Claude wrote over 90% of the contributions to Anthropic’s codebase in September 2026, Anthropic says in the Form S-1 it filed with the SEC on 26 October 2026. “Over 90% of the contributions to our codebase in September 2026 were written by Claude”, the filing says.
The same section says the average daily volume of code contributed per engineer in the three months to 30 September 2026 was over 45 times the 2024 average, while headcount grew over five times from 31 December 2024 to 30 September 2026.
Business Pill · DATA FLYWHEEL
A short Business Engineer explainer on the data flywheel. It teaches the general idea only and says nothing about Anthropic or its filing.
The key insight: As we read it, the new element is that the S-1 describes Claude as the main author of Anthropic’s own code: over 90% of codebase contributions in September 2026, with engineers directing and reviewing the work.
What the Filing Says Claude Does Inside Anthropic
The passage sits in the S-1’s Business section, under a heading about Claude accelerating Anthropic’s ability to close the loop. It says: “Many aspects of the research, training, and testing of subsequent model generations now involve Claude, including running experiments, generating and processing training data, automating evaluations and safety audits, and writing and reviewing code.”
Anthropic calls this “closing the loop”. It names the deployment of Claude Code across its engineering organisation as a clear example, and says: “This has enabled our engineers to increasingly direct and review its work rather than write code line by line.”
| Measure | What the S-1 reports | Period |
|---|---|---|
| Codebase contributions written by Claude | Over 90% | September 2026 |
| Claude’s success rate on open-ended engineering problems | Approximately 90%, up approximately 50 percentage points in nine months | September 2026, internal benchmarks |
| Average daily code contributed per engineer | Over 45 times | Three months to 30 Sep 2026 vs year to 31 Dec 2024 |
| Headcount | Grew over five times | 31 Dec 2024 to 30 Sep 2026 |
| Quarterly revenue | 86 times | Three months to 30 Sep 2026 vs three months to 31 Dec 2024 |
| Value of internal Claude usage at list prices | Grown over 250 times | January 2026 to September 2026 |
Source: Anthropic, PBC Form S-1, filed with the SEC on 26 October 2026, Business section. The success rate is based on Anthropic’s internal benchmarks. The multiples cover different periods.

When Internal Use Passed the Payroll
The S-1 puts a price on its own use of Claude. It says the value of that usage at list prices grew over 250 times from January to September 2026, and that “in April 2026, this monthly value first exceeded our monthly personnel expense, excluding stock-based compensation”.
Anthropic links its headcount and revenue figures to the same tool: “We believe this operating leverage has been driven by the use of Claude throughout our organization.”
| Period | Revenue ($M) | Personnel expense ($M) |
|---|---|---|
| Year to 31 December 2024 | 386 | 454 |
| Year to 31 December 2025 | 4,585 | 1,839 |
| Six months to 30 June 2025 | 1,163 | 538 |
| Six months to 30 June 2026 | 16,286 | 3,152 |
Source: Anthropic, PBC Form S-1, filed with the SEC on 26 October 2026, Summary Consolidated Financial Data (US$ millions). Personnel expense includes stock-based compensation; full years audited, six-month periods unaudited.
Connecting the Coverage
Our revenue-per-employee piece reported the S-1’s figure of over $6.6 million of trailing twelve-month revenue per employee, with approximately 5,300 staff. Our crossover piece showed revenue passing compute and infrastructure expense in the first quarter of 2026. Our S-1 piece covered the headline numbers. What is new here is how the filing says the work itself is done: mostly by Claude, directed and reviewed by engineers.
The Structural Read
As we read it, the filing pairs two growth rates: code contributed per engineer up over 45 times against 2024, and headcount up over five times since the end of 2024.
The S-1 also prices the tool it sells against its own payroll: in April 2026, the monthly list-price value of internal Claude use first exceeded monthly personnel expense, excluding stock-based compensation. As we read it, Anthropic is presenting its own staff as users of the product it sells.
As we read it, the same filing frames the loop as a risk too: the Risk Factors say Anthropic has observed its models contributing to the pace of its own research.
Anthropic, PBC Form S-1, 26 October 2026
“This has enabled our engineers to increasingly direct and review its work rather than write code line by line.”
Three Implications
FOR ENGINEERING LEADERS The S-1 reports over 90% of Anthropic’s September 2026 codebase contributions as written by Claude, with engineers directing and reviewing the work.
FOR INVESTORS The filing attributes its operating leverage to Claude, stated as Anthropic’s belief, and reports quarterly revenue up 86 times against headcount up over five times.
FOR SAFETY RESEARCHERS The S-1 says using AI to help build more capable AI could pose a considerable safety risk, and that this falls under its Responsible Scaling Policy.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Four Intelligence Moats.
The framework names one moat for each data pipeline, among them “The Harness Moat — built by agentic-loop infrastructure, active and where most current moat construction is happening.”
As we read it, the S-1’s description of Claude Code working across Anthropic’s engineering organisation is a company-level case of that agentic loop. The framework was published in July 2026 and does not discuss this filing.
What Is Not Established
The success rate comes from Anthropic’s internal benchmarks, which the passage does not publish. The passage gives the contribution share for one month, September 2026, and does not say in that passage how a contribution is counted. The value of internal usage is measured at list prices, not as cash spent. The link between Claude and operating leverage is stated as Anthropic’s belief. The Risk Factors add: “We have observed that our models are contributing to the pace of our own research”, and the Business section says “Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress.”
The Bottom Line
Anthropic’s S-1 says Claude wrote over 90% of the contributions to its codebase in September 2026, that code contributed per engineer rose over 45 times against 2024 while headcount grew over five times, and that the monthly value of its own Claude use first exceeded its monthly personnel expense, excluding stock-based compensation, in April 2026.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. All figures are from Anthropic, PBC’s Form S-1 filed with the SEC on 26 October 2026, mainly its Business section; the success-rate figure is based on Anthropic’s internal benchmarks. Earlier reports are as covered in our pieces linked above. This is not investment advice. We did not contact Anthropic.
Sources: Anthropic, PBC Form S-1, filed 26 Oct 2026 (SEC EDGAR)









