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

Episode Summary

Executive Summary: Benedict Evans argues AI is a major platform shift comparable to the internet or mobile, but not necessarily bigger. He says adoption and impact will unfold unevenly, with software already transformed while most other industries lag. The real question is not job extinction but where value accrues, how distribution changes, and which workflows become commoditized versus strategic.

Main Topics: AI as a platform shift, not an apocalypse (Priority: 5/5): Evans frames AI as as significant as the internet or mobile, but rejects doom narratives that predict immediate mass layoffs or total economic upheaval. He emphasizes that the pattern of technology is repeated automation plus new job creation. The difference between tasks and jobs (Priority: 5/5): A central theme is that AI may automate tasks, but many jobs contain the hard-to-automate parts: judgment, context, politics, customer discovery, and deciding what to build. He uses accounting, law, and consulting as examples. Software is the leading edge of AI change (Priority: 5/5): Evans says software development is already the most visibly transformed profession, akin to accountants seeing VisiCalc in the late 1970s. Other industries are in a much earlier, uneven adoption phase. Value capture and model-layer economics (Priority: 5/5): He questions whether foundation model companies will retain pricing power. His view is that models may become commoditized infrastructure, with value accruing higher in the stack, especially in applications and distribution. Distribution becomes more important as products commoditize (Priority: 4/5): As AI products become easier to build, brand, defaults, and embedded distribution from incumbents like Google, Apple, and Meta may matter more than raw model quality. He compares this to browsers, cloud, and mobile platform dynamics. Backlash, politics, and uneven social impact (Priority: 4/5): Evans sees anti-AI sentiment as a mix of real harms, misunderstandings, and culture-war reactions. He argues the data on job loss is inconclusive, but the social and political backlash is already real. Practical adaptation over moral purity (Priority: 5/5): His advice is to engage deeply with AI rather than reject it. People should learn how it works, understand where it helps, and become the kind of person companies want to hire in the new environment.

Key Arguments: AI is a huge deal, but it is more like the internet or mobile than the industrial revolution; debating whether it is 20% or 100% bigger is not useful. Technology waves usually destroy some jobs, create new ones, and expand total economic value; the jobs that disappear are obvious, while the new ones are hard to predict beforehand. AI adoption is uneven: software developers are already seeing dramatic changes, while many professionals use AI only weekly or not at all. You cannot meaningfully decompose a profession into fixed percentages of automatable work; jobs are not just sums of tasks, and the important parts are often the least automatable. Professional services will not disappear; AI will likely change how firms deliver value, but the hardest part is understanding what clients really need, not generating a deck or memo. Foundation models may not have durable pricing power if they become commoditized and multiple providers compete; the value may shift to apps, workflows, and distribution. Incumbents with strong distribution may win more than startups because AI products are often thin wrappers around commoditized capabilities. Anti-AI backlash is driven by a fuzzy mix of labor fears, real harms like deepfakes, niche creator concerns, and broader cultural resistance. The lack of transparent usage and productivity data means many AI claims are still vibes-based rather than empirically settled. The right response for workers is to learn and experiment with AI, not dismiss it morally; engagement is a career advantage.

Data Points: AI market comparison: as big a deal as the internet or mobile - Evans’s framing of AI’s importance relative to previous platform shifts Adoption among 13–18-year-olds: 15–20% daily active users; another ~20% weekly active users - He cites survey data showing wide variance in AI usage even among teens Data center water use in the U.S.: 0.017% of U.S. water consumption - Used to argue that many anti-data-center claims overstate the water problem Enterprise software sales cycle: ~18 months if you're lucky - Illustrates why AI adoption inside large companies will not happen overnight Mobile industry revenue: about $1 trillion/year - Example of a huge infrastructure layer that still captures relatively little of the ecosystem value Mobile telecom CapEx: about $200 billion/year - Supports his argument about utility-like, low-margin infrastructure economics Mobile data consumption growth: about 1,500–2,000x since 2010 globally - Shows massive infrastructure usage growth without proportional equity returns AI usage scale: 900 million chat users - Used to show how AI can scale quickly because it sits on top of existing internet distribution Mainframe install base peak: ~70,000–80,000 units - Historical comparison to show how platform scale changed across eras U.S. supermarket product count: chart shows large increase since the 1950s - Used as an example of how barcodes and inventory systems expanded what could be stocked Livermore study estimate: US data center water consumption at 0.017% - Referenced as evidence against dramatic water-use panic Recorded music revenue: fell about 50% from 2000 to 2015, then recovered to ~75% of peak - Illustrates how a technology can first destroy an old business model and later create a new one

Pivotal Quotes: "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: His core thesis on AI’s scale and limits "You can't look at a senior partner at a law firm and say, well, 17% of their work could be automated. This is horseshit." — Benedict Evans: He rejects simplistic task decomposition models for professions "Don't stick your head in the sand and say, I hate all of this stuff... What helps is you diving into this and coming out understanding what you can do with it." — Benedict Evans: Advice for workers and students adapting to AI

Implications: AI is likely to reshape software, services, and distribution before it fully transforms labor markets. Winners will be those who learn the tools, redesign workflows, and capture value above the model layer rather than assuming AI itself is the moat.