Writing a program once took years of learning before any result. Today, AI assistants change the game: they write, explain and fix code. Here is what it is about, without mystery.

What is an “AI that codes”?

These are language models trained on huge amounts of code (and text). From a plain-language request — “write me a function that…” — they suggest code, explain it, spot errors and propose fixes.

They do not “understand” like a human: they predict the most likely continuation, with striking efficiency. But they can also be confidently wrong.

The main players

• Claude (Anthropic) — a general assistant known for reasoning and code; usable in chat or integrated into tools.
• Codex / Copilot (OpenAI / GitHub) — Copilot integrates into the editor and completes as you type; Codex refers to OpenAI’s code models.
• Others exist (Gemini, open-source models…), with different strengths.

These are not the same companies: Claude ≠ Codex. They feel similar in use, but come from competing players.

How they are used

• In the editor: the assistant completes code as you type (smart autocomplete).
• In conversation: you ask a question (“why this error?”, “how do I do X?”) and get code + explanation.
• In agent mode: the AI chains several steps (read files, edit, test) under supervision.

What they do well — and their limits

Strengths:

• Start fast: a first working draft in seconds.
• Learn: they explain a piece of code, line by line.
• Debug: they often spot the error faster than by hand.

Limits:

• Hallucinations: wrong code that looks correct. Test it.
• Security: generated code may contain flaws; vigilance stays human.
• Context: they ignore what you do not tell them (business rules, secrets).

Beginner or pro: to each their use

• Beginner: a patient tutor that explains and gives examples. As long as you seek to understand, not copy-paste blindly.
• Professional: a productivity boost on repetitive tasks, time kept for design and decisions.

The golden rule

AI suggests, the human checks and decides.

It is a copilot, not an autopilot. It speeds you up, but the responsibility — accuracy, security, choices — stays yours. The same logic applies to automating content with AI.