Anthropic says Claude now writes most merged code internally
Anthropic says Claude wrote over 80% of merged code by May 2026, while warning that code lines are an imperfect measure of productivity.
#
Key takeaways
- Anthropic says Claude authored more than 80 percent of code merged into its codebase by May 2026.
- The company says this rose from a low single-digit share before Claude Code entered research preview.
- Anthropic warns that lines of code are an imperfect productivity measure.
- Human review still remains part of the process.
- The company says automated review could have caught about one-third of bugs behind earlier claude.ai incidents.
Anthropic updates its internal AI engineering analysis
Anthropic updated its analysis of AI-assisted research and engineering on September 18, 2026. The company says Claude authored more than 80 percent of the code merged into Anthropic's codebase as of May 2026. It says that share was in the low single digits before Claude Code entered research preview in February 2025.
The company also says the typical engineer merged eight times as many lines of code per day in the second quarter of 2026 as in 2024. Anthropic explicitly cautions that lines of code are an imperfect productivity measure. It says the metric likely overstates the true gain.
What the company says improved
Anthropic reports stronger results in its internal coding sessions across trivial, routine, substantial, and open-ended work. It says the largest improvement came in open-ended tasks. The company also says Claude's success rate rose across those task types over time.
In a March survey of 130 Anthropic research employees, the median respondent estimated about four times as much output on comparable projects when using an internal model. That figure comes from employee estimates, not an independent benchmark. Anthropic says these are its own internal measurements.
Human review still matters
Anthropic says human review remains part of the process. It also says Claude-written code was considered worse than human-written code in late 2025, but roughly at parity by the time of the update. The company adds that staff views differ.
The company now uses an automated Claude reviewer before changes can merge. Anthropic says a retrospective analysis found that such review could have caught roughly one-third of bugs behind earlier claude.ai incidents. That is a company-reported internal finding, not an external audit.
What this report does and does not show
This source describes Anthropic's own internal measurements. It does not provide independent productivity findings. It also does not establish a general rule for all teams or all coding environments.
The report is useful because it shows how Anthropic measures AI-assisted work inside its own engineering process. It is less useful as a universal benchmark. The company itself warns that one of its main metrics may overstate the gain.
Morocco relevance
The source reports no Morocco-specific fact. For readers, the global lesson is that internal AI productivity claims should be read with care, especially when the metric is lines of code.
Why this matters
The update highlights a shift from experimental use to deeper workflow integration. Claude is not only generating code, but also helping review it before merge. That suggests AI tools can move into more of the software development pipeline when teams build review and oversight into the process.
At the same time, Anthropic's own caution is important. A higher volume of code does not automatically mean better engineering outcomes. The report points to a familiar governance issue: teams need measures that capture quality, not just output.
Follow us on Google
Add Intelligence Artificielle Maroc as a preferred source to see more of our relevant stories in Google Search.
What would you like to build?
We build custom AI platforms, SaaS products, intelligent business applications, and automation systems.
This form is for project inquiries, not general questions about artificial intelligence.
