Skip to content
Français
Account: not signed in, sync available My learning
Select theme
Français

Secure and govern AI-generated applications

  • 10 min

By updated

Improve this page

This course teaches you to supervise applications built with AI, the ones an employee produces by describing what they want to a coding assistant, without always reading the result. It is aimed at decision-makers (CISO, IT department, management), at the people who build these applications and at the engineers who will have to host them. Module 1 requires no technical prerequisite: it explains what vibe coding covers, why these applications are a problem for the company, how to classify them and who must decide on putting them into service. The following modules need a terminal, Python, pytest and pip, or teach them first.

Who is this course for?

Three audiences meet around the same application, and each sees a different part of it. The decision-maker must know which applications exist, which ones touch sensitive data, and who accepts the risk they carry. The person who builds the application with AI wants to move fast, but without exposing data or leaving a flaw that nobody will see. The platform engineer must be able to host these applications somewhere, with controls that do not depend on everyone's goodwill.

The module "Understand AI-generated applications in business" is shared by the three, and it requires no technical knowledge. It gives everyone the same vocabulary: a governance decision only makes sense if the person who takes it and the person who applies it are talking about the same thing.

Which modules make up the course?

The course has six modules, in this order. The durations are the sum of the lessons announced in the course outline.

  • "Overview": this page, 10 min.
  • "Course": the programme and the progress tracking, 5 min.
  • "Understand AI-generated applications in business": seven lessons, 1 h 55 min.
  • "Technical foundations": eleven lessons, 4 h 30 min.
  • "Vibe coding with method": two lessons, 40 min.
  • "Frame and declare an application": two lessons, 45 min.

The module "Understand AI-generated applications in business" follows the order in which a company discovers the subject. It starts from the definition, goes through what already exists without having been declared, then reaches the decisions to take.

  1. The lesson on vibe coding gives the two meanings of the word, the main families of coding assistants, and the border between assisted development and code accepted without review.
  2. The lesson on shadow AI shows how to discover the uses of AI that nobody has declared, then how to choose between banning, supervising or hosting.
  3. The lesson on risks starts from an incident on a fictitious application, the purchasing portal, and traces it back to its five causes, from the dependency invented by the model to the departure of the author.
  4. The lesson on the hosting architecture describes the zones and trust boundaries of a platform that hosts these applications without exposing the rest of the information system.
  5. The lesson on classification applies the C1 to C4 grid: it starts from the data, flows, users and criticality, and derives the controls required.
  6. The lesson on the legal framework says what the GDPR, the AI Act, NIS2, DORA and the CRA require, and from whom.
  7. The lesson on governance splits the decisions, sets the evidence to require, and explains how to prevent a rule that is too heavy from pushing usage out of sight.

Each lesson of this module ends with a quiz of eight questions. The same case, the purchasing portal built by a buyer to read the orders of an ERP, serves as the running example from one lesson to the next.

The module "Technical foundations" gathers the terminal, Git and Python that the rest of the course needs: navigate a folder, save and reread a change with Git, write a small program, isolate its dependencies and run tests. It is optional for those who already know how to do all this, and for a reader who does not code. Its first lesson, which installs WSL2 to get a local Linux, concerns only people who work on Windows.

The module "Vibe coding with method" teaches how to produce code with AI while staying in control: a verifiable specification before generating, tests written first, a reviewed diff that refuses to let a test be weakened, and dependencies checked before any installation. It requires a terminal, Python, pytest and pip, which the module "Technical foundations" teaches. Its labs are recommended: they are the proof of the gesture.

The module "Frame and declare an application" covers what happens once the application is written. You declare the application in an application sheet validated by a schema, then see why the decision to put it into service is signed outside the repository: the application cannot authorise itself. The second lesson, shared with the Claude Code course, shows what an organisation imposes on developers' workstations and on the repositories where a code agent works.

How do you follow the course?

The course outline, in the left-hand column, gives the order of the modules and lessons, and the page "My course" keeps your progress in this browser, with no account to create. The lessons are read in order: each one builds on the vocabulary of the previous one, and the classification grid is then used by the lesson on governance.

If you are new to the subject, start with the lesson on vibe coding, even if the word seems familiar: the course distinguishes two meanings of the word, which articles often mix up. If you are an engineer and already know the risks of generated code, the lesson on the hosting architecture and the one on classification are the ones that concern you most directly.

What does this course not cover?

The course covers applications built with the help of AI, whether or not they embed a model. It does not deal with the security of a language model integrated into an application (prompt injection, leaks through the model): that subject has its own guide, securing an application that uses an LLM (in French). Nor does it replace legal advice: the lesson on the legal framework says which texts may apply and which questions to ask, but the answer for your case belongs to your legal department.

Key takeaways

  • An application built with AI becomes a company matter as soon as it reads real data or other people use it.
  • Module 1 requires no technical prerequisite and serves as a common base for decision-makers, application builders and engineers. The following modules need a terminal, Python, pytest and pip.
  • The module "Technical foundations" is optional for those who already know how, and its first lesson (WSL2) concerns only Windows.
  • The C1 to C4 grid links the nature of an application to the controls that are required of it.
  • The decision to put into service belongs to an authorised human, never to the requester themselves.