Understand this story
On November 16, 2026, Addy Osmani wrote an article introducing the Claude Code team's verification methods. According to an article published the same day on Anthropic's Claude.dev developer blog, the Claude Code team says that about 90% of its code is written and reviewed by Claude, so it treats the verification loop as a core engineering practice. The article describes three types of testing—end-to-end testing, property testing, and differential testing—as well as practices such as the verify skill, a 39-item reliability checklist, and commit gates. The team also mentioned that in internal tests of structured file generation, adding an independent grader to check Claude's output against grading criteria raised the task success rate by up to 10 percentage points compared with a standard prompt loop. The next day, Claude Devs said that when Claude can verify its own work, the code it writes is of higher quality, and that checks such as tests, benchmarks, and end-to-end runs, which were costly in the past, have now become cheap, and included a link to the Claude Code team's explanation of its verification approach.
AI-generated from public reports
Latest development · Report
Claude Code team on why verification makes code betterDevelopments
1 confirmed developments
Addy Osmani writes about the Claude Code team's verification methods
Source:English text is being prepared.
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 17
Claude DevsFirst-party
Claude Code team on why verification makes code better
ClaudeDevs says that when Claude can verify its own work, the code it writes is of higher quality; checks that used to be costly, such as tests, benchmarks, and end-to-end runs, have now become cheap. The post also includes a link to the Claude Code team's explanation of their verification approach.
Nov 16
Anthropic: Claude.dev developer blogFirst-partySelected
How the Claude Code team uses verification loops to make agents self-check their code
The Claude Code team says that about 90% of its code is written and reviewed by Claude, so it treats verification loops as a core engineering practice. The article introduces three types of tests—end-to-end tests, property tests, and differential tests—as well as practices such as the verify skill, a 39-item reliability checklist, and commit gates. The team also mentioned that in internal tests of structured file generation, adding an independent grader to check Claude's output against grading criteria improved task success rates by up to 10 percentage points compared with a standard prompt loop.