Arvind Narayanan and Sayash Kapoor · May 1, 2025
Arvind Narayanan and Sayash Kapoor argue in Normal Tech that AGI is not a milestone
Markdown · JSON · Original source
Sayash Kapoor and Arvind Narayanan published this essay on May 1, 2025. They argue that is not a milestone with a clear capability threshold or sudden consequences: a system’s capabilities do not by themselves determine its effects, which depend on how it is used and adopted. They distinguish capability—the ability to do specified tasks—from power, which depends on the environment and authority given to a system. They also argue that AI’s economic effects depend on diffusion: developing useful products, training workers, changing organisations and establishing laws and norms. These processes, they write, unfold over decades, so a declaration of AGI would not tell businesses or policymakers what to do. Their conclusion is that AI’s effects will come through “millions of boring little business process adaptations and policy tweaks,” rather than a sprint toward a magic-bullet technology. They urge businesses to evaluate AI products carefully and policymakers to focus on productive and safe diffusion. The essay’s forecasts about diffusion and its argument about control are the authors’ views; the pace and eventual effects remain uncertain.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Arvind Narayanan and Sayash Kapoor · Article
Source details
Original source date: May 1, 2025 (2025-05-01T11:47:59.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-14T09:58:23.606Z. First observed: 2026-10-23T10:33:03.996Z. Retrieved: 2026-10-23T10:36:09.550Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The essay does not analyze military AI, which the authors say they plan to address in future essays. Text as retrieved on October 23, 2026; originally published May 1, 2025.
Open question
What evidence would show that diffusion, rather than a capability threshold, is driving AI’s economic effects?
Source-stated
Sayash Kapoor and Arvind Narayanan argue that AGI does not mark a capability threshold with particular significance or immediate consequences.
In this essay, we argue that AGI is not a milestone. It does not represent a discontinuity in the properties or impacts of AI systems.
Source-stated
The authors argue that AI’s economic impact depends on adoption and complementary changes, a process they expect to unfold over decades.
Like past general-purpose technologies, we expect the economic impacts of AI to be realized over decades, as this process of diffusion unfolds.
Source-stated
The authors distinguish an AI system’s capability from the power it has through its operating environment, which they say people can design.
Capability is an intrinsic property of an AI system, whereas power is a matter of how we design the environment in which AI systems operate. And humans have agency over this design.
Source-stated
The authors conclude that AI’s effects will arise through many small changes to business processes and policy, not a sprint toward a single technology.
AI's impact on the world will be realized not through a sprint toward a magic-bullet technology but through millions of boring little business process adaptations and policy tweaks.
Source-stated
The essay is dated May 1, 2025.
May 01, 2025
Source and contributor context
Arvind Narayanan and Sayash Kapoor · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Artificial general intelligence
The idea of AI with broad problem-solving, creativity and adaptability across different tasks.
Editorial provenance
Version 1. Publication ID: 2802f397-d497-5d9a-ed66-45e019e87cf2.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-14T09:54:18.879Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_screen.
Observation: 26f940b0f2cd1d952fae0d0364e990f7dd3cd736a74a5e7559b680649d96e8e5. Source text hash: c765b81ea6e3a002bdb8fd26884e46ad1cf4170cf2d9964ea849770f997cd687.
Revision and source-check history
Version 1 · 2026-11-14T09:58:23.606Z
Arvind Narayanan and Sayash Kapoor argue in Normal Tech that AGI is not a milestone
Sayash Kapoor and Arvind Narayanan published this essay on May 1, 2025. They argue that is not a milestone with a clear capability threshold or sudden consequences: a system’s capabilities do not by themselves determine its effects, which depend on how it is used and adopted. They distinguish capability—the ability to do specified tasks—from power, which depends on the environment and authority given to a system. They also argue that AI’s economic effects depend on diffusion: developing useful products, training workers, changing organisations and establishing laws and norms. These processes, they write, unfold over decades, so a declaration of AGI would not tell businesses or policymakers what to do. Their conclusion is that AI’s effects will come through “millions of boring little business process adaptations and policy tweaks,” rather than a sprint toward a magic-bullet technology. They urge businesses to evaluate AI products carefully and policymakers to focus on productive and safe diffusion. The essay’s forecasts about diffusion and its argument about control are the authors’ views; the pace and eventual effects remain uncertain.
Original source
Arvind Narayanan and Sayash Kapoor · Article
Source details
Original source date: May 1, 2025 (2025-05-01T11:47:59.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-14T09:58:23.606Z. First observed: 2026-10-23T10:33:03.996Z. Retrieved: 2026-10-23T10:36:09.550Z.
Limits
The essay does not analyze military AI, which the authors say they plan to address in future essays. Text as retrieved on October 23, 2026; originally published May 1, 2025.
Open question
What evidence would show that diffusion, rather than a capability threshold, is driving AI’s economic effects?
Source-stated
Sayash Kapoor and Arvind Narayanan argue that AGI does not mark a capability threshold with particular significance or immediate consequences.
In this essay, we argue that AGI is not a milestone. It does not represent a discontinuity in the properties or impacts of AI systems.
Source-stated
The authors argue that AI’s economic impact depends on adoption and complementary changes, a process they expect to unfold over decades.
Like past general-purpose technologies, we expect the economic impacts of AI to be realized over decades, as this process of diffusion unfolds.
Source-stated
The authors distinguish an AI system’s capability from the power it has through its operating environment, which they say people can design.
Capability is an intrinsic property of an AI system, whereas power is a matter of how we design the environment in which AI systems operate. And humans have agency over this design.
Source-stated
The authors conclude that AI’s effects will arise through many small changes to business processes and policy, not a sprint toward a single technology.
AI's impact on the world will be realized not through a sprint toward a magic-bullet technology but through millions of boring little business process adaptations and policy tweaks.
Source-stated
The essay is dated May 1, 2025.
May 01, 2025
Source and contributor context
Arvind Narayanan and Sayash Kapoor · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Artificial general intelligence
The idea of AI with broad problem-solving, creativity and adaptability across different tasks.
Editorial provenance
Version 1. Publication ID: 2802f397-d497-5d9a-ed66-45e019e87cf2.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-14T09:54:18.879Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_screen.
Observation: 26f940b0f2cd1d952fae0d0364e990f7dd3cd736a74a5e7559b680649d96e8e5. Source text hash: c765b81ea6e3a002bdb8fd26884e46ad1cf4170cf2d9964ea849770f997cd687.
passed · 2026-11-14T09:54:18.879Z · anthropic/claude-opus-5.5
Publication: 2802f397-d497-5d9a-ed66-45e019e87cf2
attribution — pass: The essay is correctly attributed to Sayash Kapoor and Arvind Narayanan, the bylined authors of this single essay on their Normal Tech publication. The reading attributes the views to them as their argument, not as established fact.
Sayash Kapoor and Arvind Narayanan May 01, 2025
summary — pass: The summary keeps the authors' central thesis and force. That thesis is that AGI is not a milestone because there is no threshold, impact comes through decades of diffusion, and capability differs from power. The reading explains diffusion and capability versus power in plain language, and the definitions list explains AGI. The title and question follow from the essay. One wording point is minor: describing power as depending on the authority given to a system is a slight gloss on the source's 'ability to modify the environment', but it does not distort the meaning.
AGI is not a milestone because it is not actionable.
date — pass: evidence.originalPublished is 2025-05-01T11:47:59Z, and the source dateline is May 01, 2025. The reading states May 1, 2025, at day precision, which matches. There is no hindsight framing. Forecasts are attributed to the authors, and the evidence kind is a publisher page, so no archive disclosure is needed. The modification date equals the publication date.
May 01, 2025
support — pass: Each claim is supported by its quoted passage. The recommendations to businesses and policymakers match the source, and the forecasts are labelled as the authors' views. The reading adds no unsupported benefits, costs or stakeholder claims.
rather than fixating on AGI, they should focus on enabling productive and safe diffusion, including of existing AI.
media — pass: No media was selected, and no verified media candidates were supplied, so nothing is misidentified.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
The idea of AI with broad problem-solving, creativity and adaptability across different tasks.
Definition referenceThe idea of AI with broad problem-solving, creativity and adaptability across different tasks.
Definition reference