AymericDev← Back to homeFrançais ↗

From understanding the need to production

How I work

I start with your needs, experiment and build in small steps. Across AI and web development, I aim to make decisions clear, progress visible and results verifiable.

01 / Working together

Making progress, one step at a time.

  1. Understand the need

    We clarify goals, use cases, constraints and success criteria. I break down the work into concrete priorities to define a useful first step.

  2. Design and experiment

    I choose an approach that fits the context and test assumptions with a prototype when needed. For AI projects, this also means examining the data and evaluation criteria.

  3. Build in small steps

    I build in small increments, with tests suited to the expected behaviour and risks. Regular check-ins give us opportunities to review progress and adjust what comes next.

  4. Deliver and support

    I prepare deployment, useful documentation and handover. Feedback from real use then informs fixes and further improvements, within the support we agree on.

02 / The development approach

Informed by Extreme Programming.

My approach draws on Extreme Programming (XP): short iterations, frequent conversations and continuous attention to software quality. Each step should produce something we can review and improve.

  • Concrete needs and small releases to get useful feedback quickly.
  • Tests to clarify expected behaviour and make changes with confidence.
  • Simple code, improved gradually through refactoring and integrated frequently.
Extreme Programming diagram in French: needs, small iteration, tests first, simple code, continuous integration, delivery and feedback, then back to the needs.Open the full-size diagram
Extreme Programming — a cycle of learning and delivery.Diagram in French; English summary below.

Reading the diagram

Six steps connect the initial need to user feedback. Within each iteration, the test → code → refactoring cycle helps verify and improve the solution. Communication, simplicity, feedback, courage and respect guide this work.

I adapt these practices to the project and team. The diagram also introduces XP’s values and team practices, such as pair programming; their use depends on how we work together.

Extreme Programming practices

03 / Version control

Git, with short-lived branches.

To track code changes, I favour a Trunk-Based Development workflow with Git: one reference branch, main, and short-lived working branches that are integrated frequently.

  • A small change branched from main, with focused commits that each describe a specific evolution.
  • A pull request to explain the change, run automated checks (CI) and support code review.
  • A prompt merge after validation, with the aim of keeping main stable and deployable.
Git Trunk-Based Development diagram in French: a feature/login branch starts from main, receives a few commits and rejoins main after a pull request, checks and review.Open the full-size diagram
Git — a change’s journey back to the main branch.Diagram in French; English summary below.

Reading the diagram

Branch from main → make small commits → push → open a pull request → run CI and review → merge promptly. Branches stay short-lived, commits focused and main deployable.

This rhythm keeps changes easy to review and limits divergence between branches. Naming conventions and validation rules are agreed to suit the project.

How short-lived branches work

04 / Working with Codex

Developing with Codex, staying in control.

I delegate a substantial share of code production to Codex while leading the goals, architecture and key technical decisions. I define the scope, examine the results and remain responsible for the quality of the software I deliver.

  • The repository holds the project’s memory: code, README, architecture documentation, AGENTS.md instructions, decisions and current progress remain available between sessions.
  • I break the work into small tasks with observable outcomes and organise sessions around a specific feature.
  • Before a structural change, Codex proposes a plan for my approval. It can then carry out the task autonomously within the agreed scope.
  • I ask for evidence suited to the change: tests actually run, CI results or a demonstration, together with an explanation of what changed.
French diagram of a nine-step Codex workflow, from product vision to merge, with human approval of the plan, verification and review before integration. The loop returns to the next specific task.Open the full-size diagram
Using Codex to develop a project from start to finish.Diagram in French; English summary below.

Reading the diagram

Product vision → repository and documentation → specific task → Codex plan → human approval → implementation → verification → review and understanding → merge and iterate. The loop returns to defining the next task. The repository preserves the project’s memory; Codex executes within an agreed scope while I retain control of decisions and integration.

Significant changes go through a review session separate from implementation. Before integration, I read the code, verify the results and make sure I understand the decisions and their consequences. Codex proposes, codes, tests and documents; I decide and validate.

Codex in the OpenAI documentation

Your project

Let’s start with what you need.

A goal, an idea to test or a system to improve: let’s discuss the first step.

Let’s discuss your project