Home >AI >Tools >An overview of AI assistants for programmers

An overview of AI assistants for programmers

The types of AI tools a programmer can use and an overview of the best-known assistants. The offer changes quickly – the overview of tools is as of September 2026, but the division by way of working holds for the long term.

Types of tools

Chat – a web, desktop or mobile app in which you write questions and paste code or files.
Assistant in the editor – an extension of the development environment (VS Code, JetBrains and others) that completes code, edits selected parts and answers questions about the project.
Agent – works in the terminal or in the editor, reads and edits files itself, runs commands and tests.
API – a programming interface through which you build a model into your own app or website. You pay by the tokens used.
Local models – open models running directly on your computer (for example through the Ollama or LM Studio tools). The data never leaves the computer, but the models tend to be weaker than the best online ones and need powerful hardware.

Well-known tools (as of September 2026)

Claude (Anthropic) – a chat app, the Claude Code coding agent (terminal, editors, desktop app) and the Claude API.
ChatGPT (OpenAI) – a chat app, the Codex coding agent and the OpenAI API.
Gemini (Google) – a chat app connected to Google services, tools for programmers (Gemini Code Assist, Gemini CLI) and the Gemini API.
GitHub Copilot (GitHub, Microsoft) – an assistant in editors with code completion, chat and an agent mode. It offers models from several providers.
Cursor – a standalone editor built on VS Code with built-in AI and a choice of models from several providers.

Most services offer a free version with limitations and paid plans with higher limits (as of September 2026). Prices, limits and available models change often – you will always find the current terms directly at the provider.

What changes the fastest

the names and versions of models (a newer one appears every few months),
prices, plans and limits on the number of messages,
which models a particular tool offers,
the terms for keeping data and using it for training.

The principles of working – how to write prompts, how to review code and what not to send – hardly differ between tools. What you learn in one, you will use in the others too.