The past week delivered a stark reminder of how fast the AI industry is moving — with the centerpiece being a rare, data-rich paper from the Anthropic Institute that pulls back the curtain on how AI is accelerating its own development.
The concept at the heart of the paper: Recursive Self-Improvement (RSI) — a scenario where an AI system can autonomously design and develop its successor. "We are not there yet," Anthropic writes, "and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for."
What's the big deal?
The paper reveals a staggering metric: Anthropic engineers now ship 8x more code per quarter on average than they did from 2021–2025. The length of tasks AI models can reliably complete on their own has been doubling every four months — from tasks that take humans about four minutes in March 2024 to tasks lasting twelve hours by March 2026.
If the trend holds, days-long tasks could come into range this year. By 2027, AI systems could be capable of completing tasks that take a skilled person weeks.
The wider news cycle
Separately, Sam Altman reportedly discussed with the Trump administration the possibility of the US government taking an ownership stake in OpenAI — pitching the idea as a way to channel AI's economic benefits to the public. NOTUS reported that Altman first floated the idea to President Trump early last year.
Google signed a compute agreement with SpaceX to meet what it called "surging customer demand" for Gemini Enterprise. The deal mirrors a similar arrangement Anthropic signed with SpaceX in May. Google described it as a short-term measure to keep up with demand that has been "even higher than we expected."
On the regulatory front: New York state passed a bill barring AI companies from letting chatbots suggest they're human when interacting with minors. The bill follows lawsuits against OpenAI and Character.AI over claims that their chatbots harmed teen users. At the federal level, bipartisan lawmakers released a 269-page draft bill as a starting point for national AI regulation.
Reid Hoffman, LinkedIn's co-founder, announced he is leaving Microsoft's board to focus on Manas, the AI drug discovery startup he co-founded last year.
Google also shut down Pixel Studio, the AI image-generation app that launched alongside the Pixel 9 in 2024, directing users to Gemini instead.
The bottom line
This week marks a rare moment when a leading AI company opens a window onto what's happening behind closed doors. Anthropic's data points to a steady, non-linear acceleration — and a future where the line between AI tool and independent AI researcher may blur faster than most expect.
