According to various studies, 60 to 75 % of developers are already using AI to code. Whether they are Generative AI (ChatGPT, Claude, Gemini...), IDEs (VS Code, Jetbrains, Cursor...) or specialist extensions (GitHub Copilot, Tabnine, Gemini Code Assist, SonarQube AI...). So how do you choose? For which uses? Should you use AI with your eyes closed and do some vibe coding? Find out how to make the right choices and speed up your workflows!
The software development landscape has undergone an unprecedented transformation since our article " AI: code generators, a revolution for developers "In February 2024, we published an article exploring the advantages and limitations of code AI.
Once confined to specific fields, artificial intelligence has established itself as an essential partner for developers. And in the space of a year, new tools dedicated to code have appeared: Claude Code, Gemini Code Assist, Replit, etc.
According to various recent studies, between 60 and 75 % of code professionals already use AI in their daily workflowrevolutionising working methods that had been established for decades.
This massive adoption can be explained by the growing diversity of solutions available. Generalist generative AIs such as ChatGPT, Gemini or Claude specialised extensions such as GitHub Copilot or Tabnineand integrated development environments such as CursorToday, developers have an arsenal of tools with impressive capabilities.
Intelligent autocompletion, automatic refactoring, test generation, customised commits... the possibilities seem endless. AI has infiltrated every stage of development. Every second commit now includes automated suggestions.
And yet, in the face of such abundance, a crucial question arises: how do you choose the right tool for your specific needs? Because while 'vibe coding' - the practice of letting AI generate the majority of the code - is growing in popularity, it would be unwise to use these technologies blindly.
Each solution has its strengths, weaknesses and preferred use cases, which need to be identified in order to optimise your development workflow.
The main AI code assistants
In the burgeoning ecosystem of AI assistants for development, four major solutions stand out for their adoption and capabilities.
GitHub Copilot: the pioneer integrated into the GitHub ecosystem
Developed jointly by GitHub and OpenAI, GitHub Copilot has established itself as the benchmark for AI coding assistants.
Its native integration into the GitHub ecosystem gives it a considerable advantage for the millions of developers using this platform. Trained on the immense database of public code available on GitHub, Copilot offers remarkably relevant contextual suggestions.
The tool stands out for its ability to adapt to the developer's coding style. It analyses the immediate context, adjacent files and programming history to suggest code complements that fit harmoniously into the existing base. This progressive customisation significantly improves the quality of suggestions over time.
Available as an extension in a number of environments (Visual Studio Code, Visual Studio, Vim, Neovim, JetBrains, etc.), Copilot offers a fluid experience that integrates naturally into developers' workflows. Its integrated chat function can also be used to obtain detailed explanations or reformulate requests to refine suggestions.
ChatGPT: the versatile conversational approach
Unlike solutions designed specifically for development, ChatGPT offers a conversational approach that goes beyond the traditional boundaries of coding assistants. Paradoxically, this versatility is their main strength in the field of programming.
Its most advanced model, GPT 4.5, excels particularly in explaining complex concepts and generating code from detailed natural language descriptions. Their ability to understand the full context of a problem means that they provide complete solutions, including documentation and explanatory comments, making them particularly valuable for learning and problem solving.
The arrival of the Canvas option for sharing code and a reduction in the " hallucinations "(errors of logic or syntax).
Although not natively integrated with development environments, ChatGPT's models offer unrivalled flexibility for exploring different approaches, debugging complex code or generating prototypes quickly. Their generalist nature also enables frictionless transition from development to documentation, testing or architectural planning.
Gemini Code Assist: Google's AI code assistant
Launched at the end of 2024, Gemini Code Assist represents the major evolution of Google's offering in the field of code assistants. This new assistant benefits from the advanced capabilities of the Gemini 2.0 Pro model, optimised specifically for programming.
Highlights:
- Native integration into the Google Cloud environment and popular IDEs (VS Code, IntelliJ, Android Studio)
- Multimodal comprehension, enabling you to analyse code, screenshots and diagrams at the same time
- Code generation from sketches or hand-drawn user interfaces
- Exceptional contextualisation that takes account of the entire project and its dependencies
- Real-time optimisation suggestions based on best practices specific to each language
- Advanced support for AI and machine learning with customised model generation
Price: 20 $/month (included in Google One AI Premium, with a limited version available free of charge)
Claude Code: Anthropic's AI code assistant
Launched in early 2025, Claude Code is a command-line tool developed by Anthropic that allows developers to delegate programming tasks directly from their terminal. Based on the Claude 3.7 Sonnet model, it stands out for its ability to understand entire projects and generate context-sensitive code that integrates seamlessly into existing code bases.
Highlights:
- Analysis of complete projects with an overall view, thus reducing one of the limitations mentioned in the original article
- Generation of particularly robust unit tests
- Excellent ability to optimise code performance
- Native integration with Git and the main CI/CD systems
Price: 30 $/month (included in the Claude Pro subscription)
Devin (Cognition): an AI agent that wants to replace engineers
Unveiled in March 2024, Devin describes itself as the "first autonomous AI software engineer". Unlike conventional assistants, Devin can carry out entire development projects with minimal supervision.
Highlights:
- Able to plan, code, test and deploy complete applications
- Can navigate between different tools (IDE, terminal, browser)
- Automatically detects and corrects errors at runtime
- Learns from its mistakes and gradually improves its performance
Price: from 40$ /month for individual developers
Replit Ghostwriter X: capable of generating complete applications
A major evolution of Ghostwriter, this new version launched at the end of 2024 is based on a LLM specially trained on high-quality code.
Highlights:
- Integrated cloud development environment
- Support for over 50 programming languages
- Peer programming" mode, which simulates a real coding partner
- Ability to generate complete applications from simple descriptions
Price: 25 $/month (free limited version)
Specialised AI throughout the development cycle
In addition to general AI, there is a whole range of AI specialised for specific tasks in the software development cycle:
- Project planning and management
Objective: Organising tasks, defining objectives, writing specifications and monitoring project progress- Notion AI / pmMilestones / pmSpecs : specification generation and epic decomposition
- ChatGPT / Claude (in brainstorming mode) : transforming requests into concrete tasks
- Monday.com / Jira / Trello : Kanban boards and agile management enhanced by AI
- ClickUp / Wrike / Smartsheet : dynamic planning and deadline monitoring
- Taskade / Bitrix24 : fully automated project management using AI
- Project Planner (PAI) : software solution with AI to optimise workflows
- Coding and code generation
Objective: Write, complete and generate code quickly.- GitHub Copilot / Amazon Q Developer : contextual completion and code optimisation
- Google Gemini Code Assist : assistance with complex languages.
- Claude Code (CLI): Executing commands via natural requests
- AskCodi : Efficient code generation with AI
- DeepCode (Snyk Code) : real-time analysis and suggestions
- Intel AI Tools Samples : structured workflow for advanced coding
- Managing commits and commit messages
Objective: Write commit messages that comply with standards.- Commitizen / commitlint : validation via VS Code plugins
- GitHub Copilot : message generation from diffs
- Gitmoji : standardised icons for commits.
- Claude Code : automated commits via voice commands
- GitFlow : integration with Conventional Commits for automated pipelines.
- Code review and static analysis
Objective: Detect bugs and improve quality.- Sourcegraph Cody / DeepCode (Snyk Code) : analysis of vulnerabilities.
- CodeClimate / SonarQube : continuous assessment of quality metrics
- DeepSource : detection of inefficient patterns
- ReSharper : real-time refactoring suggestions
- Qodana : Code quality inspection platform integrated into CI/CD pipelines
- Unit test generation and automation
Objective: create automated tests.- GitHub Copilot / Amazon Q Developer : generating tests from comments
- ChatGPT : creation of test suites using natural queries
- Pytest / Jest : integration with AI for parameterised testing
- AutoDev : test automation in CI/CD pipelines
- Selenium : Automated interface tests with AI.
- Refactoring and code optimisation
Objective: Improving readability and performance.- ReSharper / DeepSource : refactoring suggestions.
- Claude Code (CLI): execution of refactoring commands.
- ESLint / Prettier : automated formatting via AI.
- Intel AI Tools Samples : optimising performance via AI workflows
- Black : code formatting with rules predefined by the AI
- Continuous Integration and Deployment (CI/CD)
Objective: automate builds, tests and deployments.- GitLab CI / Jenkins : pipelines with Conventional Commits.
- ChatGPT : generation of YAML/Dockerfile files.
- Claude Code : automated deployment via CLI
- Argo CD : continuous deployment with AI validation
- GitHub Actions : integration with AI-generated workflows
- Documentation and comments
Objective: maintain clear, up-to-date documentation.- Swimm / Notion AI : automatic generation of documentation based on code and real-time updates
- IDE extensions : automatic insertion of structured docstrings to improve code comprehension
- Docusaurus / MkDocs : creation of guides and documentation sites using user-friendly IA tools
- Read the Docs : hosting and automatic generation of documentation in real time with AI integration
- Maintenance, monitoring and version management
Objective: monitor performance and manage software versions effectively.- New Relic / SonarQube : real-time monitoring of applications with proactive detection of anomalies
- GitVersion : automated semantic versioning for consistent version management
- Datadog : advanced monitoring with predictive analysis to optimise performance.
- SemVer : automatic generation of changelogs based on Conventional Commits.
- CLI automation for repetitive tasks
Objective: speed up and automate day-to-day command-line operations.- Claude Code : execution of Git and other commands via natural language queries, simplifying complex interactions
- AutoDev : comprehensive framework for automating refactoring and testing tasks, integrated with CI/CD pipelines
- Taskade CLI : task automation and project management directly from the command line with AI support
- Bitrix24 CLI : complete project management and team collaboration via online orders, optimising workflows
Which tool should you choose?
Despite their impressive capabilities, current AI models have a limited understanding of code and can generate security flaws or vulnerabilities. They also struggle to understand the overall architecture of a project and lack critical judgement.
The choice of an AI assistant depends largely on the specific usage scenarios. For continuous development and team projects using GitHub, GitHub Copilot is a preferred choice thanks to its native integration and its ability to learn the team's style.
The choice also depends on the main programming language. As we saw in the article AI: code generators, a revolution for developers, the languages mastered by these AIs depend on their training data.
The nature of the project is also a key factor. For web and mobile applications, GitHub Copilot offers a fluid experience. Tabnine is relevant for projects with strict confidentiality requirements. For native cloud applications, particularly on AWS, Amazon CodeWhisperer is a logical choice.
Major developments since 2024
Improved understanding of complex projects
New LLMs such as Claude 3.7 Sonnet or GPT-4.5 have considerably improved their ability to understand complex codebases, overcoming one of the main limitations mentioned in the 2024 article. These models can now analyse entire projects and maintain context on tens of thousands of lines of code.
Reduced hallucinations
Hallucinations, a major problem mentioned in the initial article, have been significantly reduced in the new models specialising in code. For example, tests show that GitHub Copilot has reduced its errors by 30 % since 2023, while Claude Code has a hallucination rate of less than 5 % on programming tasks.
Better management of ethical and intellectual property issues
In response to the ethical and legal concerns mentioned in the article, most AI tools for code now offer enhanced confidentiality options:
- Offline mode for sensitive environments
- Automatic filtering of potentially problematic suggestions
- Greater transparency on learning sources
- Options for excluding certain deposits from the drive
Democratising programming
The new code assistants have considerably lowered the entry barrier to programming. Recent studies show that beginners using these tools reach in 3 months the level of competence that previously required more than a year of learning.
The future of code AI
The prediction by Thomas Dohmke (CEO of GitHub) mentioned at the beginning of theArticle from 2024 seems to be confirmed: "AI will write 80% of code within 5 years". By 2025, it is already estimated that 40 to 50 % of commercial code will be generated or heavily assisted by AI.
The most promising future developments include:
- AI that understands business intent Future generations of code assistants will be able to translate business requirements directly into complete technical solutions.
- Specialist assistants by field AIs specifically trained for certain sectors (finance, healthcare, e-commerce) that understand the regulatory constraints and best practices in the field.
- Autonomous maintenance systems capable of monitoring code in production, detecting problems and proposing (or even implementing) corrective measures.
- More natural developer-IA collaboration The interface between the developer and the assistant will become more intuitive, with multimodal capabilities enabling communication via text, voice and even sketches.
The changing role of developers
If code AI is evolving fast, the role of developers must adapt accordingly. AI is transforming development practices, with the emergence of "vibe coding" and "prompt-based development".
The ability to formulate precise instructions for AI assistants is becoming an essential skill. Developers are focusing more on architecture, user experience and business logic.
AI is being integrated into the entire development cycle, from design to maintenance. The customisation of models to meet business needs is a major trend. AI is contributing to the democratisation of development, enabling non-technical profiles to create applications using no code.
The role of the developer is evolving towards supervision, architecture and management of man-machine collaboration.
Conclusion
Choosing an AI tool for coding depends on a range of factors specific to your situation. We recommend a phased approach to adoption. It is crucial to systematically check the code generated and to consider AI as a partner, not a replacement.
What's more, mastering the art of the prompt is still essential to avoid wasting time. The future of software development is taking shape around a man-machine symbiosis, where each party brings its complementary strengths to the table. Developers who embrace these technologies while retaining a critical eye will be best placed to thrive in this new era.
