# Ramez Naam argues that agent swarms may add speed faster than intelligence [Canonical reading](https://escapemeridian.com/readings/f44f6c53e679fc7ddcd957e046d2d2506d082a568abcecfc7ec6dfb5f16fa5a5) Ramez Naam argues that large teams of AI agents can speed up parallelizable work, but current experiments show much smaller gains in harder problem-solving as teams grow. He points to examples where performance gains flatten sharply, including a formal-math benchmark where expanding from ten Opus 5.5 agents to a hundred added only two score points. He cautions that benchmark scores are task-specific, and that the experiments do not establish why gains flatten or how well future systems might scale. The distinction matters for recursive self-improvement (RSI): the idea that AI helps develop improved AI. Naam’s view is that more copies of today’s models may help search or execution, but may not supply the conceptual breakthroughs needed for an accelerating feedback loop. He says the evidence is limited and proposes tracking verified performance, speed relative to compute, and where gains weaken as team sizes increase. 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 Noah Smith · Article Original source date: Oct 19, 2026 (2026-10-19T04:33:30+00:00). Date basis: Date stated by the publisher. Added to Escape Meridian: 2026-10-19T09:02:10.506Z. First observed: 2026-10-19T04:41:22.287Z. Retrieved: 2026-10-19T09:01:20.050Z. - [Can a Million AI Agents Create Superintelligence?](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Limits The article is a guest post by Ramez Naam on Noah Smith’s blog; the opening commentary is by Smith, while the main argument and analysis are presented as Naam’s. The article says the evidence base is limited, and many cited results are specific benchmarks rather than direct measures of AI research or general intelligence. The reported benchmark gains do not establish why scaling flattens or whether future models and methods will change the pattern. ## Open question Will future AI models produce larger verified problem-solving gains as agent teams grow, especially on tasks requiring conceptual breakthroughs? ## Source-stated Naam argues that parallel agents appear more useful for speeding up work than for increasing the difficulty of problems a team can solve. > So far, adding agents seems much more effective at speeding up execution than at raising the level of problem a team can solve. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source-stated In one formal-math benchmark, the first expansion from one to ten agents added 27 score points, while expanding from ten to a hundred added two. > The first tenfold expansion adds 27 score points; the next adds just two. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source-stated Naam argues that current evidence does not show that adding copies of today’s models can reliably produce the conceptual breakthroughs that recursive self-improvement may require. > Adding more copies of today’s models hasn’t shown the sustained intelligence gains that may be needed for repeated conceptual breakthroughs. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source-stated Naam says the evidence is limited and that future measurements should test how verified performance, speed-cost tradeoffs, and flattening vary with team size. > As we discussed in the previous post, we’re going off of scant data here. This model could be wrong. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source and contributor context Noah Smith · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media was declared by the source, but no asset passed the supported availability checks. Open the original source to view it. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## Ramez Naam Explicitly tests self-improvement and diminishing-return assumptions; connects AI to energy and economics. - [Explore this context](/entities/ramez-naam) - [Reference](https://www.diamandis.com/podcast/ramez-naam) - [Reference](https://www.rameznaam.com/p/ai-rsi-isnt-leading-to-super-intelligence) - [Reference](https://rameznaam.com/) ## Noah Smith Economic interpretation can separate technological usefulness from financing and deployment dynamics. - [Explore this context](/entities/noah-smith) - [Reference](https://lexfridman.com/ezra-klein-and-derek-thompson-transcript) - [Reference](https://www.dwarkesh.com/p/notes-on-china) - [Reference](https://www.dwarkesh.com/p/patrick-collison) - [Reference](https://www.noahpinion.blog/p/ai-vs-the-railroads) - [Reference](https://www.noahpinion.blog/) ## Editorial provenance Version 1. Publication ID: db60d4f0-daf9-44d2-8246-d78ea3107cd3. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-19T09:02:10.506Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 4ae17f9a7adb8a9f09ef1acfaceaaef7f9a78d7919dba4d0e3bcccc762b2a787. Source text hash: 99e316bda8c2a8cce45f6d26f4088bc6f6f7bb90d6abb11b589bba297f801acb. ## Revision and source-check history ### Version 1 — 2026-10-19T09:02:10.506Z Ramez Naam argues that large teams of AI agents can speed up parallelizable work, but current experiments show much smaller gains in harder problem-solving as teams grow. He points to examples where performance gains flatten sharply, including a formal-math benchmark where expanding from ten Opus 5.5 agents to a hundred added only two score points. He cautions that benchmark scores are task-specific, and that the experiments do not establish why gains flatten or how well future systems might scale. The distinction matters for recursive self-improvement (RSI): the idea that AI helps develop improved AI. Naam’s view is that more copies of today’s models may help search or execution, but may not supply the conceptual breakthroughs needed for an accelerating feedback loop. He says the evidence is limited and proposes tracking verified performance, speed relative to compute, and where gains weaken as team sizes increase. ## Original source Noah Smith · Article Original source date: Oct 19, 2026 (2026-10-19T04:33:30+00:00). Date basis: Date stated by the publisher. Added to Escape Meridian: 2026-10-19T09:02:10.506Z. First observed: 2026-10-19T04:41:22.287Z. Retrieved: 2026-10-19T09:01:20.050Z. - [Can a Million AI Agents Create Superintelligence?](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Limits The article is a guest post by Ramez Naam on Noah Smith’s blog; the opening commentary is by Smith, while the main argument and analysis are presented as Naam’s. The article says the evidence base is limited, and many cited results are specific benchmarks rather than direct measures of AI research or general intelligence. The reported benchmark gains do not establish why scaling flattens or whether future models and methods will change the pattern. ## Open question Will future AI models produce larger verified problem-solving gains as agent teams grow, especially on tasks requiring conceptual breakthroughs? ## Source-stated Naam argues that parallel agents appear more useful for speeding up work than for increasing the difficulty of problems a team can solve. > So far, adding agents seems much more effective at speeding up execution than at raising the level of problem a team can solve. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source-stated In one formal-math benchmark, the first expansion from one to ten agents added 27 score points, while expanding from ten to a hundred added two. > The first tenfold expansion adds 27 score points; the next adds just two. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source-stated Naam argues that current evidence does not show that adding copies of today’s models can reliably produce the conceptual breakthroughs that recursive self-improvement may require. > Adding more copies of today’s models hasn’t shown the sustained intelligence gains that may be needed for repeated conceptual breakthroughs. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source-stated Naam says the evidence is limited and that future measurements should test how verified performance, speed-cost tradeoffs, and flattening vary with team size. > As we discussed in the previous post, we’re going off of scant data here. This model could be wrong. - [Supporting source](https://www.noahpinion.blog/p/can-a-million-ai-agents-create-superintelligence) ## Source and contributor context Noah Smith · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media was declared by the source, but no asset passed the supported availability checks. Open the original source to view it. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## Ramez Naam Explicitly tests self-improvement and diminishing-return assumptions; connects AI to energy and economics. - [Explore this context](/entities/ramez-naam) - [Reference](https://www.diamandis.com/podcast/ramez-naam) - [Reference](https://www.rameznaam.com/p/ai-rsi-isnt-leading-to-super-intelligence) - [Reference](https://rameznaam.com/) ## Noah Smith Economic interpretation can separate technological usefulness from financing and deployment dynamics. - [Explore this context](/entities/noah-smith) - [Reference](https://lexfridman.com/ezra-klein-and-derek-thompson-transcript) - [Reference](https://www.dwarkesh.com/p/notes-on-china) - [Reference](https://www.dwarkesh.com/p/patrick-collison) - [Reference](https://www.noahpinion.blog/p/ai-vs-the-railroads) - [Reference](https://www.noahpinion.blog/) ## Editorial provenance Version 1. Publication ID: db60d4f0-daf9-44d2-8246-d78ea3107cd3. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-19T09:02:10.506Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 4ae17f9a7adb8a9f09ef1acfaceaaef7f9a78d7919dba4d0e3bcccc762b2a787. Source text hash: 99e316bda8c2a8cce45f6d26f4088bc6f6f7bb90d6abb11b589bba297f801acb. ### passed — 2026-10-19T09:02:10.506Z — anthropic/claude-opus-5.5 Publication: db60d4f0-daf9-44d2-8246-d78ea3107cd3 attribution — pass: The argument is correctly attributed to Ramez Naam. The limitations correctly note that this is a guest post on Noah Smith's blog and that Smith wrote the opening commentary. > In this post, he expresses skepticism about the potential for agent swarms — large numbers of AI agents working in parallel — to offer dramatic increases in AI capabilities. summary — pass: The title, summary and question faithfully convey Naam's distinction between the speed gains and the intelligence gains that more agents provide. RSI is explained in plain language, and the caveats about scant data and task-specific benchmarks are preserved. The question is a genuine open issue that the source itself raises. > So far, adding agents seems much more effective at speeding up execution than at raising the level of problem a team can solve. date — pass: originalPublished is 2026-10-19, which matches the byline date on the page and the feed entry. Nothing in the source indicates an earlier first publication of this post. > Ramez Naam Oct 19, 2026 support — pass: Each claim is directly supported by the source: speed versus capability, the 27-point and then 2-point gains on the formal-math benchmark, the lack of demonstrated conceptual breakthroughs, and the limited data with proposed metrics. There are no unsupported benefits or costs. > The first tenfold expansion adds 27 score points; the next adds just two. media — pass: The selected media is the verified direct link to the original article. It is the most direct way for a reader to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/f44f6c53e679fc7ddcd957e046d2d2506d082a568abcecfc7ec6dfb5f16fa5a5/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/f44f6c53e679fc7ddcd957e046d2d2506d082a568abcecfc7ec6dfb5f16fa5a5.json)