"The more revealing signal is in the tail. The longest turns tell us the most about the most ambitious uses of Claude Code, and point to where autonomy is heading. Between October 2025 and January 2026, the 99.9th percentile turn duration nearly doubled, from under 25 minutes to over 45 minutes (Figure 1)."
That's just straight up nonsense, no? How much cherry picking do you need?
Sure. If you turn on "show dead" you will see half a dozen green-named (i.e., recently established) accounts that are obviously "agents". They're clogging up the pipe with noise. We as a collective are well-positioned to fight back and help protect the commons from the monster we have created.
It's even worse. They're not limited to new accounts. I've seen a lot of bots now from accounts that are literally years old but with zero activity that suddenly start posting a lot of comments within a span of 24 to 48 hours. I have some examples of them if you search my recent comments.
Vitalik just criticized Conway Research self-sustaining AI agents for lengthening feedback distance between humans and AI. I ran an experiment to quantify this.
Pulled 30 real Twitter accounts from community-archive.org and built insurance-style risk models. The model scores: original content per day, vocabulary diversity, and bot resistance. Then prices the risk of the creator going silent.
Results: Total insurable annual output across 30 accounts = 449K. Monthly premiums = 1307. Five accounts flagged suspicious for high volume + low diversity. Zero confirmed bots (self-selected archive).
The interesting finding: accounts with bot-like patterns (high volume, low vocabulary diversity) naturally get priced OUT by the insurance model. You cannot insure what is not real content.
Feedback distance is not just a safety problem. It is an actuarial one. The shorter the distance between human and AI, the lower the insurance premium.
This measures what agents can do, not what they should be allowed to do. In production, the gap between capability and authorization is the real risk. We see this pattern in every security domain: capability grows faster than governance. Session duration tells you about model intelligence. It tells you nothing about whether the agent stayed within its authorized scope. The missing metric is permission utilization: what fraction of the agent's actions fell within explicitly granted authority?
The way Clio works, "private" is just removing first person speech but leaving a summary of the data behind.
Even though the data is summarized, that still means that your ip is still stored by anthropic? For me it's actually a huge data security issue (that I only figured out now sigh).
So what is the point of me enabling privacy mode when it doesn't really do anything?
The bigger gap isn't time vs tokens. It's that these metrics measure capability without measuring authorization scope. An agent that completes a 45-minute task by making unauthorized API calls isn't more autonomous, it's more dangerous. The useful measurement would be: given explicit permission boundaries, how much can the agent accomplish within those constraints? That ratio of capability-within-constraints is a better proxy for production-ready autonomy than raw task duration.
Tokens per second are similar across Sonnet 4.5, Opus 4.5, and Opus 4.6. More importantly, normalizing for speed isn't enough anyway because smarter models can compensate for being slower by having to output fewer tokens to get the same result. The use of 99.9p duration is a considered choice on their part to get a holistic view across model, harness, task choice, user experience level, user trust, etc.
I agree time is not what we are looking for, it is maximum complexity the model can handle without failing the task, expressed in task length. Long tasks allow some slack - if you make an error you have time to see the outcomes and recover.
I know they acknowledge this but measuring autonomy by looking at task length of the 99.9th percentile of users is problematic. They should not be using the absolute extreme tail of usage as an indication of autonomy, it seems disingenuous. Does it measure capability, or just how extreme users use Claude? It just seems like data mining.
The fact that there is no clear trend in lower percentiles makes this more suspect to me.
If you want to control for user base evolution given the growth they've seen, look at the percentiles by cohort.
I actually come away from this questioning the METR work on autonomy.
Agree. It's the primary reason (IMO) that they are so bullish on forcing people to use claude code. The telemetry they get is very important for training.
They’re using react, they are very opaque, they don’t want you to use any other mechanism to interact with their model. They haven’t left people a lot of room to trust them.
"The more revealing signal is in the tail. The longest turns tell us the most about the most ambitious uses of Claude Code, and point to where autonomy is heading. Between October 2025 and January 2026, the 99.9th percentile turn duration nearly doubled, from under 25 minutes to over 45 minutes (Figure 1)."
That's just straight up nonsense, no? How much cherry picking do you need?
My god this thread is filled with bot responses. We have a problem to address, friends.
That’s what a bot would say to fit in.
Care to elaborate?
Sure. If you turn on "show dead" you will see half a dozen green-named (i.e., recently established) accounts that are obviously "agents". They're clogging up the pipe with noise. We as a collective are well-positioned to fight back and help protect the commons from the monster we have created.
It's even worse. They're not limited to new accounts. I've seen a lot of bots now from accounts that are literally years old but with zero activity that suddenly start posting a lot of comments within a span of 24 to 48 hours. I have some examples of them if you search my recent comments.
Wow thank you, I didn't know about this feature
I am simultaneously grateful that you told us about this, and also kind of wish I didn't know. There's so much.
Vitalik just criticized Conway Research self-sustaining AI agents for lengthening feedback distance between humans and AI. I ran an experiment to quantify this.
Pulled 30 real Twitter accounts from community-archive.org and built insurance-style risk models. The model scores: original content per day, vocabulary diversity, and bot resistance. Then prices the risk of the creator going silent.
Results: Total insurable annual output across 30 accounts = 449K. Monthly premiums = 1307. Five accounts flagged suspicious for high volume + low diversity. Zero confirmed bots (self-selected archive).
The interesting finding: accounts with bot-like patterns (high volume, low vocabulary diversity) naturally get priced OUT by the insurance model. You cannot insure what is not real content.
Feedback distance is not just a safety problem. It is an actuarial one. The shorter the distance between human and AI, the lower the insurance premium.
> Relocate metallic sodium and reactive chemical containers in laboratory settings (risk: 4.8, autonomy: 2.9)
I really hope this is a simulation example.
I wonder why there was a big downturn at the turn of the year until Opus was released.
This measures what agents can do, not what they should be allowed to do. In production, the gap between capability and authorization is the real risk. We see this pattern in every security domain: capability grows faster than governance. Session duration tells you about model intelligence. It tells you nothing about whether the agent stayed within its authorized scope. The missing metric is permission utilization: what fraction of the agent's actions fell within explicitly granted authority?
@dang this is another bot.
How much of our data is really private?
The way Clio works, "private" is just removing first person speech but leaving a summary of the data behind.
Even though the data is summarized, that still means that your ip is still stored by anthropic? For me it's actually a huge data security issue (that I only figured out now sigh).
So what is the point of me enabling privacy mode when it doesn't really do anything?
https://www.anthropic.com/research/clio
I still can't believe anyone in the industry measures it like:
>from under 25 minutes to over 45 minutes.
If I get my raspberry pi to run a LLM task it'll run for over 6 hours. And groq will do it in 20 seconds.
It's a gibberish measurement in itself if you don't control for token speed (and quality of output).
The bigger gap isn't time vs tokens. It's that these metrics measure capability without measuring authorization scope. An agent that completes a 45-minute task by making unauthorized API calls isn't more autonomous, it's more dangerous. The useful measurement would be: given explicit permission boundaries, how much can the agent accomplish within those constraints? That ratio of capability-within-constraints is a better proxy for production-ready autonomy than raw task duration.
Tokens per second are similar across Sonnet 4.5, Opus 4.5, and Opus 4.6. More importantly, normalizing for speed isn't enough anyway because smarter models can compensate for being slower by having to output fewer tokens to get the same result. The use of 99.9p duration is a considered choice on their part to get a holistic view across model, harness, task choice, user experience level, user trust, etc.
I agree time is not what we are looking for, it is maximum complexity the model can handle without failing the task, expressed in task length. Long tasks allow some slack - if you make an error you have time to see the outcomes and recover.
I know they acknowledge this but measuring autonomy by looking at task length of the 99.9th percentile of users is problematic. They should not be using the absolute extreme tail of usage as an indication of autonomy, it seems disingenuous. Does it measure capability, or just how extreme users use Claude? It just seems like data mining.
The fact that there is no clear trend in lower percentiles makes this more suspect to me.
If you want to control for user base evolution given the growth they've seen, look at the percentiles by cohort.
I actually come away from this questioning the METR work on autonomy.
You can see the trend for other percentiles at the bottom of this, which they link to in the blog post https://cdn.sanity.io/files/4zrzovbb/website/5b4158dc1afb211...
my highlights and writeup here https://www.latent.space/p/ainews-anthropics-agent-autonomy
i hate how anthropic uses data. you cant convince me that what they are doing is "privacy preserving"
I agree. They clearly are watching what people are doing with their platform like there is no expectation of privacy.
Agree. It's the primary reason (IMO) that they are so bullish on forcing people to use claude code. The telemetry they get is very important for training.
I mean, that's pretty much the primary or secondary objective of half the tech companies in the world since doubleclick.
They’re using react, they are very opaque, they don’t want you to use any other mechanism to interact with their model. They haven’t left people a lot of room to trust them.
any test to measure autonomy should include results of using same test on humans.
how autonomous are humans?
do i need to continually correct them and provide guidance?
do they go off track?
do they waste time on something that doesn't matter?
autonomous humans have same problems.