Anthropic Filed to Go Public, Says AI Is Starting to Build Itself

In Brief

Anthropic took its first formal step toward an IPO and, on the same week, published a paper arguing that AI is now doing more and more of the work of building AI. 

The company says more than 80% of the code it uses is now written by its own model, Claude, and that the future it considers most likely is one where a 100-person company can do the work of 10,000.

What Happened

Anthropic, the AI lab behind the Claude models, confidentially filed documents with the SEC as a first formal step toward going public. Because the filing is confidential, the company has not disclosed share counts or pricing yet. But it comes right after a $65 billion funding round that valued Anthropic at roughly $965 billion, with run-rate revenue (annualized sales based on its current pace) reported to have crossed $47 billion.

The more striking news came alongside it. Anthropic's in-house research arm published a piece called "When AI builds itself," co-authored by company co-founder and head of policy, Jack Clark. The argument: AI is increasingly building the next version of AI, a trend called recursive self-improvement. This is a system that can eventually design and develop its own successor with humans largely out of the loop.

To make the case, Anthropic says more than 80% of the code it now merges into its own software is written by Claude, and that its engineers built roughly eight times more code per day in the second quarter of 2026 than they did in 2024.

On Episode 218 of The Artificial Intelligence Show, SmarterX founder and CEO Paul Roetzer broke down what this advancement means for businesses. 

The Key Numbers

$965 billion - Anthropic's valuation after its May raise

$47 billion - Reported run-rate revenue behind the IPO filing

80%+ - Share of Anthropic's own code now written by Claude

4 months - How long it now takes for AI to double the length of tasks it finishes on its own

76% - Claude's success rate on the most open-ended tasks in May 2026

Why Business Leaders Should Pay Attention to AI Building Itself 

The pace of improvement is accelerating. Anthropic reports that the length of tasks its models can reliably finish on their own is now doubling roughly every four months, up from an earlier pace of every seven months. The progression is steep: in March 2024, Claude could handle software tasks that take a human about four minutes. A year later, about an hour and a half. A year after that, 12-hour tasks. Roetzer's point is that this is not just a coding story. "They're focused on AI research because that's what all the labs are trying to build first," says Roetzer. "But whatever happens there, you as a listener, you need to extract that into your own industry. Something that takes you weeks, you're going to start to be able to do in minutes through simple prompts."

The paper lays out three possible futures. The first is that progress stalls. Another is full recursive self-improvement, where AI designs its successors and humans move to simply overseeing AI's work. But the other scenario, the one Anthropic calls most likely, is where AI development gets heavily automated while humans still set direction. With this scenario, "100-person companies could do the work of 10,000 or 100,000 person organizations." Roetzer agrees, and applies it directly to his own company. "When I think about SmarterX and what we're building, I think we can be a $100 million a year company with 100 or fewer people," he says. That works out to $1 million in revenue per employee, and he thinks two to three times that is realistic for a company built around AI from the ground up.

Don't be fooled by one line. Much of the press coverage focused on a line near the end of the paper where Anthropic said it would "likely be a good thing" to have the option to slow frontier AI development. Roetzer is blunt that this will not happen. "That is not going to happen because China won't agree to it," he says adding, "The US government is not going to slow down no matter what they say publicly they're doing." The takeaway he wants leaders to focus on is the competitive consequence.

"Someone is building the 100-person version of your 1,000-person company, and how are you going to compete with them?

"There's not very many obstacles to someone just building a smarter, more efficient, more profitable version of your company. I don't even see that as debatable. It's inevitable across every industry."

— Paul Roetzer, founder and CEO of SmarterX, Episode 218 of The Artificial Intelligence Show

SmarterX Take

The instinct for most leaders reading about recursive self-improvement is to file it under science fiction. But what if you consider it's 5% true. You do not need to believe AI will build its own successor next year to take the competitive threat seriously. You only need to believe the scenario Anthropic calls most likely: AI handles most of the work, humans set the direction. Even then, you have a lot of work to do to prepare.

That scenario is already showing up inside the companies closest to the technology. Anthropic is using itself as an example, publishing its own internal data instead of pointing at outside studies. The honest move for any leader is to ask the same inward-looking question: If a competitor built an AI-native version of your business with a fraction of the headcount, what would actually stop them? For some companies the answer is product, distribution, or brand. For many, the answer is much less. 

What to Watch

The race to the public markets is on. SpaceX, which has absorbed Elon Musk's xAI lab, is expected to go public first, making it effectively the first AI lab to list. After that, it becomes a contest between Anthropic and OpenAI for who reaches the market next and who carries the higher valuation. Expect far more financial detail on these labs to surface as the filings move forward.

AI-native challengers taking on established companies and brands. The clearest near-term proof of the most likely impact of AI building itself is AI-first companies entering established industries and operating with a fraction of the employees. When that pattern shows up in your sector, time will have run out to rethink your team structure and cost base.

Where Is Your Organization on the Path to Scaling AI?

Only 25% of organizations have reached the Scaling phase of AI adoption, according to the 2026 State of AI for Business Report. The largest group, 47%, is still Piloting, and another 28% is still just Understanding AI. That gap is the whole point of Anthropic's warning: the companies most exposed to AI-native competitors are the ones still stuck in the early phases.

The report is built on more than 2,100 responses from professionals across roles, functions, and industries, and it maps exactly where organizations stand on adoption, governance, training, and tooling. If you want an honest read on whether your company is positioned for the future Anthropic describes, check out our data to benchmark your business. Read the full report →