Doctolib
Claude Code and remote coding agents for engineering
Doctolib's engineers opened 47% more pull requests after it rolled out Claude Code
Problem
Doctolib's engineers spent time on repeatable work such as documentation, tests, reviews and technical debt instead of harder healthcare problems. By its own estimate, writing code took 40-45% of an engineer's time.
Approach
Doctolib ran a pilot with 30 engineers and then rolled Claude Code out to the whole development team. About 90% of its roughly 900 tech and product staff use it daily. The company also runs its own remote coding agents on Claude Code through GitHub Actions, covering changes from ticket to post-release check, plus review, risk assessment and bug fixes. Claude Opus 5.5 is the default model after A/B tests against Sonnet 5.0.
Results
Takeaway
Doctolib measured throughput and lead time and compared model cost per pull request. The time-share estimate for coding is internal and the company did not describe how it was calculated.
Categories
- Industry
- Healthcare and life sciences
- Business function
- IT and R&D
- Innovation type
- Process innovation
- AI technology
- Generative AIAI agents
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Anthropic's customer story describes Doctolib's rollout of Claude Code and its own remote coding agents. It reports 47% more pull requests per engineer, 25% shorter change lead time, remote agents opening 10% of pull requests, and 2.5x lower cost per pull request with Opus 5.5. The page carries no date; the Internet Archive first saved it on November 7, 2025.
Full details are in the original sources.