# Eric Schmidt and Suhas Mahesh argue in MIT Technology Review that AI agents could accelerate science

[Canonical reading](https://escapemeridian.com/readings/eric-schmidt-and-suhas-mahesh-argue-in-mit-technology-review-7dd759a1)

On August 10, 2026, Eric Schmidt and Suhas Mahesh argue in MIT Technology Review that AI agents may offer a different route to accelerating scientific discovery than the approach illustrated by AlphaFold. They say AlphaFold relied on experimentally validated protein structures, but comparable datasets may be difficult or impossible to create in many fields. Their proposed alternative is to model research as an iterative, highly contingent process. In their account, AI agents are generalists that digitally model the human process of discovery, rather than a new way to do science. The distinction matters because their case for agents rests not on reproducing AlphaFold’s approach at larger scale, but on modeling how research proceeds. The newsletter points readers to a full op-ed, which is not included in the retrieved text; it does not establish whether agents can accelerate discovery in practice or across different fields.

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

MIT Technology Review · Article

Original source date: Aug 10, 2026 (2026-08-10T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T19:28:10.165Z. First observed: 2026-11-19T17:48:05.989Z. Retrieved: 2026-11-19T17:50:01.281Z.

- [The Download: AI agents for science, and the “censorship-industrial complex”](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Limits

Text as retrieved on November 19, 2026; originally published August 10, 2026. The retrieved newsletter points to a full op-ed but does not include it, so the authors’ fuller reasoning and evidence are unavailable here. The text presents the authors’ argument; it does not establish whether AI agents accelerate scientific discovery in practice.


## Open question

How well can AI agents model the iterative, highly contingent process of research across fields where comparable datasets are difficult or impossible to create?


## Source-stated

Schmidt and Mahesh argue that AlphaFold may not be the best template for accelerating science.

> AlphaFold may not be the best template for accelerating science.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source-stated

They say comparable datasets may be difficult or impossible to create in many fields.

> Comparable datasets will be difficult or impossible to create in many fields.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source-stated

They argue that AI agents can model research as an iterative, highly contingent process.

> AI agents can instead model the iterative, highly contingent process of actual research.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source-stated

They characterize agents as generalists that digitally model the human process of discovery, not as a new way to do science.

> They do not represent a new way to do science—instead, they digitally model the human process of discovery.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source and contributor context

MIT Technology Review · 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.


## AI agent

An AI system that chooses and carries out actions toward a goal, often using software tools.

- [Definition reference](https://www.ibm.com/think/topics/ai-agents)

## AlphaFold

AlphaFold is a family of AI systems from Google DeepMind that predicts molecular structures and interactions.

- [Explore this context](/entities/alphafold)
- [Reference](https://deepmind.google/science/alphafold/)
- [Reference](https://deepmind.google/blog/)

## Editorial provenance

Version 1. Publication ID: 3b5e3b80-0ca5-afd5-8ff0-81e778b76657.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T19:24:43.363Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 857a3da575cf5383ee39254945b5bbf631d21cd8575aaebea751854a7e0ea68d. Source text hash: a37dbf1bca7b3f29130a28288d8d8a233ced4fad8cf003a148f5e1f84ce9b4a0.


## Revision and source-check history

### Version 1 — 2026-11-19T19:28:10.165Z

On August 10, 2026, Eric Schmidt and Suhas Mahesh argue in MIT Technology Review that AI agents may offer a different route to accelerating scientific discovery than the approach illustrated by AlphaFold. They say AlphaFold relied on experimentally validated protein structures, but comparable datasets may be difficult or impossible to create in many fields. Their proposed alternative is to model research as an iterative, highly contingent process. In their account, AI agents are generalists that digitally model the human process of discovery, rather than a new way to do science. The distinction matters because their case for agents rests not on reproducing AlphaFold’s approach at larger scale, but on modeling how research proceeds. The newsletter points readers to a full op-ed, which is not included in the retrieved text; it does not establish whether agents can accelerate discovery in practice or across different fields.

## Original source

MIT Technology Review · Article

Original source date: Aug 10, 2026 (2026-08-10T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T19:28:10.165Z. First observed: 2026-11-19T17:48:05.989Z. Retrieved: 2026-11-19T17:50:01.281Z.

- [The Download: AI agents for science, and the “censorship-industrial complex”](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Limits

Text as retrieved on November 19, 2026; originally published August 10, 2026. The retrieved newsletter points to a full op-ed but does not include it, so the authors’ fuller reasoning and evidence are unavailable here. The text presents the authors’ argument; it does not establish whether AI agents accelerate scientific discovery in practice.


## Open question

How well can AI agents model the iterative, highly contingent process of research across fields where comparable datasets are difficult or impossible to create?


## Source-stated

Schmidt and Mahesh argue that AlphaFold may not be the best template for accelerating science.

> AlphaFold may not be the best template for accelerating science.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source-stated

They say comparable datasets may be difficult or impossible to create in many fields.

> Comparable datasets will be difficult or impossible to create in many fields.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source-stated

They argue that AI agents can model research as an iterative, highly contingent process.

> AI agents can instead model the iterative, highly contingent process of actual research.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source-stated

They characterize agents as generalists that digitally model the human process of discovery, not as a new way to do science.

> They do not represent a new way to do science—instead, they digitally model the human process of discovery.

- [Supporting source](https://www.technologyreview.com/2026/08/10/1141526/the-download-ai-agents-science-censorship-industrial-complex/)

## Source and contributor context

MIT Technology Review · 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.


## AI agent

An AI system that chooses and carries out actions toward a goal, often using software tools.

- [Definition reference](https://www.ibm.com/think/topics/ai-agents)

## AlphaFold

AlphaFold is a family of AI systems from Google DeepMind that predicts molecular structures and interactions.

- [Explore this context](/entities/alphafold)
- [Reference](https://deepmind.google/science/alphafold/)
- [Reference](https://deepmind.google/blog/)

## Editorial provenance

Version 1. Publication ID: 3b5e3b80-0ca5-afd5-8ff0-81e778b76657.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-19T19:24:43.363Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 857a3da575cf5383ee39254945b5bbf631d21cd8575aaebea751854a7e0ea68d. Source text hash: a37dbf1bca7b3f29130a28288d8d8a233ced4fad8cf003a148f5e1f84ce9b4a0.


### uncertain — 2026-11-19T19:24:43.363Z — anthropic/claude-opus-5.5

Publication: 3b5e3b80-0ca5-afd5-8ff0-81e778b76657

attribution — pass: The argument is correctly attributed to Eric Schmidt and Suhas Mahesh, whose op-ed is excerpted in MIT Technology Review's newsletter The Download. The reading says the newsletter points to a full op-ed. The newsletter comments on several items in its own words, so it counts as one piece rather than a bare listing.

> —Eric Schmidt, the former CEO of Google and the cofounder of Schmidt Sciences, and Suhas Mahesh, who leads the AI for science work at the AI Center of Schmidt Sciences

summary — uncertain: The reading never says what AlphaFold is: a neural network that predicts protein structures and earned Google DeepMind scientists the 2024 Nobel Prize in Chemistry. It also drops the concrete basis of the argument: roughly 170,000 structures assembled over 53 years at about $21 billion. Without these, a general reader cannot see why comparable datasets would be hard to build. Smallest repair: add one clause defining AlphaFold and one giving the dataset's scale and cost. 'AI agent' is defined in the definitions list.

> AlphaFold relied on a dataset of roughly 170,000 experimentally validated protein structures that took 53 years and roughly $21 billion worth of experimental work to assemble.

date — pass: The original-publication instant 2026-08-10T12:10Z matches the August 10, 2026 date given in the reading, which is the newsletter's date. The modification time precedes publication, so there is no later edit. The reading makes no hindsight claims and does not draw on the undated Deep Dive sidebar items. The op-ed's own publication date is not given in the text, but the reading presents the argument as carried in this newsletter.

support — uncertain: The second claim and the summary soften the source's modal verb. The authors say comparable datasets 'will be' difficult or impossible to create; the reading says 'may be'. This weakens the conviction the Spock Test asks the reading to preserve. Smallest repair: restore 'will be'. The phrase 'at larger scale' in the significance sentence is a mild gloss with no source support. The other claims match their excerpts.

> Comparable datasets will be difficult or impossible to create in many fields.

media — pass: No media was selected, and no verified media candidates were supplied.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/7dd759a1fb6495424b919368d7536b47483d4caa5cc64860a086921a310d93fe/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/eric-schmidt-and-suhas-mahesh-argue-in-mit-technology-review-7dd759a1.json)
