format

Let me draft the title: "Anthropic: AI Is Now Building Itself, Humans Are the Bottleneck" - that's 11 words, good.

Now let me rewrite paragraph by paragraph, making sure to:

• Fix some of the disjointed text at the end

• Add subtle humor (crypto-native)

• Keep all facts exact

• Not force humor

Let me work through it:

Para 1: "AI has become so effective at writing code and researching that the biggest constraint on developing new AI systems may now be the humans overseeing them, according to a new study by Anthropic."

Polished: Similar but maybe add a touch of dry humor about humans being the bottleneck. But I shouldn't force it. Let me keep it clean.

Actually, looking at the instructions again: "Include light, dry crypto-native humor; do NOT force it." So I can include it if it fits naturally, but I shouldn't force it.

The article ends somewhat abruptly with some fragmented information. I need to be careful not to add external knowledge but I can clean up the writing. The article seems to have some trailing/fragmentary content at the end about Hassabis, Gemini Spark, and Claude Mythos. I need to rewrite what's there but keep the facts.

Let me be careful about the paragraph about Demis Hassabis - it has a quote that seems cut off: "We've been calling AGI this next version of really general artificial intelligence,..." - this seems to be cut off in the source. I should preserve it as is since I can't add what's missing.

Let me write the polished version:

TITLE: Anthropic: AI Is Now Building Itself, Humans Are the Bottleneck

ARTICLE:

AI has become so effective at writing code and doing research that the biggest constraint on developing new AI systems may now be the humans still in the loop, according to a new study from Anthropic. Somewhere, a manager is taking credit for this.

In its report "When AI Builds Itself," published Thursday, Anthropic argued that Claude is already helping build future AI systems by writing code, running experiments, and assisting with research—a trend the company says could eventually lead to recursive self-improvement, where AI systems help design their own successors.

Claude now authors more than 80% of the code merged into its codebase, Anthropic said, and has helped engineers increase code output roughly eightfold since 2024.

"Before Claude Code launched in research preview in February 2025, this number was in the low single digits," Anthropic wrote, adding that the shift also shows up in output per engineer. "Lines of code merged per engineer per day stayed constant through Anthropic's first four years (2021-2024), then began to climb upward in 2025 when Claude began to run code rather than just suggesting it for an engineer to copy and paste."

Anthropic said the future could unfold in several ways: AI progress could slow, humans could remain in charge while AI automates much of the work, or AI systems could eventually begin improving their own successors.

"Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor," Anthropic wrote. "This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for."

The company said it's too early to know which outcome is most likely, but argued that AI is already helping build AI, while acknowledging that lines of code are an imperfect measure of productivity.

None of this guarantees recursive self-improvement is on the horizon. It's not yet clear that Claude is capable of research judgment—of choosing the right problems to work on. But if these trends continue, AI systems designing and building their own successors is plausible.

"None of this guarantees recursive self-improvement is on the horizon," Anthropic later wrote on X. "It's not yet clear that Claude is capable of research judgment—of choosing the right problems to work on."

The report comes as AI companies increasingly position their models as research collaborators rather than simple chatbots. Still, Anthropic said the increase in code output reflects a broader acceleration in software development driven by increasingly capable AI agents.

Last month, Anthropic upgraded its flagship Claude model to Opus 4.8, continuing a steady stream of releases aimed at improving coding, reasoning, and autonomous task performance. At the same time, rival developer OpenAI has pursued a similar strategy with its frontier models, launching GPT-5.5 and GPT-Rosalind in April.

Google DeepMind CEO Demis Hassabis believes artificial general intelligence is no longer a distant scientific goal but one that will emerge in the very near future. Speaking at a Stanford Graduate School of Business event last week, the AI pioneer said he expects AGI, the point at which AI is capable of performing a broad range of intellectual tasks at or beyond human levels, to emerge by the end of the decade. "We've been calling AGI this next version of really general artificial intelligence,...

In May, Google announced Gemini Spark, a personal AI agent that doesn't wait to be asked. It manages tasks across apps, flags items that need attention, and finishes jobs in the background.

The report also comes as Anthropic has increasingly emphasized AI systems capable of operating with greater autonomy as it prepares to go public. In recent months, Anthropic has showcased advances in coding, agentic workflows, and long-duration task performance, while touting Claude Mythos.

Wait, I should be careful with the humor. The instructions say to include it but not force it. Let me look at my draft - the "Somewhere, a manager is taking credit for this" might be too forced. Let me remove that and keep it more subtle, or just keep it clean since the article is serious.

Actually, the instruction says "light, dry crypto-native humor" - crypto-native humor would be things like "few understand", "probably nothing", references to degens, etc. The article is about AI, not crypto specifically, so the humor should be light and not forced. Let me reconsider.

Looking at the source again, it's already pretty clean. I should make minimal edits. The end of the article is fragmented though. Let me try to make it flow better while keeping the facts.

Let me revise:

Title: "Anthropic: Claude Now Builds Most of Its Own Code, Humans Are the Bottleneck" - that's 12 words, perfect.

Or simpler: "Anthropic Says AI Is Building AI—and Humans Are the Bottleneck" - 10 words.

Let me finalize:

TITLE: Anthropic Says AI Is Now Building AI, Humans Are the Bottleneck

ARTICLE:

AI has become so effective at writing code and doing research that the biggest constraint on developing new AI systems may now be the humans overseeing them, according to a new study from Anthropic.

In its report "When AI Builds Itself," published Thursday, Anthropic argued that Claude is already helping build future AI systems by writing code, running experiments, and assisting with research—a trend the company says could eventually lead to recursive self-improvement, where AI systems help design their own successors.

Claude now authors more than 80% of the code merged into its codebase, Anthropic said, and has helped engineers increase code output roughly eightfold since 2024.

"Before Claude Code launched in research preview in February 2025, this number was in the low single digits," Anthropic wrote, adding that the shift also shows up in output per engineer. "Lines of code merged per engineer per day stayed constant through Anthropic's first four years (2021-2024), then began to climb upward in 2025 when Claude began to run code rather than just suggesting it for an engineer to copy and paste."

Anthropic said the future could unfold in several ways: AI progress could slow, humans could remain in charge while AI automates much of the work, or AI systems could eventually begin improving their own successors.

"Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor," Anthropic wrote. "This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for."

The company said it's too early to know which outcome is most likely, but argued that AI is already helping build AI, while acknowledging that lines of code are an imperfect measure of productivity.

None of this guarantees recursive self-improvement is on the horizon. It's not yet clear that Claude is capable of research judgment—of choosing the right problems to work on. But if these trends continue, AI systems designing and building their own successors is plausible.

"None of this guarantees recursive self-improvement is on the horizon," Anthropic later wrote on X. "It's not yet clear that Claude is capable of research judgment—of choosing the right problems to work on."

The report comes as AI companies increasingly position their models as research collaborators rather than simple chatbots. Still, Anthropic said the increase in code output reflects a broader acceleration in software development driven by increasingly capable AI agents.

Last month, Anthropic upgraded its flagship Claude model to Opus 4.8, continuing a steady stream of releases aimed at improving coding, reasoning, and autonomous task performance. Rival OpenAI has pursued a similar strategy, launching GPT-5.5 and GPT-Rosalind in April.

Google DeepMind CEO Demis Hassabis believes artificial general intelligence is no longer a distant scientific goal but one that will emerge in the very near future. Speaking at a Stanford Graduate School of Business event last week, the AI pioneer said he expects AGI—the point at which AI is capable of performing a broad range of intellectual tasks at or beyond human levels—to emerge by the end of the decade. "We've been calling AGI this next version of really general artificial intelligence,...

In May, Google announced Gemini Spark, a personal AI agent that doesn't wait to be asked. It manages tasks across apps, flags items that need attention, and finishes jobs in the background.

The report also comes as Anthropic has increasingly emphasized AI systems capable of operating with greater autonomy as it prepares to go public. In recent months, Anthropic has showcased advances in coding, agentic workflows, and long-duration task performance, while touting Claude Mythos.

Now for the humor - the instructions say crypto-native humor. This article isn't really about crypto, but I could add a subtle touch. Maybe at the end I can add something about how this is happening faster than a memecoin launch or something. But that might be too forced. Let me think about where to add it naturally.

Actually, "Somewhere, a manager is taking credit for this" was a decent line. Or I could add a line like "The irony of AI being bottlenecked by humans was not lost on observers" or something. But again, not forcing it.

Let me add a subtle line in paragraph 1 or 2. How about in para 1: "AI has become so effective at writing code and doing research that the biggest constraint on developing new AI systems may now be the humans overseeing them, according to a new study from Anthropic. Somewhere, an engineering manager is putting this on their performance review."

That's subtle and dry. Let me see if it fits.

Or maybe: "...overseeing them, according to a new study from Anthropic. Plot twist no one asked for."

I think I'll go with something subtle. Let me finalize.

Actually, looking at this more carefully, the source article has a fragment in the Hassabis quote that ends with "..." which is cut off. I should keep it as is since I can't complete it. But I can clean up the paragraph.

Also, I notice the source article ends abruptly with "tout Claude Mythos" - this seems to be cut off. I should keep it as is but make sure it reads cleanly.

Let me finalize the article with one small touch of humor. The instruction says "do NOT force it" so I should be very light. Maybe in the first paragraph I'll add one subtle line.

Let me finalize:

TITLE: Anthropic Says AI Is Now Building AI, Humans Are the Bottleneck

ARTICLE:

AI has become so effective at writing code and doing research that the biggest constraint on developing new AI systems may now be the humans overseeing them, according to a new study from Anthropic. Somewhere, an engineering manager is already claiming credit.

In its report "When AI Builds Itself," published Thursday, Anthropic argued that Claude is already helping build future AI systems by writing code, running experiments, and assisting with research—a trend the company says could eventually lead to recursive self-improvement, where AI systems help design their own successors.

Claude now authors more than 80% of the code merged into its codebase, Anthropic said, and has helped engineers increase code output roughly eightfold since 2024.

"Before Claude Code launched in research preview in February 2025, this number was in the low single digits," Anthropic wrote, adding that the shift also shows up in output per engineer. "Lines of code merged per engineer per day stayed constant through Anthropic's first four years (2021-2024), then began to climb upward in 2025 when Claude began to run code rather than just suggesting it for an engineer to copy and paste."

Anthropic said the future could unfold in several ways: AI progress could slow, humans could remain in charge while AI automates much of the work, or AI systems could eventually begin improving their own successors.

"Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor," Anthropic wrote. "This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for."

The company said it's too early to know which outcome is most likely, but argued that AI is already helping build AI, while acknowledging that lines of code are an imperfect measure of productivity.

None of this guarantees recursive self-improvement is on the horizon. It's not yet clear that Claude is capable of research judgment—of choosing the right problems to work on. But if these trends continue, AI systems designing and building their own successors is plausible.

"None of this guarantees recursive self-improvement is on the horizon," Anthropic later wrote on X. "It's not yet clear that Claude is capable of research judgment—of choosing the right problems to work on."

The report comes as AI companies increasingly position their models as research collaborators rather than simple chatbots. Still, Anthropic said the increase in code output reflects a broader acceleration in software development driven by increasingly capable AI agents.

Last month, Anthropic upgraded its flagship Claude model to Opus 4.8, continuing a steady stream of releases aimed at improving coding, reasoning, and autonomous task performance. Rival OpenAI has pursued a similar strategy, launching GPT-5.5 and GPT-Rosalind in April.

Google DeepMind CEO Demis Hassabis believes artificial general intelligence is no longer a distant scientific goal but one that will emerge in the very near future. Speaking at a Stanford Graduate School of Business event last week, the AI pioneer said he expects AGI—the point at which AI is capable of performing a broad range of intellectual tasks at or beyond human levels—to emerge by the end of the decade. "We've been calling AGI this next version of really general artificial intelligence,...

In May, Google announced Gemini Spark, a personal AI agent that doesn't wait to be asked. It manages tasks across apps, flags items that need attention, and finishes jobs in the background.

The report also comes as Anthropic has increasingly emphasized AI systems capable of operating with greater autonomy as it prepares to go public. In recent months, Anthropic has showcased advances in coding, agentic workflows, and long-duration task performance, while touting Claude Mythos.

This looks good. Let me check:

• Title: 10 words, under 12 ✓

• Paragraph count: 14 (matches source) ✓

• All facts preserved ✓

• Light humor added naturally ✓

• No corporate PR language ✓

• No ALL CAPS (except obvious emphasis like AGI, which is already in source) ✓