Arvind Narayanan and Sayash Kapoor · Jul 13, 2026
Arvind Narayanan argues in Normal Tech that AI will change work
Markdown · JSON · Original source
Arvind Narayanan’s keynote post is dated July 13, 2026. He argues that AI’s effects on work depend not only on model capabilities but also on reliable deployment and the slow reorganization of workplaces. Drawing on the history of electricity, he says this adaptation takes decades, and that no decision by an AI company in a lab will suddenly put everyone out of work. He also argues that AI can amplify people’s abilities, while warning that jobs will change and substantial adaptation will be needed. Narayanan’s central image is a shift from “rowing” to “steering”: as AI handles more verifiable tasks, he expects more human effort to go toward evaluating systems, exercising judgment and deciding what to build. He argues that evaluation helps people retain control over AI’s direction. He does not present this future as guaranteed: he says outcomes depend on human choices, and argues that AI systems running companies would already threaten human dignity and democratic governance, whether or not they are aligned. He closes by calling for people to keep working toward human/AI “co-.”
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: Jul 13, 2026 (2026-07-13T18:47:06.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-06T08:48:23.812Z. First observed: 2026-10-23T17:21:49.323Z. Retrieved: 2026-10-23T17:23:44.717Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 23, 2026; originally published July 13, 2026. Narayanan labels some examples and proposed changes as speculation; the keynote does not establish how future work will actually change. Narayanan describes evaluation findings and employment examples, but this reading does not establish that they predict outcomes across occupations.
Open question
What evidence would show whether AI is shifting work from building systems toward evaluating and directing them, rather than simply reducing the amount of work people do?
Source-stated
Arvind Narayanan argues that AI’s economic effects depend on downstream deployment and organizational change, not only on model capability.
the barriers are downstream, and therefore model improvements won’t rapidly change the economy.
Source-stated
Narayanan argues that no lab milestone will suddenly put everyone out of work, while predicting that work will change through gradual adaptation.
there’s no world in which something that an AI company will decide to do in a lab will put us all out of work.
Source-stated
Narayanan argues that as AI handles more verifiable tasks, more human effort may shift toward evaluating systems and deciding their direction.
some or much of the effort shifts from building to evaluation.
Source-stated
Narayanan argues that allowing AI systems to run companies would itself threaten human dignity and democratic governance, whether or not those systems are aligned.
if the AI system is not aligned, terrible things might happen — the AI could turn into a paperclip maximizer or do other catastrophic stuff.
Source-stated
Narayanan presents human/AI “co-superintelligence” as a future he wants to work toward, while acknowledging that it is not guaranteed.
this is not guaranteed, but this is the future that I want to work towards
Source-stated
The source dates the keynote post to July 13, 2026.
Arvind Narayanan Jul 13, 2026
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.
Superintelligence
A hypothetical AI far more capable than people across most intellectual tasks.
Editorial provenance
Version 1. Publication ID: d792e120-60bd-e04b-18e3-7eb32589408c.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-06T08:44:38.598Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: f1efd0d0ba5a913f59ba25d84705b2a3cda05f4c5b4d305810c421dd12daafc6. Source text hash: b43c1e8aa971d5b9a09abcd7eb66d4475845549c29eeb3644f1e5c25dcb5df76.
Revision and source-check history
Version 1 · 2026-11-06T08:48:23.812Z
Arvind Narayanan argues in Normal Tech that AI will change work
Arvind Narayanan’s keynote post is dated July 13, 2026. He argues that AI’s effects on work depend not only on model capabilities but also on reliable deployment and the slow reorganization of workplaces. Drawing on the history of electricity, he says this adaptation takes decades, and that no decision by an AI company in a lab will suddenly put everyone out of work. He also argues that AI can amplify people’s abilities, while warning that jobs will change and substantial adaptation will be needed. Narayanan’s central image is a shift from “rowing” to “steering”: as AI handles more verifiable tasks, he expects more human effort to go toward evaluating systems, exercising judgment and deciding what to build. He argues that evaluation helps people retain control over AI’s direction. He does not present this future as guaranteed: he says outcomes depend on human choices, and argues that AI systems running companies would already threaten human dignity and democratic governance, whether or not they are aligned. He closes by calling for people to keep working toward human/AI “co-.”
Original source
Arvind Narayanan and Sayash Kapoor · Article
Source details
Original source date: Jul 13, 2026 (2026-07-13T18:47:06.000Z).
Date basis: Date supplied by the publisher’s feed.
Added to Escape Meridian: 2026-11-06T08:48:23.812Z. First observed: 2026-10-23T17:21:49.323Z. Retrieved: 2026-10-23T17:23:44.717Z.
Limits
Text as retrieved on October 23, 2026; originally published July 13, 2026. Narayanan labels some examples and proposed changes as speculation; the keynote does not establish how future work will actually change. Narayanan describes evaluation findings and employment examples, but this reading does not establish that they predict outcomes across occupations.
Open question
What evidence would show whether AI is shifting work from building systems toward evaluating and directing them, rather than simply reducing the amount of work people do?
Source-stated
Arvind Narayanan argues that AI’s economic effects depend on downstream deployment and organizational change, not only on model capability.
the barriers are downstream, and therefore model improvements won’t rapidly change the economy.
Source-stated
Narayanan argues that no lab milestone will suddenly put everyone out of work, while predicting that work will change through gradual adaptation.
there’s no world in which something that an AI company will decide to do in a lab will put us all out of work.
Source-stated
Narayanan argues that as AI handles more verifiable tasks, more human effort may shift toward evaluating systems and deciding their direction.
some or much of the effort shifts from building to evaluation.
Source-stated
Narayanan argues that allowing AI systems to run companies would itself threaten human dignity and democratic governance, whether or not those systems are aligned.
if the AI system is not aligned, terrible things might happen — the AI could turn into a paperclip maximizer or do other catastrophic stuff.
Source-stated
Narayanan presents human/AI “co-superintelligence” as a future he wants to work toward, while acknowledging that it is not guaranteed.
this is not guaranteed, but this is the future that I want to work towards
Source-stated
The source dates the keynote post to July 13, 2026.
Arvind Narayanan Jul 13, 2026
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.
Superintelligence
A hypothetical AI far more capable than people across most intellectual tasks.
Editorial provenance
Version 1. Publication ID: d792e120-60bd-e04b-18e3-7eb32589408c.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-06T08:44:38.598Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: f1efd0d0ba5a913f59ba25d84705b2a3cda05f4c5b4d305810c421dd12daafc6. Source text hash: b43c1e8aa971d5b9a09abcd7eb66d4475845549c29eeb3644f1e5c25dcb5df76.
uncertain · 2026-11-06T08:44:38.598Z · anthropic/claude-opus-5.5
Publication: d792e120-60bd-e04b-18e3-7eb32589408c
attribution — pass: The piece is a single keynote write-up by Arvind Narayanan on the Normal Tech site, which he runs with Sayash Kapoor. The reading attributes the arguments to Narayanan and keeps them framed as his claims.
I had the honor of giving a keynote at the International Conference on Machine Learning in Seoul last week
summary — pass: The reading conveys the main theses: deployment and slow adaptation, no sudden lab-driven job loss, the rowing-to-steering shift toward evaluation, and co-superintelligence. It also keeps the conditional framing. The title is bland but accurate. Terms like verifiable tasks and aligned are lightly explained at most, but they do not block a general reader from following the argument. Omitting the RSI/AGI/ASI distinctions is a compression, not a distortion.
Effort shifts from “rowing the boat” to “steering the ship, navigating the ship, and figuring out where we even want to go”.
date — pass: originalPublished is 2026-07-13T18:47Z, which matches the source dateline of Jul 13, 2026, and is stated at day precision. The page is a publisher page, so no archive disclosure is needed. The August 2026 update note about the video is a later addition, and the reading neither uses nor relies on it. There is no hindsight framing.
Arvind Narayanan Jul 13, 2026
support — uncertain: Claim 4 is supported elsewhere in the source, but its cited excerpt is the pessimists' view that unaligned AI could do catastrophic things. That excerpt does not support the claim that AI running companies is a threat whether or not it is aligned. Smallest repair: cite the passage quoted here instead. The other claims are adequately supported by their excerpts.
that’s already dystopian. It doesn’t matter if that AI is aligned or not.
media — pass: No media was selected, and no verified media candidates were supplied. With nothing chosen, there is no identity or relevance problem.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A hypothetical AI far more capable than people across most intellectual tasks.
Definition referenceA hypothetical AI far more capable than people across most intellectual tasks.
Definition reference