Some snippets from Anthropic’s RSI article for my own reference.
What they mean by “fully RSI” (#3 below) seems pretty conservative if you’ve mostly gotten the idea from (the memeplex that includes LW’s conception of it):
Possible futures
What happens next depends on two things: whether the trend continues, and what we choose to do if it does. We can imagine at least three future scenarios:
The trend stalls, but today’s AI capabilities are widely diffused. … We include this scenario for completeness, but we don’t believe it’s likely. Every capability we can measure, including those that feel “squishier,” like quality of code and success on open-ended tasks, has so far followed the same curve. We have not yet seen that curve bend. … We are more worried about the next two, which would move faster and leave far less room for preparation.
AI labs continue to see compounding efficiency gains. … The evidence we’ve laid out here suggests that we’re likely heading into this scenario. But speeding up one part of a process often just shifts the bottleneck elsewhere: overall pace is capped by the parts that haven’t sped up. … The rate at which organizations can spot and fix these bottlenecks may be a skill that improves over time, and it may become the most important skill for any organization.
AI systems themselves become capable of full recursive self-improvement, and begin building their successors. … Even if model development became fully automated and recursive, we can’t predict what that would mean for most humans’ daily lives. Amdahl’s law applies here as well. Recursive intelligence could lead to achieving many of the benefits outlined in Machines of Loving Grace, quickly in some domains. … But achieving recursive improvement alone does not suggest an immediate change in how industrial production occurs, societies organize, or markets function. More intelligence can’t learn what a drug does over decades of use, can’t hold elections sooner than a constitution dictates, and can’t turn a stranger into an old friend in a weekend. For most people, the felt pace of this future will still be set by the bottlenecks, even if the laboratory upstream runs at the speed of compute. That collision, where recursive intelligence building itself ever faster meets the world of humans, relationships, and governance, is another part of this future we can’t predict.
A research taste proxy graph. The jumps from Opus 4.5 (51%) to 4.6 (55%) to 4.7 (59%) are somewhat larger and steadier than I expected; naive extrapolation would place 4.8 closer to Mythos Preview than I’d guessed.
Claude is getting better at steering research sessions towards research findings. We examined real Claude Code sessions (between January and March 2026) where Anthropic researchers were working with Claude on an open-ended investigative problem, like figuring out why a training run kept crashing, or why a model scored poorly on a benchmark. In each case, we found a moment where the researcher took a detour: they pursued a direction that sent the session sideways before it eventually got back on track. We then showed various Claude models only the work from before the session went off-course and asked what it would do next. A separate Claude that was able to see how the session eventually turned out then judged whether the AI or the human suggested the better next step.8
(Fn: As a check on judge bias, we ran the same test on a separate set of 127 moments where the human’s next move was already strong (as opposed to the original set, where the human’s direction had room for improvement). There, the models’ suggestions were judged better only about 20% of the time.)
Because we deliberately picked moments (n=129) where we know the human’s choice had room for improvement, this isn’t a like-for-like comparison between model and human judgement. What these moments give us is a set of realistic, challenging situations where the right next step is not obvious, and where the human’s choice serves as a useful yardstick to compare model performance over time. On this measure, our best model in November 2025 (Opus 4.5) beat the human choice 51% of the time; in April 2026 (Mythos Preview), this grew to 64%. The day-to-day work of research is largely a chain of these next-step decisions, making this a relevant measure of the model’s ability to eventually run an investigation of its own. We view this result as an early signal that AI systems are getting better at making the kinds of judgement calls that AI research depends on.
How to read this: The practical ceiling line measures an “ideal” answer written by a model that could see the whole session (including how it ended).
An area of human comparative advantage, for now, is research taste and judgment, including choosing which problems matter, which results to trust, and when an approach is a dead end.
What if we’re wrong?
A natural objection to the evidence presented above is that the work that is still in human hands—choosing which problems to work on—is what matters most. Without that judgment, Claude is a capable assistant, but not a system that could drive AI progress on its own.
It is genuinely unclear whether today’s training methods and architectures could unlock that capacity. But AI is rarely advanced by “eureka!” moments. There have been a few of these in AI’s recent history, like the Transformer architecture, or mixture-of-experts models, but paradigm-shifting ideas arrive years apart. In between, most progress is incremental: we scale something up, see what breaks, fix it, and try again. That is exactly the kind of workflow Claude now excels at. Edison said that genius is 1% inspiration and 99% perspiration. But we see perspiration becoming increasingly automated. It’s becoming clear that much of what advances the frontier is automatable; large-scale research progress is mostly a function of tools and resources, which dictate how fast you can run experiments, how many you can run at once, and how quickly you can get results.
Even if we suppose that Claude never achieves good research taste, a conservative reading of our evidence still implies compounding acceleration. If humans spend most of their time on the single-digit fraction of work that is direction-setting, while Claude handles the rest, that means each engineer or researcher is steering far more work than before. The evidence we see suggests that people at Anthropic are both moving faster and covering a broader surface. In practice, this means that AI already makes Anthropic move much faster than it did before the advent of effective AI tools.
The less conservative reading is that the early evidence on Claude’s improving research judgment—narrow as it is today—is an indicator that this capability is improving as well. “Research taste” might be just another AI capability that AI systems fail at for a time, then get good at. We’ve seen a similar pattern with other qualitative skills, like AI systems being able to explain why a joke is funny, demonstrate theory of mind, and solve linguistic riddles.
Related: how AI Futures models automated research taste, survey data grounding, Oliver Sourbut’s model of research taste, some papers from which you could probably argue that (quote) “even models from about one year ago, with reasonable scaffolding/fine-tuning, seem already roughly in the range of a PhD student from a top institution on research taste, if not higher, in the ML research domain”.
How Anthropic thinks about “good code”, which may be different from others. Session success on open-ended problems is where Mythos vs Opus becomes particularly pronounced. Vibes-wise, Anthropic staff think Claude-written code was clearly worse than their own late last year, roughly at parity today, and expected to be better within the year.
The code that Claude writes is “good” and improving. “Good code” means two things: it works, and it is written in a manner that allows another engineer to understand it and build upon it. On the first criterion, the evidence is clear. The rate at which Anthropic staff correct, redirect, or take over mid-task from Claude has been falling steadily for a year, including on the most complex and open-ended tasks. This means problems with no clear specification, where the engineer isn’t sure what the answer looks like. This is evident in Claude’s success rate over time on tasks of different difficulties, as shown in the graph below. Claude writes code that works.
How to read this: Session success is determined by a Claude judge; a session is deemed successful if the Claude Code agent clearly succeeded at the user’s tasks without requiring corrections. Changes in workloads can lead to short-term fluctuations in success rates.
On the most open-ended tasks, Claude’s success rate reached 76% in May 2026, up 50 percentage points in six months. To give an example of tasks in this difficulty tier, a routine upgrade began crashing tens of thousands of training jobs. An engineer pointed Claude at the live incident with little more than some text content and cluster access. Working through the running jobs and testing one environment setting at a time, Claude isolated the single obscure debugging flag that was triggering the crash, reproduced it reliably, and confirmed a fix. In about two hours, Claude delivered what would normally be two to three days of work.
The second criterion is writing code that another engineer can understand and build on. Here the gap between humans and AI persists, but is closing fast. There isn’t full consensus among staff at Anthropic, but many believe that the Claude-written code was still worse in quality than human-written code at Anthropic in late 2025, and is roughly at parity today. We expect it to be better within the year.
This has changed the way that Anthropic now reviews its own code. Proposed changes to our codebase are now read by an automated Claude reviewer that looks for bugs, security flaws, and other defects before it can merge. Using this tool, we ran a retrospective analysis, and found that an automated Claude review of every change to our codebase would have caught roughly a third of the bugs behind past incidents on claude.ai before they ever reached production. The engineers who wrote that code are among the best in the world at building these systems. Claude is now catching the mistakes that they missed.
Productivity uplift:
In a March 2026 poll of 130 employees from across Anthropic research teams, the median respondent estimated that they produced around 4x as much output with Mythos Preview as they would have without access to any AI models, on the kinds of projects they would have been working on regardless.5 (Fn: Additional details on the methodology of this survey are discussed in section 2.3.5 of the Claude Opus 4.7 System Card.) We expect that the true degree of uplift in March was somewhat lower.6 (Fn: Many respondents may not have thought carefully about how to account for various biases or subtleties in the question definition, and recent research by METR shows that developer estimates of AI productivity uplift can be overestimated.) Nevertheless, we find the overall claim plausible, and in line with our other observations: a significant fraction of Anthropic technical staff is accomplishing their core work multiple times faster than they could without AI assistance.
We also see evidence that people at Anthropic are using Claude to do work that simply wouldn’t have happened otherwise, like building exploratory tooling and addressing long-deferred cleanup. For example, in April 2026, Claude shipped over 800 fixes that reduced a class of API errors by a factor of one thousand. The engineer overseeing Claude estimated that a human would have taken four years to complete this work; solving other people’s bugs is slow and painstaking, and humans struggle to hold that much unfamiliar context in their head at once.
I’m a little surprised by how conservative the article’s edge-case (not median) examples of productivity uplift are. Awhile back I collected various examples of tokenmaxxing and the SemiAnalysis ones struck me most; I was expecting more such examples in that tier here. The big caveat is that those examples came from Dylan Patel himself. Boris Cherny’s example struck me too, but I guess not everyone can context-switch like Boris.
Some quotes, which “reflect individual views as of May 2026, not official company positions”:
“Work (and life) ran on a gift economy of small favors between humans. ‘Can you help me get this script running?’ [...] each one created a little debt, a little mutual awareness. [Claude is] faster, it creates zero debt, but each of these is a lost bid for human collaboration.”
“On days where everything works well, I can’t help but think nothing I do matters, everything is automated and better and faster than I ever will be. But then there are days where everything breaks and I don’t understand why and I realize I have no idea what I’ve been up to anymore.”
Some snippets from Anthropic’s RSI article for my own reference.
What they mean by “fully RSI” (#3 below) seems pretty conservative if you’ve mostly gotten the idea from (the memeplex that includes LW’s conception of it):
A research taste proxy graph. The jumps from Opus 4.5 (51%) to 4.6 (55%) to 4.7 (59%) are somewhat larger and steadier than I expected; naive extrapolation would place 4.8 closer to Mythos Preview than I’d guessed.
Related: how AI Futures models automated research taste, survey data grounding, Oliver Sourbut’s model of research taste, some papers from which you could probably argue that (quote) “even models from about one year ago, with reasonable scaffolding/fine-tuning, seem already roughly in the range of a PhD student from a top institution on research taste, if not higher, in the ML research domain”.
How Anthropic thinks about “good code”, which may be different from others. Session success on open-ended problems is where Mythos vs Opus becomes particularly pronounced. Vibes-wise, Anthropic staff think Claude-written code was clearly worse than their own late last year, roughly at parity today, and expected to be better within the year.
Productivity uplift:
I’m a little surprised by how conservative the article’s edge-case (not median) examples of productivity uplift are. Awhile back I collected various examples of tokenmaxxing and the SemiAnalysis ones struck me most; I was expecting more such examples in that tier here. The big caveat is that those examples came from Dylan Patel himself. Boris Cherny’s example struck me too, but I guess not everyone can context-switch like Boris.
Some quotes, which “reflect individual views as of May 2026, not official company positions”: