Skip to content
Agents tracked: 502 Releases (7d): 613 Agent status: 4 with issuesSince last limit reset: Claude 7d · Codex 5d For AI agents: Markdown Updated Nov 18, 2026
Case study DoctolibResults as reported by Anthropic

Doctolib got 47% more pull requests per engineer with Claude Code

Doctolib, the European healthcare booking and practice-software platform, rolled Claude Code out to its whole development team and then built its own remote coding agents on Claude Code in headless mode through GitHub Actions. Between January and September 2026 the average engineer handled 47% more pull requests and change lead time fell by 25%, as reported by Anthropic; the agents now open 10% of all pull requests.

The problem

Documentation, tests, pull request reviews and technical debt kept pulling Doctolib's developers away from harder healthcare problems. Once Claude Code made writing code faster, the bottleneck moved to everything around it: reviewing each change with the care healthcare software needs, checking it for privacy and security risk, and confirming it behaves once it ships. Running every check on every change by hand was not realistic.

How they did it

  1. Start with a pilot group of engineers, then roll Claude Code out to the whole development team; engineers set it up themselves from VS Code, JetBrains or the command line.
  2. Keep a shared repository of prompts, custom commands and subagents that every engineer pulls at setup, and revisit the default model and settings as new models arrive (Doctolib uses Opus plan mode for interactive work).
  3. Run remote agents on Claude Code in headless mode through GitHub Actions; Anthropic's docs set this up with the claude-code-action workflow and an ANTHROPIC_API_KEY repository secret, and the /install-github-app command does it from inside Claude Code.
  4. Chain the agents around a ticket - generate the change with context, update the pull request, work through review comments, then run a check once the code is in production - instead of running single-purpose bots.
  5. Pick the model behind the agents with A/B tests on a set of your own real tasks, comparing result quality and cost per pull request.
  6. Add scheduled jobs for the boring parts, such as updating technical documentation with every code change and running automated Claude reviews on the main infrastructure repository.
  7. Keep humans on the merge decision; at Doctolib the team that owns the affected product accepts or rejects each agent-proposed fix.

From the official docs

claude   # then run /install-github-app inside the repository

Results as reported by Anthropic

  • The average engineer handled 47% more pull requests between January and September 2026, as reported by Anthropic.
  • Change lead time fell by 25% over the same period, as reported by Anthropic.
  • Doctolib's remote coding agents now open 10% of all pull requests, and half of the fixes they propose for identified bugs are accepted and merged, as reported by Anthropic.
  • In Doctolib's A/B tests, Opus 5.5 produced better results than Sonnet 5.0 at 2.5 times lower cost per pull request, as reported by Anthropic.
  • By Doctolib's own estimates, writing code went from 40 to 45% of an engineer's time to roughly 20% in about a year, with more time on review and specifications, as reported by Anthropic.
  • About 90% of Doctolib's roughly 900 tech and product staff use Claude every day, as reported by Anthropic.

These figures are as reported by Anthropic in Anthropic customer story ↗. Anthropic sells the product used here. AgentGid did not measure or independently verify them.

Takeaway. When coding agents make writing code cheap, the next gains come from automating review, risk checks and post-merge verification around each change, not from writing more code.
AgentGid's take

This fits a large engineering organisation with mature CI, code owners and a platform team that can maintain shared prompts, subagents and agent workflows; the headline numbers come from that system, not from installing Claude Code alone. A small team gets the interactive part cheaply but may not need the remote-agent layer. Per AgentGid's Claude Code page (pricing checked Nov 4, 2026), Claude Code needs a paid plan: Pro from $17 a month on annual billing ($20 monthly), Max $100, Ultra $200, Team premium seats $100 per seat a month annual ($125 monthly), and Enterprise seats $20 per seat a month annual with a 20-seat minimum plus usage at API rates. GitHub Actions runs bill API tokens (or draw on a subscription token) on top of Actions minutes. All figures come from Anthropic's own customer story, with no publication date and no independent measurement; the time-split numbers are internal estimates. For an open-source terminal agent you can wire into CI yourself, OpenCode has a free tier, with paid use from $10 a month, per our catalog.

The agent used here

Similar use cases

Guides