--- title: 30 Months on the Exponential (in Charts) | daily.dev description: Ethan Mollick's new book 'Co-Existence' is out, 30 months after 'Co-Intelligence', and he charts how much AI has changed since. On the METR Long Horizon... canonical: https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: 30 Months on the Exponential (in Charts) | daily.dev og:description: Ethan Mollick's new book 'Co-Existence' is out, 30 months after 'Co-Intelligence', and he charts how much AI has changed since. On the METR Long Horizon... og:url: https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti og:image: https://api.daily.dev/og/posts/LupPoY7tI.png og:image:alt: 30 Months on the Exponential (in Charts) 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. # 30 Months on the Exponential (in Charts) **[Tech Lead Digest](https://daily.dev/sources/techleaddigest)** · 7 min read · 1 upvotes · 0 comments ## Summary Ethan Mollick's new book 'Co-Existence' is out, 30 months after 'Co-Intelligence', and he charts how much AI has changed since. On the METR Long Horizon Benchmark, the length of tasks an AI could complete doubled eight times, going from 4 minutes in 2024 to 17 hours by April 2026, before the benchmark ran out of headroom. The Google Proof Q&A benchmark cost dropped from 30 cents per correct answer to four hundredths of a cent. 180 AI models now beat GPT-4's original release, and 19 can run on a smartphone. Mollick made the post's charts and an otter video by delegating to Opus 5.5 acting as an orchestrator agent in Claude Projects, a mode of work he calls 'conjurer'. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much did AI task-completion ability improve on the METR Long Horizon Benchmark between 2024 and 2026? The amount of human work an AI could complete with at least 50% success doubled eight times, going from a 4-minute task in 2024 to 17 hours of work by April 2026. After that, METR said it could no longer accurately measure AI capability with its existing methods, and other studies suggest the real figure may be as high as 17 weeks of work. _Developers gauging how fast agentic AI tools are closing capability gaps can follow benchmark shifts like this on daily.dev._ ### How much has the cost of running the Google Proof Question and Answer exam with AI dropped? The cost of getting a correct answer on the GPQA exam, which is designed to stump graduate students, fell from about 30 cents in AI usage to roughly four hundredths of a cent for the same result. This cost collapse happened over the same 30-month period in which task-completion ability on other benchmarks rose dramatically. _Teams budgeting AI usage costs track pricing drops like this one on daily.dev before committing to a model._ ### What is 'conjurer' work in the context of using AI agents, as opposed to centaur or cyborg work? Conjurer work is a way of collaborating with AI where you summon agents, give them general instructions, and evaluate what they return, rather than staying continuously in the loop. It contrasts with centaur work (dividing tasks between AI and humans) and cyborg work (blending AI and human contributions), both described in Co-Intelligence; conjuring reflects newer agentic AI like an orchestrator agent in Claude Projects delegating to sub-agents. _Developers rethinking their workflow around autonomous coding agents can track emerging patterns like this on daily.dev._ ## Similar posts on daily.dev - [Measuring AI Ability to Complete Long Software Tasks](https://daily.dev/posts/measuring-ai-ability-to-complete-long-software-tasks-zt241vimh) · Metadata · 1 upvotes · 0 comments - [The twilight of the chatbots](https://daily.dev/posts/the-twilight-of-the-chatbots-xd1w1acfw) · Tech Lead Digest · 1 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#ai-agents](https://daily.dev/tags/ai-agents), [#claude](https://daily.dev/tags/claude), [#vibe-coding](https://daily.dev/tags/vibe-coding) [View this post on daily.dev](https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti) ```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":"30 Months on the Exponential (in Charts)","url":"https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti"},"datePublished":"2026-10-26T00:21:24.305Z","dateModified":"2026-10-26T00:22:06.121Z","description":"Ethan Mollick's new book 'Co-Existence' is out, 30 months after 'Co-Intelligence', and he charts how much AI has changed since. On the METR Long Horizon...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b88f5f09cd338144f714caec99cc8f7e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b88f5f09cd338144f714caec99cc8f7e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Tech Lead Digest","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":"Tech Lead Digest","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/6ed1ee1a2e5c49f698265a6b2912b80d","url":"https://daily.dev/sources/techleaddigest"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,ai-agents,claude,vibe-coding","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Tech Lead Digest","item":"https://daily.dev/sources/techleaddigest"},{"@type":"ListItem","position":3,"name":"30 Months on the Exponential (in Charts)"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/30-months-on-the-exponential-in-charts--luppoy7ti#faq","mainEntity":[{"@type":"Question","name":"How much did AI task-completion ability improve on the METR Long Horizon Benchmark between 2024 and 2026?","acceptedAnswer":{"@type":"Answer","text":"The amount of human work an AI could complete with at least 50% success doubled eight times, going from a 4-minute task in 2024 to 17 hours of work by April 2026. After that, METR said it could no longer accurately measure AI capability with its existing methods, and other studies suggest the real figure may be as high as 17 weeks of work. Developers gauging how fast agentic AI tools are closing capability gaps can follow benchmark shifts like this on daily.dev."}},{"@type":"Question","name":"How much has the cost of running the Google Proof Question and Answer exam with AI dropped?","acceptedAnswer":{"@type":"Answer","text":"The cost of getting a correct answer on the GPQA exam, which is designed to stump graduate students, fell from about 30 cents in AI usage to roughly four hundredths of a cent for the same result. This cost collapse happened over the same 30-month period in which task-completion ability on other benchmarks rose dramatically. Teams budgeting AI usage costs track pricing drops like this one on daily.dev before committing to a model."}},{"@type":"Question","name":"What is 'conjurer' work in the context of using AI agents, as opposed to centaur or cyborg work?","acceptedAnswer":{"@type":"Answer","text":"Conjurer work is a way of collaborating with AI where you summon agents, give them general instructions, and evaluate what they return, rather than staying continuously in the loop. It contrasts with centaur work (dividing tasks between AI and humans) and cyborg work (blending AI and human contributions), both described in Co-Intelligence; conjuring reflects newer agentic AI like an orchestrator agent in Claude Projects delegating to sub-agents. Developers rethinking their workflow around autonomous coding agents can track emerging patterns like this on daily.dev."}}]} ```