How AI transforms software development
“You won’t be replaced by AI, but by a colleague who uses it better” — Samuel Durand
A powerful phrase that resonates strongly today — but for how much longer?
Main developer challenges Link to heading
Developers face several major challenges in their work:
- Produce code quickly, maintainable, scalable, quality, secure and reliable
- Identify and fix bugs quickly
- Automate certain repetitive tasks
- Learn continuously and develop skills
What are we talking about? — LLM agents and generative AI Link to heading
An LLM (Large Language Model) agent is an AI based on advanced models like GPT, Gemini or Claude. These agents are capable of:
- Writing, correcting or commenting code
- Answering technical questions
- Accompanying the developer in their work process
Why now? The tools have reached maturity, with direct integration into development environments (IDE, terminal, GitLab…). Solutions like GitHub Copilot or Gemini CLI become daily assistants.
Tools in action Link to heading
🧪 Gemini CLI – Usage example: bug fix Link to heading
Here’s a typical command you can pass to Gemini to fix a referenced bug:
gemini "1. Fix the bug from Jira ticket FBDB-3203
2. Create a branch named after the ticket identifier
3. Add a test that fails because of the bug (use mockModule if needed)
4. Make necessary corrections until the test passes
5. Commit and push the branch
6. Create a merge request to the 'main' branch
7. Add critical feedback in the MR description"
🤖 GitHub Copilot Link to heading
Copilot integrates into the IDE to propose code in real time, complete functions, generate tests or even comment code blocks. A powerful assistant fully integrated into the work tool.
What AI changes… and what it doesn’t change Link to heading
✔️ Benefits:
- Time savings and productivity
- Help with code quality (tests, review, consistency)
- Easier onboarding for new developers
⚠️ Limitations to keep in mind:
- Always plan for human proofreading, which takes time
- Stay attentive to confidentiality and licensing issues
- Don’t become passive to AI suggestions
When to use AI and when to do without Link to heading
Artificial intelligence is particularly useful in the following cases:
- Simple projects or prototypes: to quickly launch an idea, create an MVP or respond to a specific need.
- Exploring new technologies: when you want to test without investing time in complete mastery of a framework or language.
- Understanding assistance: get explanations on existing code, detect bugs, or propose refactorings and improvements.
- Documentation and tests: quickly generate code comments, READMEs, or unit test cases.
It may be preferable to limit or even avoid using AI in the following contexts:
- Complex or critical applications: when you want to finely control architecture, make non-standard design choices, or meet strong performance, security or compliance requirements.
- Deep learning: if the objective is to truly understand how a language, paradigm or system works, AI can become an obstacle by masking fundamental mechanisms.
- Strategic or creative decisions: when value comes from human arbitration (ex: UX choices, business prioritization, technical compromises), AI can only suggest, not decide.
- Highly collaborative environment: in a team, excessive automation can harm code readability or collective consistency if contributions are not sufficiently reviewed.
For now, in the context of my projects, AI gives me the impression of a junior developer: it requires regular proofreading and corrections, but can sometimes produce very relevant results.
The use of AI is not binary. We can consider a progressive and conscious approach, where AI is a support tool — never an autopilot.
Next steps Link to heading
- Better understand and position on the issues of a usage policy related to confidentiality
- Continue trying the tools
- Choose a uniform tool for development
- Progressively integrate interesting use cases into workflows.
An inspiring example: learning to code thanks to AI Link to heading
Marion’s story, who got into development without a technical background, illustrates the power of current tools to develop skills quickly: