water consumption.
“it came out at about 0.017% of US water consumption”
Listen at 49:21
U.S. data-center energy use may grow by one percentage point annually for five years.
“data centers are what, like 5% of U.S. energy and might grow 1% a year for the next, next five years, 1 percentage point a year”
Listen at 49:39
There is no clear consensus that AI is currently affecting employment.
“there's no clear consensus that we're seeing an impact on jobs”
Listen at 51:19
AI-related job-market conditions will settle substantially within about five years, unpredictably.
“staff will have settled down a lot by then in probably unpredictable ways”
Listen at 53:51
AI now enables teenagers to create and distribute deepfake pornography at large scale.
“And now they can.”
Listen at 56:32
AI will reconnect harmful people, bad instincts, and societal problems as social media did.
“And so that will happen again with AI.”
Listen at 57:03
AI will automate away some jobs that mainly consist of automatable tasks.
“some jobs where no, that is just a task and that job gets automated away”
Listen at 1:00:09
Current methods cannot reliably quantify what percentage of a profession AI can automate.
“You can't describe a profession like that. Well, at any rate, we can't.”
Listen at 1:03:31
AI can generate training routines and monitor whether users perform exercises correctly.
“I ask an AI to build me a training routine and watch me and tell me if I'm doing it right.”
Listen at 1:04:38
Workers should immerse themselves in AI to understand its practical capabilities and career implications.
“What helps is you diving into this, completely submerging yourself in it and coming out, understanding what you can do with it”
Listen at 1:07:47
AI currently performs precise information retrieval relatively poorly.
“the kind of stuff that I would want a machine to do for me is the stuff that AI kind of can't do for me very, very, very well at the moment”
Listen at 1:09:20
Generative AI can work very well for visualizing apartment redecorations.
“I used it redecorating my apartment. That worked fantastic.”
Listen at 1:09:30
High marginal costs make it difficult for consumer AI apps to combine free scale with a viable revenue model.
“because of marginal cost more than anything else, you can't make it free and get 50 million users and then have a revenue model”
Listen at 1:15:22
AI language models still produce hallucinations.
“They still hallucinate.”
Listen at 1:18:54
AI language models remain useful despite hallucinations.
“But that doesn't mean they're not useful.”
Listen at 1:19:02
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.