--- title: Andrej Karpathy says only 5,000 people have seen what AI... description: Andrej Karpathy argues that a vast capability gap exists between the few thousand people with access to frontier AI systems inside labs and the billions who've... canonical: https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Andrej Karpathy says only 5,000 people have seen what AI can really do | daily.dev og:description: Andrej Karpathy argues that a vast capability gap exists between the few thousand people with access to frontier AI systems inside labs and the billions who've... og:url: https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj og:image: https://api.daily.dev/og/posts/eYdbZwnqj.png og:image:alt: Andrej Karpathy says only 5,000 people have seen what AI can really do og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Andrej Karpathy says only 5,000 people have seen what AI can really do **[The New Stack](https://daily.dev/sources/newstack)** · 5 min read · 0 upvotes · 0 comments ## Summary Andrej Karpathy argues that a vast capability gap exists between the few thousand people with access to frontier AI systems inside labs and the billions who've only casually used chatbots. Box CEO Aaron Levie builds on this, framing it as evidence of how early AI diffusion really is, and pitches agent swarms as the next evolution of enterprise AI workflows, while acknowledging cost, reliability, and change-management obstacles still block wide adoption. Karpathy also explains why coding and math benefit most from reinforcement learning: results are digital and verifiable, unlike writing or advice. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why does Andrej Karpathy think most people underestimate what AI can currently do? He estimates that of roughly 6 billion people worldwide, most have barely used a large language model, 1-2 billion use chatbots casually, about 20 million use frontier models professionally for math and code, and only around 5,000 people have access to frontier systems inside AI labs, creating a widening capability gap. _Developers tracking how fast coding agents are actually advancing can follow analysis like this on daily.dev._ ### Why are coding and math tasks improving faster with AI than writing or advice? Capability gains concentrate in domains that are digital, verifiable, and economically valuable, according to Karpathy. Coding and math results can be tested and measured for reinforcement learning, while writing and advice are harder to evaluate objectively, and more commercial incentive exists to improve technical work, so labs focus effort there. _Anyone deciding where to trust AI agents most can keep up with these capability debates on daily.dev._ ### What are agent swarms and why does Box think they matter for enterprise AI? Agent swarms are systems that divide complex work among multiple AI agents running in parallel, which Box CEO Aaron Levie calls possibly the ultimate form factor for agents inside workflows. Adoption remains rare because models aren't equally capable across every task, running many agents at scale is expensive, and management tooling is still immature. _Enterprise teams weighing multi-agent workflows can follow how this space evolves on daily.dev._ ## Similar posts on daily.dev - [OpenAI Co-Founder: AI Agents Are Still 10 Years Away](https://daily.dev/posts/openai-co-founder-ai-agents-are-still-10-years-away-jwyarmzzq) · The New Stack · 19 upvotes · 1 comments - [Andrej Karpathy — AGI is still a decade away](https://daily.dev/posts/andrej-karpathy-agi-is-still-a-decade-away-bvlwybhbu) · Tech Lead Digest · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents) [View this post on daily.dev](https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Andrej Karpathy says only 5,000 people have seen what AI can really do","url":"https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj"},"datePublished":"2026-10-12T15:59:01.136Z","dateModified":"2026-10-15T08:31:33.918Z","description":"Andrej Karpathy argues that a vast capability gap exists between the few thousand people with access to frontier AI systems inside labs and the billions who've...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/fbc716bb85bf0580464f6cbbbc15a5b0?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/fbc716bb85bf0580464f6cbbbc15a5b0?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Andrej Karpathy says only 5,000 people have seen what AI can really do"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/andrej-karpathy-says-only-5-000-people-have-seen-what-ai-can-really-do-eydbzwnqj#faq","mainEntity":[{"@type":"Question","name":"Why does Andrej Karpathy think most people underestimate what AI can currently do?","acceptedAnswer":{"@type":"Answer","text":"He estimates that of roughly 6 billion people worldwide, most have barely used a large language model, 1-2 billion use chatbots casually, about 20 million use frontier models professionally for math and code, and only around 5,000 people have access to frontier systems inside AI labs, creating a widening capability gap. Developers tracking how fast coding agents are actually advancing can follow analysis like this on daily.dev."}},{"@type":"Question","name":"Why are coding and math tasks improving faster with AI than writing or advice?","acceptedAnswer":{"@type":"Answer","text":"Capability gains concentrate in domains that are digital, verifiable, and economically valuable, according to Karpathy. Coding and math results can be tested and measured for reinforcement learning, while writing and advice are harder to evaluate objectively, and more commercial incentive exists to improve technical work, so labs focus effort there. Anyone deciding where to trust AI agents most can keep up with these capability debates on daily.dev."}},{"@type":"Question","name":"What are agent swarms and why does Box think they matter for enterprise AI?","acceptedAnswer":{"@type":"Answer","text":"Agent swarms are systems that divide complex work among multiple AI agents running in parallel, which Box CEO Aaron Levie calls possibly the ultimate form factor for agents inside workflows. Adoption remains rare because models aren't equally capable across every task, running many agents at scale is expensive, and management tooling is still immature. Enterprise teams weighing multi-agent workflows can follow how this space evolves on daily.dev."}}]} ```