LLenny's Podcast
← All episodes
Benedict Evans31 May 2026

A rational conversation on where AI is actually going

6Frameworks
13Insights

Frameworks in this episode

Marketing4 steps

Distribution Beats a Commodity Product

When products in a category are basically interchangeable, an adequate product with superior distribution wins

Read →
Self-Mastery4 steps

Dive-In Career Strategy for a Platform Shift

Facing a disruptive technology, submerge yourself in it and aim to hold at least two of three career pillars

Read →
Mindset4 steps

Presume Radical Uncertainty (The 1997 Lens)

Treat an emerging platform as if it were 1997 for the internet: assume most of it doesn't work and you can't yet name the winners

Read →
Innovation4 steps

Price-Elasticity Three-Response Model

When a technology makes something cheaper, work out which of three demand responses your market will take

Read →
Strategy4 steps

Task vs. Job Automation Test

Before predicting a role's automation, ask whether the automatable task IS the job or just one piece of it

Read →
Strategy5 steps

Value-Up-The-Stack Test

To find where the money accrues in a platform shift, test the infrastructure layer for network effects, differentiation, and pricing power

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster20:30

Why the Doomers Are Wrong: No One Fires Everyone in Two Weeks

Evans dismantles the doomer fantasy that every big company will buy ChatGPT tomorrow and fire all their staff two weeks later. Enterprise reality is an 18-month-plus sales cycle, longer than a startup's funding rounds, especially in aerospace or healthcare. Companies won't rip out SAP overnight; the whole estate may look radically different in three to ten years, but it takes time sector by sector for people to work out what they can do with the technology.

  • The idea that companies buy ChatGPT and fire everyone in two weeks reflects total ignorance of how business works
  • Enterprise software sales cycles are ~18 months if you're lucky, longer than venture funding rounds
  • Nobody tears out SAP and replaces it overnight; change is sector by sector over years
  • SaaS companies founded post-ChatGPT often could have been built 15 years earlier; the delay was people realizing the problem and solution

“you talk to these doomers on Twitter and they would like act like you know every big company is going to buy chat EBT tomorrow…”

Benedict Evans · 21:00

“enterprise software sales cycle is like 18 months if you're lucky”

Benedict Evans · 21:30
#enterprise#ai-adoption#doomers#jobs

Hot Take· 2

Hot Take02:30

AI Is Only as Big a Deal as the Internet or Mobile

Benedict Evans frames his most controversial view: AI is as big a deal as the internet or mobile, and only that big, pushing back on people who insist it's the industrial revolution. His deeper point is that we're at a '1997 moment' where most of what AI will do hasn't been built yet and nobody knows how it will work. He argues the productive conversation isn't whether it's 20% or 100% bigger than the internet, but accepting that a fundamental shift is underway whose shape is unknown.

  • AI is as big as the internet or mobile, not bigger like the industrial revolution
  • We're roughly in the equivalent of 1997: exciting, but most stuff doesn't work yet
  • Most of what people will do with AI hasn't been built and it's unclear how it will work
  • Adoption is spread very wide, from Mac-mini cluster owners to people using it every week or two
  • His own 80-slide presentation essentially argues 'we don't know'

“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…”

Benedict Evans · 00:00

“it's like we're in 1997. Like it's very exciting. Most stuff kind of doesn't work yet.”

Benedict Evans · 03:00
#ai#technology-cycles#forecasting#internet
Hot Take32:30

Foundation Models Are Commodities and the Value Moves Up the Stack

Evans argues the foundation-model layer looks like a low-margin commodity utility, not a Windows-style toll booth. He mocks Sam Altman's line about selling intelligence on a meter like electricity by pointing out utilities have terrible margins. His telecoms analogy is damning: the global mobile industry is objectively amazing infrastructure whose stocks went nowhere for 25 years because it's low-margin commodity, while all the cool stuff got built further up the stack.

  • Selling AI 'on a meter like water or electricity' means accepting utility-industry margins
  • Models show no network effects, so competition and price pressure should persist indefinitely
  • Telecom stocks went nowhere for 25 years despite exponential data growth and huge sophistication
  • It may end up like AWS (customers don't care which cloud) rather than like Windows
  • If it needs apps the labs won't build, value accrues up the stack, not to the model companies

“there's this quote from Sam Alman where he said you know we're going to be selling electricity we're going to be selling AI AI intelligence…”

Benedict Evans · 32:30

“And the stocks have gone nowhere in 25 years because it's an Xgrowth low margin commodity utility where they're selling this inc this objectively amazing…”

Benedict Evans · 34:00
#foundation-models#value-capture#commodity#telecom

Explainer· 5

Explainer09:30

Why the Top AI Labs Are Investing in Consultants, Not Replacing Them

Everyone assumed AI would wipe out consultants, yet OpenAI and Anthropic are the ones most heavily investing in professional services and forward-deployed engineers. Evans explains that reimagining a company's internal workflows around AI is itself a big project that needs five to ten people for a month or two, and companies don't have idle staff to do it. That's exactly why firms hire Bain, Accenture, or a branding agency, and why the labs are building that muscle.

  • A forward-deployed engineer is like an Accenture outsourced developer who works in San Francisco
  • Reimagining workflows for AI and then implementing it are two separate multi-person projects
  • Companies don't keep spare people around to run big new projects, so they hire consultants
  • The most cutting-edge AI labs are investing in the very consultancies people thought AI would kill

“companies do not have lots of people sitting around waiting to do a build a big new project or do a big new piece of…”

Benedict Evans · 10:30

“Who's going to do that? Because you don't have a bunch of people sitting around not doing anything.”

Benedict Evans · 11:30
#consulting#professional-services#ai-adoption#enterprise
Explainer12:30

The Task vs the Job: What You Actually Hire McKinsey For

Evans separates the task from the job: Claude can write the code, but figuring out what code you want, who your customer is, and what the right product is is the actual job. He uses Amazon as an analogy: Amazon gets you the SKU, but knowing which SKU you want is a separate job. What you pay Bain for isn't the 75-slide deck; it's walking your whole company, understanding the politics, and talking to your customers.

  • Sometimes the task is the job (an elevator button); usually the hard part is something else
  • Amazon gets you the SKU, but knowing which SKU you want is a different job
  • AI can make features, but deciding what features, which customer, and go-to-market is the job
  • An AI-made McKinsey deck misses the point; the deck was never what you paid for
  • You pay consultants to walk the enterprise, navigate the politics, and talk to real customers

“the claw code can write you the code, but what code do you want? It can make you the features, sure, but what features do…”

Benedict Evans · 15:00

“you'll get all these kind of AI grifters on LinkedIn and and Twitter and so on saying, "Hey, I made a McKenzie deck with Claude."…”

Benedict Evans · 15:30
#automation#jobs#consulting#product
Explainer17:30

The Job Apocalypse and the Lump-of-Labor Fallacy

Evans pushes back on the coming-job-apocalypse narrative using 200 years of history: every technology automates some jobs and unlocks new ones you can't yet imagine. In 1800, 90% of people were peasants worried about crops failing; automation since then has made everyone richer. He notes AI adoption is faster, but so was the internet, because each wave stands on the shoulders of the last, and cautions against arguments from authority on labor economics from AI-lab CEOs.

  • Every technology automates jobs and then unlocks new jobs that didn't exist yet
  • In 1800, ~90% were peasants; two centuries of automation made everyone richer
  • Vanished jobs (typesetters, telephone operators, typists) tend to be crap jobs in hindsight
  • AI adoption is fast because it stands on the shoulders of existing internet infrastructure
  • Running an AI lab doesn't make you an authority on theories of labor and comparative advantage

“you go back to 1800 like 90% of us were peasants and our major concern was would like the crops going to fail”

Benedict Evans · 19:00

“you can always see the job that's going to go going to go away and you don't know the new job because it doesn't exist…”

Benedict Evans · 19:00
#jobs#automation#economics#labor
Explainer26:30

AGI Is a Moving Target: 'AI Is Whatever Machines Can't Do Yet'

Evans notes that we have no theory of human intelligence, no theory of why the models work, and no theory of how much better they'll get, so everyone is 'vibes forecasting.' He cites Larry Tesler's line that AI is whatever machines can't do yet, because once they can do it people call it just software. AGI is now being quietly redefined to mean 'a percentage of economically valuable work', which an IBM mainframe could do in 1975, versus 'it has a soul and it's alive.'

  • We have no theory of intelligence, of why models work, or of how far they'll improve
  • Larry Tesler: AI is whatever machines can't do yet; once done, people call it just software
  • AGI is being redefined to mean doing a percentage of economically valuable work
  • An IBM mainframe in 1975 already did meaningful economically valuable work
  • Even if models stopped improving today, it's still a transformative technology

“an AI scientist called Larry Tesla who said AI is whatever machines can't do yet”

Benedict Evans · 27:00

“now clearly you can see people redefining AGI to mean the stuff that works now.”

Benedict Evans · 28:00
#agi#definitions#forecasting#ai
Explainer48:30

The Anti-AI Backlash: Sorting the Real Concerns From the Fake Ones

Evans unpacks the growing anti-AI sentiment as a fuzzy mess of very real and very fake concerns. Rising electricity bills are real but hyper-local; the data-center water panic is essentially nonsense, at roughly 0.017% of US water consumption per a Livermore Lab study. On jobs there's no clear consensus yet in the data, and the rest is a culture war over AI slop and creative work, much like the compressed, partly-true, partly-false backlash around social media.

  • Electricity bills going up is real but only in a very small number of places
  • The water panic is fake: US data centers are ~0.017% of US water consumption (Livermore Lab, 2024)
  • There's no clear consensus in the data yet that AI is hurting jobs
  • A slowdown in 18-24 year old employment appears regardless of degree or AI exposure
  • It echoes the social-media backlash: compressed, some true, some sort-of-true, some flatly false

“the water thing is weird because it's just like completely fake”

Benedict Evans · 48:30

“I actually went and dug into this at the Livermore lab did did a study at the end of 2024 where they estimated US data…”

Benedict Evans · 49:00
#anti-ai#data-centers#jobs#culture-war

Story· 1

Story57:00

The UK Post Office Scandal: Every Technology Can Ruin Lives

Evans tells the story of the UK Post Office scandal to show that every wave of technology brings new ways to ruin people's lives, by accident or design. A buggy Fujitsu point-of-sale system showed cash shortfalls at franchised post offices, and the Post Office concluded the operators were stealing. Hundreds were prosecuted, some went to prison, some went bankrupt or lost their homes, and there were suicides, while officials swore in court the system had no bugs, using 1970s-era technology.

  • UK post offices are mostly franchises, often run by second-generation Indian immigrants
  • A Fujitsu point-of-sale system had bugs that falsely showed cash shortfalls
  • The Post Office assumed theft; hundreds were prosecuted, with bankruptcies and suicides
  • Officials swore in court the system had no bugs despite it being 1970s-era technology
  • Every technology wave brings ways to ruin lives, so stay conscious without panicking

“the post office rolled out this new computer system built by them by Fujitsu that had a bunch of bugs in it that showed short…”

Benedict Evans · 57:30

“Hundreds of people get prison. Bunch of suicides, bunch of bankruptcies, people lose their homes.”

Benedict Evans · 57:30
#post-office-scandal#software-bugs#harm#accountability

Q&A· 1

Q&A1:09:00

The Analyst's Dilemma: Why Benedict Evans Struggles to Use AI

In the AI corner, Evans admits he's like the lawyer looking at a spreadsheet: the work he'd want to automate is precise information retrieval, which is exactly what today's models are worst at. He finds it genuinely useful for proofreading and images, and it worked fantastically for virtually redecorating his apartment. But as someone whose job is synthesizing lots of material into new ideas, he struggles to find AI use cases, quoting the line that AI is good at what computers are bad at and bad at what computers are good at.

  • His core work is precise information retrieval, which is exactly what models are worst at
  • He finds AI useful for proofreading and for images
  • Virtually redecorating his apartment (repaint, add furniture, change colors) worked fantastically
  • AI is good at stuff computers are bad at and bad at stuff computers are good at
  • Much of his writing starts as dictated voice memos that get transcribed, blurring AI and automation

“I I struggle with this question because I'm sort of the lawyer looking at chat GBT. So, you know, the stuff that I would do…”

Benedict Evans · 1:09:00

“somebody said AI is good at stuff that computers are bad at and bad at stuff that computers are good at.”

Benedict Evans · 1:09:30
#ai-use-cases#productivity#images#workflow

Takeaway· 3

Takeaway43:00

When the Product Is a Commodity, Distribution Is the Moat

If the underlying models are commodities, distribution becomes what matters, which is why Google sprays Gemini across its surfaces and Meta pushed its assistant everywhere. Evans compares it to browsers: the browser product is a thin wrapper on a rendering engine, and Microsoft won browsers via distribution only for it to not matter because value was further up the stack. The model, he says, is just 'the dumb thing underneath' that powers a feature.

  • When the product is a commodity, distribution is what matters
  • Google and Meta win by defaulting an adequate model onto every surface
  • The browser is a thin wrapper on a rendering engine; Microsoft won browsers and it didn't matter
  • The model is the commodity underneath; the feature and distribution decisions sit on top
  • Incumbents with existing distribution have a structural advantage over startups

“that if the product is a commodity, then the distribution is what matters.”

Benedict Evans · 43:30

“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…”

Benedict Evans · 47:00
#distribution#moats#gemini#commodity
Takeaway1:03:00

You Can't Predict What AI Will Disrupt: The Taxi and Personal-Trainer Test

Evans warns that you can't reliably predict which jobs a technology will expose. In 1997 people said taxi drivers were safe because taxis had nothing to do with the internet, yet ride-hailing transformed the whole business. Today people say personal trainers are safe, but you can prop your iPhone up, have an AI build a routine and watch your form. The lesson is humility: the big companies often come from places nobody thought were exposed.

  • In 1997, taxi drivers looked safe from the internet, yet ride-hailing remade the business
  • People now say personal trainers are safe, but an AI can build and watch a routine via your phone
  • You can't look at a senior partner and say 17% of their work is automatable; that's the expert-systems fallacy
  • The biggest companies often emerge from areas nobody predicted were exposed
  • The honest answer is usually 'it depends', which requires humility

“But the other side is well obviously like taxi drivers you couldn't automate that with the internet. It's got nothing to do with the internet.…”

Benedict Evans · 1:04:00

“So, I take my iPhone and I balance it on the metal piece with the camera pointed at me and I ask an AI to…”

Benedict Evans · 1:04:30
#disruption#prediction#jobs#humility
Takeaway1:06:30

The Only Real Advice: Dive In, Don't Stick Your Head in the Sand

Asked what people worried about their careers should do, Evans admits some professions face genuine, unclear upheaval, especially the associate tier of professional services. But hating the technology only gives you a feeling of moral superiority, not a job. The actionable move is to submerge yourself in AI, understand what it can do and how it changes things, and become a great hire, just as he and others did with the internet and mobile.

  • Some professions face a real question, especially entry-level associates in professional services
  • Refusing to engage buys moral superiority, not employability
  • Immerse yourself in AI to understand what it can do and how it changes things
  • Walking into an interview saying you'll never use AI is not the right move
  • This mirrors how people had to absorb the internet and mobile

“don't stick your head in the sand and say I hate all of this stuff because that gives you a great feeling of moral superiority…”

Benedict Evans · 1:07:30

“what helps is you diving into this completely submerging yourself in it and coming out understanding what you can do with it”

Benedict Evans · 1:07:30
#career-advice#ai-skills#jobs#adaptation