Ajeya Cotra · Aug 14, 2026
Ajeya Cotra argues in Planned Obsolescence that selected AI task milestones are ahead of forecast
Markdown · JSON · Original source
Ajeya Cotra’s post was published on August 14, 2026. She revisits ten qualitative forecasts for AI capabilities and assesses their status. Cotra says math is well ahead of her predictions, logistics and video are roughly on track, and game play and game design are somewhat ahead. Some judgments remain ambiguous: she says end-to-end birthday-party planning might require more setup and inference spending than people would want, while the research-paper forecast depends partly on whether can explain their work well enough to write a paper. Cotra says the selected task milestones seem to be arriving 30%-50% faster than she predicted, but says she would not mechanically raise her probabilities for more extreme milestones. She argues that 2026’s progress so far does not refute the possibility of an intelligence explosion this year. Cotra says, “We are profoundly unprepared,” and says she is glad there is increasing energy to build the option to deliberately pace progress. Whether the larger milestones she lists will occur by the end of the year remains unresolved.
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
Ajeya Cotra · Article
Source details
Original source date: Aug 14, 2026 (2026-08-14T14:02:22+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-06T05:53:23.136Z. First observed: 2026-10-27T23:05:56.956Z. Retrieved: 2026-10-27T23:07:49.351Z.
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 27, 2026; originally published August 14, 2026. Cotra describes judgments about selected qualitative task forecasts; the post does not establish that these tasks represent AI capabilities broadly or resolve whether the larger milestones will occur.
Open question
What evidence would show that faster progress on these selected tasks should change forecasts for the larger milestones Cotra lists?
Source-stated
Cotra says AI capability milestones across the categories she reviewed appear to be arriving faster than she predicted.
Across all these categories, it feels like we’re hitting task milestones somewhat (30%-50%) faster than I predicted at the beginning of the year.
Source-stated
Cotra judges progress in math to be ahead of her predictions, logistics and video roughly on track, and game play and game design somewhat ahead.
Overall, math feels clearly well ahead of my predictions, logistics and video feel roughly on track, and game play and game design feel a bit ahead.
Source-stated
Cotra says the apparent pace of task milestones does not by itself justify mechanically increasing her probabilities for extreme milestones.
I don’t necessarily think that I should mechanically increase my probability on these extreme milestones because the earlier ones appear to have arrived somewhat faster than I thought.
Source-stated
Cotra says progress so far does not refute the possibility of an intelligence explosion in 2026 and says society is unprepared.
But nothing about the way 2026 has played out so far refutes the basic point that the intelligence explosion really could be this year. We are profoundly unprepared
Source-stated
Cotra says she is glad there is increasing energy to build the option to deliberately pace frontier AI progress.
I am very glad that there is increasing energy to build the option to deliberately pace frontier AI progress.
Source-stated
Cotra says whether AI can write a publishable research paper remains ambiguous, in part because people questioned whether agents could explain their research well enough.
multiple people said they didn’t think AI agents would be able to explain their research well enough to autonomously write the paper for it. That makes this prediction ambiguous
Source and contributor context
Ajeya Cotra · 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.
Frontier model
One of the most capable AI models available at a given time; the set shifts as models improve.
Editorial provenance
Version 1. Publication ID: c9ce2052-c272-a480-e6c1-143a9b8cf64a.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-06T05:50:13.272Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 609b27803ea5f7ea7388f2d0020a34c662d5c0018242723251362b366d15b834. Source text hash: 1e3af066e054cddb98215ba6a3913dae7a01be9f04f29c0696cf0df67e4f1f65.
Revision and source-check history
Version 1 · 2026-11-06T05:53:23.136Z
Ajeya Cotra argues in Planned Obsolescence that selected AI task milestones are ahead of forecast
Ajeya Cotra’s post was published on August 14, 2026. She revisits ten qualitative forecasts for AI capabilities and assesses their status. Cotra says math is well ahead of her predictions, logistics and video are roughly on track, and game play and game design are somewhat ahead. Some judgments remain ambiguous: she says end-to-end birthday-party planning might require more setup and inference spending than people would want, while the research-paper forecast depends partly on whether can explain their work well enough to write a paper. Cotra says the selected task milestones seem to be arriving 30%-50% faster than she predicted, but says she would not mechanically raise her probabilities for more extreme milestones. She argues that 2026’s progress so far does not refute the possibility of an intelligence explosion this year. Cotra says, “We are profoundly unprepared,” and says she is glad there is increasing energy to build the option to deliberately pace progress. Whether the larger milestones she lists will occur by the end of the year remains unresolved.
Original source
Ajeya Cotra · Article
Source details
Original source date: Aug 14, 2026 (2026-08-14T14:02:22+00:00).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-06T05:53:23.136Z. First observed: 2026-10-27T23:05:56.956Z. Retrieved: 2026-10-27T23:07:49.351Z.
Limits
Text as retrieved on October 27, 2026; originally published August 14, 2026. Cotra describes judgments about selected qualitative task forecasts; the post does not establish that these tasks represent AI capabilities broadly or resolve whether the larger milestones will occur.
Open question
What evidence would show that faster progress on these selected tasks should change forecasts for the larger milestones Cotra lists?
Source-stated
Cotra says AI capability milestones across the categories she reviewed appear to be arriving faster than she predicted.
Across all these categories, it feels like we’re hitting task milestones somewhat (30%-50%) faster than I predicted at the beginning of the year.
Source-stated
Cotra judges progress in math to be ahead of her predictions, logistics and video roughly on track, and game play and game design somewhat ahead.
Overall, math feels clearly well ahead of my predictions, logistics and video feel roughly on track, and game play and game design feel a bit ahead.
Source-stated
Cotra says the apparent pace of task milestones does not by itself justify mechanically increasing her probabilities for extreme milestones.
I don’t necessarily think that I should mechanically increase my probability on these extreme milestones because the earlier ones appear to have arrived somewhat faster than I thought.
Source-stated
Cotra says progress so far does not refute the possibility of an intelligence explosion in 2026 and says society is unprepared.
But nothing about the way 2026 has played out so far refutes the basic point that the intelligence explosion really could be this year. We are profoundly unprepared
Source-stated
Cotra says she is glad there is increasing energy to build the option to deliberately pace frontier AI progress.
I am very glad that there is increasing energy to build the option to deliberately pace frontier AI progress.
Source-stated
Cotra says whether AI can write a publishable research paper remains ambiguous, in part because people questioned whether agents could explain their research well enough.
multiple people said they didn’t think AI agents would be able to explain their research well enough to autonomously write the paper for it. That makes this prediction ambiguous
Source and contributor context
Ajeya Cotra · 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.
Frontier model
One of the most capable AI models available at a given time; the set shifts as models improve.
Editorial provenance
Version 1. Publication ID: c9ce2052-c272-a480-e6c1-143a9b8cf64a.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-06T05:50:13.272Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 609b27803ea5f7ea7388f2d0020a34c662d5c0018242723251362b366d15b834. Source text hash: 1e3af066e054cddb98215ba6a3913dae7a01be9f04f29c0696cf0df67e4f1f65.
uncertain · 2026-11-06T05:50:13.272Z · anthropic/claude-opus-5.5
Publication: c9ce2052-c272-a480-e6c1-143a9b8cf64a
attribution — pass: The post is correctly attributed to Ajeya Cotra on her Planned Obsolescence page. The judgments, forecasts and the 'profoundly unprepared' line stay framed as her own statements.
Hurtling through 2026 Scoring ten qualitative forecasts five months early Ajeya Cotra
summary — uncertain: The reading is broadly faithful but hard to follow without background. 'Intelligence explosion' is central to Cotra's conclusion, yet it is neither explained nor in the definitions. 'Inference spending' is also left unexplained. Readers also learn only category labels, not what the forecasts were: concrete tasks like beating Pokemon, IMO problem 6 or planning a wedding, set in January as 80%/20% low- and high-end bets. The summary also omits her view that the paper forecast feels beaten in spirit. Smallest repair: briefly gloss these terms, add one or two example tasks with the 80/20 framing, and note the beaten-in-spirit view.
That makes this prediction ambiguous,5 but given that I was mostly thinking about the math itself at the time, it feels like it’s been beaten in spirit.
date — pass: The stated date, August 14, 2026, matches the dateline and the instant in evidence.originalPublished. modifiedAt is the same as the publication time. The page is a publisher page, not an archive copy. The year-end milestones are kept as unresolved forecasts, with no hindsight framing.
Ajeya Cotra Aug 14, 2026
support — pass: Each claim matches its excerpt. This covers the category judgments, the 30%-50% pace, her refusal to mechanically update, the intelligence-explosion line, the pacing line and the ambiguity of the paper forecast. Calling 'we' society is a mild gloss. The question and limitations add no unsupported assumptions.
Across all these categories, it feels like we’re hitting task milestones somewhat (30%-50%) faster than I predicted at the beginning of the year.
media — pass: No media is selected, and no verified candidates were supplied, so that choice is appropriate.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
An AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceOne of the most capable AI models available at a given time; the set shifts as models improve.
Definition referenceAn AI system that chooses and carries out actions toward a goal, often using software tools.
Definition referenceOne of the most capable AI models available at a given time; the set shifts as models improve.
Definition reference