The reference guide to building your own legal tools without writing a line of code
ParPierre Colliot•Expert Legal Tech & Vibe Coding 9 complete parts Ready-to-use prompts Secure configurationsThis document is intended for lawyers and legal professionals who want to take back control of their working tools. For two decades, legaltech operated on a one-way model: vendors designed software, firms bought it. Lawyers were stuck with poorly adapted interfaces, features designed for other professions, prices imposed without any room for negotiation.
Vibe coding reverses this dynamic. Have an idea for a tool to automate your case tracking? You can build it yourself. Want a compensation calculator tailored to your practice? A few hours are enough. The lawyer is no longer the passive consumer of technological solutions: they can become its architect.
This guide brings together the minimal technical knowledge, the proven methods and the essential precautions to get there. It promises no miracles. It offers a practicable path toward technological autonomy, with a clear-eyed view of the current limits of these tools.
On 6 February 2025, Andrej Karpathy posted a message on the social network X that would transform how we think about software development. The figure is no minor one: a co-founder of OpenAI and former director of artificial intelligence at Tesla, he ranks among the most respected engineers of his generation. His observation fits in a few sentences:
"There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. I'm building a project or webapp, but it's not really coding — I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works."
Karpathy describes a new practice: building applications by describing what you want in natural language, without examining or modifying the code generated by artificial intelligence. The developer focuses on the result, not on the technical means. They "forget that the code even exists".
The term "vibe coding" spread with unusual speed. In March 2025, the Merriam-Webster dictionary added it to its list of emerging terms. In November, Collins named it word of the year 2025. By the end of 2025, Linus Torvalds himself (the creator of Linux, a tutelary figure of the free-software world) had used this approach to develop a component of his AudioNoise project.
What distinguishes vibe coding from classic development assistance is the deliberate absence of human intervention on the code. When a developer uses GitHub Copilot, they write code and the tool completes their lines. When they ask ChatGPT about a function, they are trying to understand existing code. Vibe coding works differently: you start from an idea expressed in plain English, the AI generates the full code, you test the application, you describe the adjustments you want, the AI fixes them. At no point do you open the source code to modify it by hand.
The adoption of AI tools by professional developers has exploded. According to recent studies, 84% of developers now use AI tools that generate 41% of all code produced. Microsoft and Google have confirmed that around a quarter of the code in their products is now generated by artificial intelligence.
Reported productivity gains vary considerably from one study to another. Developers generally estimate they save between 10 and 30% of their time. Some repetitive tasks (generating tests, documentation, scaffolding) see reductions of 30 to 60%. But these optimistic figures need to be qualified.
The legal sector is following this trend with a time lag. The adoption of AI by law firms rose from 19% in 2023 to 79% in 2024, a fourfold increase in a single year. The market for legal software incorporating AI, valued at 1.2 billion dollars in 2024, is expected to reach 12 billion by 2033. McKinsey estimates that 22% of a lawyer's tasks can be automated today, and that 44% can be automated from a technical standpoint.
Vibe coding is not a universal solution. Its critics point to real risks, documented by concrete incidents.
This vulnerability (CVE-2025-48757) illustrates a structural problem. AI models generate code that meets the immediate request, but they cannot grasp the overall security context, the threats specific to your application, or your regulatory obligations. They produce what you ask for, not what you need.
Research has identified more than 2,000 vulnerabilities across roughly 5,600 applications built through vibe coding, with 400 exposed secrets (API keys, passwords) and 175 cases of publicly accessible personal data, including medical records and bank details.
For a lawyer, this distinction takes on a particular hue. Professional secrecy, ethical responsibility and potential disciplinary sanctions impose heightened vigilance. A poorly secured case-tracking tool does not just expose data: it puts your professional liability on the line.