--- title: What will be left for us to work on? | daily.dev description: Princeton professor Arvind Narayanan's ICML 2026 keynote argues that AI should be understood as a transformative but 'normal' technology — comparable to... canonical: https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: What will be left for us to work on? | daily.dev og:description: Princeton professor Arvind Narayanan's ICML 2026 keynote argues that AI should be understood as a transformative but 'normal' technology — comparable to... og:url: https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar og:image: https://api.daily.dev/og/posts/tTOKPhlaR.png og:image:alt: What will be left for us to work on? 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. # What will be left for us to work on? **[AI Snake Oil](https://daily.dev/sources/aisnakeoil)** · 38 min read · 18 upvotes · 0 comments ## Summary Princeton professor Arvind Narayanan's ICML 2026 keynote argues that AI should be understood as a transformative but 'normal' technology — comparable to electricity — that amplifies human potential rather than replacing workers. He presents three main arguments: the 'AI as Normal Technology' framework accurately describes AI's medium-term trajectory; recursive self-improvement won't suddenly eliminate jobs because AGI, ASI, and economic transformation are distinct dimensions of progress; and future jobs will shift dramatically from 'building' to 'evaluating and steering.' Drawing on research showing AI agents have improved in capability but not reliability, he explains why coding is not the bottleneck in software engineering and why Jevons' paradox suggests productivity gains historically increase employment. He advocates for human/AI 'co-superintelligence,' warns against black-box AI usage and the 'dependence spiral,' and calls for the AI research community to invest far more in evaluation over benchmark-driven building. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### why hasn't AI replaced software engineers even though coding agents have gotten much better Writing code was never the main bottleneck in software engineering; it was only about one-third of the work to begin with. The decide layer (requirements, planning) and deliver layer (integration, testing, accountability, maintenance) are not compressed by AI and are arguably expanding as coding gets faster, so overall demand for engineers has not dropped. _Track how coding agents are actually reshaping engineering roles by following analysis like this on daily.dev._ ### how much has AI agent reliability improved compared to raw capability over the last two years Capability scores on benchmarks rose dramatically over roughly the last 24 months for frontier models from three leading AI companies, but reliability metrics improved by only five to ten percentage points in the same period. Reliability was broken into consistency, robustness, calibration, and operational safety, none of which are standard in typical agent benchmarks. _Compare agent reliability data against capability hype before betting production workflows on it, with daily.dev._ ### can an AI system that improves its own architecture (recursive self-improvement) lead to superintelligence Not automatically. Recursive self-improvement, invention of better architectures or data pipelines through automated testing, is closer to glorified hyperparameter search (like AutoML) than to replacing the creativity of the worldwide AI research community. AI remains far better at verifiable tasks than unverifiable ones like creativity, so near-term RSI is unlikely to produce humanlike general intelligence or superintelligence. _Separate RSI hype from real capability limits when evaluating AI research claims on daily.dev._ ## Similar posts on daily.dev - [Why human \+ AI is the only answer that makes sense](https://daily.dev/posts/why-human-ai-is-the-only-answer-that-makes-sense-v7bhzz1uw) · All Things Open · 0 upvotes · 0 comments - [Something Big Is Happening](https://daily.dev/posts/something-big-is-happening-lgyfb7xtb) · Reid Burke · 44 upvotes · 6 comments - [How Human Work Will Remain Valuable in an AI World](https://daily.dev/posts/how-human-work-will-remain-valuable-in-an-ai-world-6lsa2mrss) · Towards Data Science · 0 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#future-of-work](https://daily.dev/tags/future-of-work) [View this post on daily.dev](https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar) ```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":"What will be left for us to work on?","url":"https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar"},"datePublished":"2026-07-13T18:51:38.321Z","dateModified":"2026-09-14T07:25:50.049Z","description":"Princeton professor Arvind Narayanan's ICML 2026 keynote argues that AI should be understood as a transformative but 'normal' technology — comparable to...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/69e20d1f81d7a53a2f08ad6e1b004833?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/69e20d1f81d7a53a2f08ad6e1b004833?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"AI Snake Oil","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":"AI Snake Oil","logo":"https://media.daily.dev/image/upload/s--hnyXgKKI--/f_auto/v1727446615/logos/aisnakeoil","url":"https://daily.dev/sources/aisnakeoil"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":18},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,llm,ai-agents,future-of-work","timeRequired":"PT38M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Snake Oil","item":"https://daily.dev/sources/aisnakeoil"},{"@type":"ListItem","position":3,"name":"What will be left for us to work on?"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/what-will-be-left-for-us-to-work-on--ttokphlar#faq","mainEntity":[{"@type":"Question","name":"why hasn't AI replaced software engineers even though coding agents have gotten much better","acceptedAnswer":{"@type":"Answer","text":"Writing code was never the main bottleneck in software engineering; it was only about one-third of the work to begin with. The decide layer (requirements, planning) and deliver layer (integration, testing, accountability, maintenance) are not compressed by AI and are arguably expanding as coding gets faster, so overall demand for engineers has not dropped. Track how coding agents are actually reshaping engineering roles by following analysis like this on daily.dev."}},{"@type":"Question","name":"how much has AI agent reliability improved compared to raw capability over the last two years","acceptedAnswer":{"@type":"Answer","text":"Capability scores on benchmarks rose dramatically over roughly the last 24 months for frontier models from three leading AI companies, but reliability metrics improved by only five to ten percentage points in the same period. Reliability was broken into consistency, robustness, calibration, and operational safety, none of which are standard in typical agent benchmarks. Compare agent reliability data against capability hype before betting production workflows on it, with daily.dev."}},{"@type":"Question","name":"can an AI system that improves its own architecture (recursive self-improvement) lead to superintelligence","acceptedAnswer":{"@type":"Answer","text":"Not automatically. Recursive self-improvement, invention of better architectures or data pipelines through automated testing, is closer to glorified hyperparameter search (like AutoML) than to replacing the creativity of the worldwide AI research community. AI remains far better at verifiable tasks than unverifiable ones like creativity, so near-term RSI is unlikely to produce humanlike general intelligence or superintelligence. Separate RSI hype from real capability limits when evaluating AI research claims on daily.dev."}}]} ```