Doctolib has radically changed its software development methods. Since early 2026, no fewer than 600 engineers have been using artificial intelligence agents continuously. This revolution impacts not only code speed but also collaboration and project management.
The shift to an « agent-first » model represents a true technological and organizational leap. AI is no longer simply a support tool; it becomes a genuine partner in the creation of functionalities. This reveals a new role for developers, who become the conductors of artificial intelligence.
Such a transformation prompts a rethinking of the developer’s role and also raises the question of real productivity gains. But one thing is certain: Doctolib is betting heavily on AI to consolidate its business model after 12 years of growth.
Doctolib: a 100% agentic development model
Since January, all Doctolib teams (600 engineers) have been working 100% with AI agents, primarily using the Claude Code tool. This widespread adoption follows a pilot phase launched in 2025 where each team tested different solutions.
“AI is becoming a contributor, not an assistant,” explains Julien Tanay, SRE staff at Doctolib. This means that workflows need to be reviewed and a major shift in the developer’s role accepted. Some engineers are now managing up to eight agents simultaneously to produce and validate code.
Technical specifications are central to the process, following a « spec-driven » approach. Markdown files allow for a clear structuring of expectations, and Doctolib relies on OpenSpec to manage this documentation. This responsible method avoids being dependent on proprietary tools.
Specialized AI agents to automate all repetitive tasks
Three families of agents work in parallel: « AI first » agents for development, « remote » agents that manage backlogs and production errors, and automations dedicated to code review. These agents handle very specific tasks such as technical migrations or ticket triage.
These intelligent robots run 100% automated pipelines, drastically reducing human intervention. This is revolutionizing the way we approach software maintenance and evolution, with a volume of actions impossible to manage otherwise.
An engineer describes this change: « Before, two new arrivals wanted to integrate each day; now 600 employees do the same. What was once a nuisance becomes a major obstacle without the right tools. »
An internal platform to democratize the use of AI
To support this reorganization, Doctolib has designed a centralized platform. It offers an internal marketplace where extensions and tools for Claude Code can be found, thanks to dedicated « skills. » The idea is also to open these services to non-developers, particularly product managers and designers.
“Centralization addresses the need to capitalize on knowledge, which is essential for the company,” explains Yankı Sesyılmaz, one of the platform’s developers. The goal is to encourage adoption and facilitate skills development through a flexible and scalable system.
Not forgetting security: before integrating a tool, Doctolib conducts thorough tests with internal experts, thus offering a balance between innovation and risk management.
Training at the heart of the technological transition
The shift to agent-first also requires a significant training effort. Doctolib has implemented a three-level training program, the most advanced of which focuses on « spec-driven » programming. This allows engineers to learn how to manage agents while maintaining a critical perspective.
Recruitment takes into account this appetite for AI and the ability to work collaboratively with it. This change illustrates the dual nature of the modern profession: mastering the technology and orchestrating artificial intelligence.
So far, traditional quality and delivery time indicators are being tracked, but it is too early to confirm a net increase in productivity. Time spent on code reviews has increased, offsetting the increased coding speed.
Doctolib is preparing the future of medical consultations with AI
While the agent-first approach is revolutionizing the behind-the-scenes development process, Doctolib is also making strides in user services. A major innovation: the AI Medical Assistant, which promises to lighten the workload of healthcare professionals and improve the patient experience.
This project, the impact of which can be seen in this feedbackThis involves, in particular, detecting the urgency of a situation through a telephone conversation with the patient. This represents a huge gain for appointment management and prioritization of care.
This integration of AI into medical practice is part of a broader growth and innovation strategy. It’s worth noting that Doctolib became profitable after more than a decade of sustained investment, a turning point for the French unicorn.
The expected benefits of AI for Doctolib and its users
- Optimization of internal processes thanks to the automation of repetitive tasks
- Quality improvement through more rigorous code reviews conducted in part by AI agents
- Open to non-developers with tools accessible to the entire product team
- Enriched patient experience through intelligent services such as the consultation assistant
- Support for profitable growth driven by a continuous optimization of resources
A look at the future of agile methods aided by artificial intelligence
Doctolib anticipates that these practices will transform the very structure of its agile teams. Developers will no longer be solely coders but coordinators, managing multiple agents simultaneously. This model could also pave the way for greater inclusion of other roles in the product lifecycle.
A promise full of challenges to preserve the human spirit behind the machine, in a context where transparency and digital sovereignty remain major issues.
What is the « agentic-first » model adopted by Doctolib?
This is a method where artificial intelligence is no longer a simple support tool, but a full-fledged contributor to software development. Developers coordinate several AI agents to produce, review, and automate code.
How does Doctolib use AI agents to manage repetitive tasks?
Doctolib deploys remote agents to process tickets, manage production errors, and automate technical changes such as migrations. These agents run fully automated pipelines, significantly reducing human intervention.
What is the role of specifications in the new development method?
Specifications are at the heart of the process, formalized in Markdown files. They guide programming according to a spec-driven approach, reinforcing rigor and clarity even before coding.
Has Doctolib already measured the productivity gains?
Not definitively yet. Teams are coding faster, but spending more time on code review. More time will be needed to confirm a clear impact on production deployment and quality.
How does Doctolib involve other professions in this transformation?
Thanks to a centralized platform, product managers and designers can access AI tools via an internal marketplace. This makes agentic practices accessible beyond just developers.
Source: www.lemondeinformatique.fr