Getting started with AI for Coding | AI Competency Centre
Getting started with AI for Coding
This page offers advice and guidance to get you started with using AI for writing code:
Why use generative AI for coding
One of the most common applications of applications powered by Large Language Models is to write computer code. Generative AI can be and is being used by both experienced and novice developers as well as people who have no programming experience at all.
You can use generative AI to create functional applications, interactive tools, and useful widgets without traditional programming knowledge but you can also use it if you’re an experienced developer. LLMs transform natural language descriptions into working code.
AI coding enables you to:
- Build practical tools without years of programming education
- Prototype ideas rapidly for testing and validation
- Automate repetitive tasks through custom applications
- Create interactive interfaces that work immediately
- Transform research workflows with bespoke analysis tools
There are still many limitations to the type of applications that can be built without any knowledge of programming but many tasks can be accomplished with minimal effort.
Four Levels of AI-Assisted Coding
Understanding AI coding tools requires recognizing that they exist on a spectrum of complexity and capability. Each level demands different knowledge and offers different possibilities for what you can create.
Choosing Your Starting Level
Note: Level 1 is not ideal for complete beginners because it requires some knowledge of how computer code works.
- Complete Beginners: Start with Level 2 (Canvas/Artifacts) using Gemini or Claude
- Some Technical Experience: Begin with Level 2, quickly move to Level 3 builders
- Existing Developers: Jump to Level 4 for productivity enhancement
- Research Focus: Level 2 for analysis tools, Level 3 for data collection platforms
The progression isn't always linear - many professional developers use tools from multiple levels depending on the task at hand.
What it is: A Large Language Model inside a chatbot generates code that you copy and paste into files or environments you manage yourself.
Knowledge Required
- Minimum: Understanding that code needs to run somewhere and basic file management
- Practical: Knowing where to put different types of files (HTML in browser, Python scripts in terminal)
- Ideal: Full coding knowledge - this level mainly saves time for experienced developers
Key Tools
- Any AI chatbot: ChatGPT, Claude, Gemini, local models
- Code-specific models: GitHub Copilot Chat, CodeT5
- Browser: Your most accessible code runner for HTML/JavaScript
What You Can Build
- Single-file web applications that run in browsers
- Utility scripts for data processing
- HTML pages with embedded functionality
- JavaScript calculators and converters
Example Workflow: Ask for a word frequency analyzer → Get HTML/CSS/JavaScript code → Save as .html file → Open in browser → Working application
What it is: LLM writes code inside a chatbot and immediately runs it within the chat interface, creating interactive applications you can use and modify in real-time.
Knowledge Required
- Minimum: Knowing this capability exists
- Practical: Understanding what types of applications are possible with web technologies
- Ideal: Familiarity with web frameworks, data formats, and API concepts
Leading Platforms
Google Gemini (Best Overall)
- Strengths: Superior sharing capabilities, can embed Gemini chat in applications, rich preset options
- Best for: Applications you want to share with others, data visualization, interactive tools
- Unique feature: Can add AI assistant functionality directly into your creations
Claude Artifacts (Most Sophisticated)
- Strengths: Advanced interface libraries, React framework support, complex data handling
- Best for: Widgets and interface prototyping
ChatGPT Canvas (Most Accessible)
- Strengths: Familiar interface, decent functionality, widely available
- Best for: Quick prototypes, learning the concepts, simple applications
- Limitation: Limited sharing and deployment options
Qwen Chat (Free Alternative)
- Strengths: Completely free, good sharing, built on open-source platform
- Best for: Experimenting with AI-assisted coding without subscription costs
- Trade-off: Slightly less powerful models than ChatGPT or Claude
What You Can Build
- Interactive data analysis dashboards
- Custom calculators with rich interfaces
- Text analysis and visualisation tools
- Educational quizzes and study materials
- Research data exploration interfaces
When to Move On: When you want applications that persist independently or need database integration
What it is: LLM-powered services that plan and execute the entire development process, from concept to deployed application, handling multiple files, frameworks, and deployment automatically.
Knowledge Required
- Minimum: Clear vision of what you want to build
- Practical: Understanding of web applications, databases, user authentication, deployment concepts
- Ideal: Knowing when you need databases, APIs, user management, and production considerations
Leading Platforms
Lovable (application prototyping)
- Strengths: Full website creation, GitHub integration, one-click publishing, professional deployment
- Best for: Client-facing applications, complete websites, team collaboration projects
- Pricing: Free tier for testing, paid for deployment and advanced features
Bolt (Rapid Development)
- Strengths: Fast iteration, real-time collaboration, excellent for prototyping
- Best for: Quick concept validation, iterative design, team brainstorming
- Focus: Speed over polish, great for early-stage development
v0 by Vercel (Interface Prototyping)
- Strengths: Outstanding UI/UX generation, component-based development, modern design patterns
- Best for: Frontend applications, design system creation, interface-heavy projects
- Specialty: Converting designs and mockups into working applications
Google AI Studio (Free Deployment)
- Strengths: No-cost deployment to Google Cloud, integration with Google services
- Best for: Academic projects, research tools, experimental applications
- Limitation: Less sophisticated than specialized builders but completely free
What You Can Build
- Complete business websites with multiple pages
- Database-driven applications with user accounts
- E-commerce platforms with payment processing
- Content management systems
- Research platforms with data collection capabilities
Learning Curve: Expect significant learning about web architecture, databases, user management, and deployment processes
What it is: AI-powered development environments that assist with complex, multi-file projects, handling entire codebases, version control, and sophisticated development workflows.
Knowledge Required
- Minimum: Understanding of development environments, file systems, terminal/command line
- Practical: Version control (Git), package managers, development frameworks, deployment pipelines
- Ideal: Software engineering principles, system architecture, security practices, team collaboration workflows
Professional Tools
Cursor (Industry Standard)
- Strengths: Excellent code completion, project-wide understanding, popular among professionals
- Best for: Complex applications, team development, maintaining large codebases
- Requirements: VS Code familiarity helpful but not essential
- Alternatives: Windsurf, Cline, others
GitHub Copilot (Most Integrated)
- Strengths: Deep GitHub integration, works within existing development workflows
- Best for: Developers already using GitHub, open-source projects, collaborative development
- Education: Free access available for students and educators
Claude Code (Command Line Tools)
- Strengths: Direct file system access, automation capabilities, command-line integration
- Best for: Automated workflows, batch processing, system administration tasks
- Requirements: Claude Pro subscription or Claude API
- Alternatives: Codex CLI, aider chat, Gemini Code
Google Colab (Research Focus)
- Strengths: Jupyter notebook integration, free GPU access, academic-friendly
- Best for: Data science, machine learning, research computing, educational contexts
- LLM Features: Now has integration with Google Gemini which can write or explain code or even create entire notebooks
What You Can Build
- Enterprise-grade applications with complex architectures
- APIs and microservices with proper documentation
- Mobile applications with cross-platform frameworks
- Data science pipelines with automated analysis
- Open-source projects with community contribution workflows
Professional Considerations: Requires understanding of software engineering principles, security best practices, testing methodologies, and production deployment strategies.
Getting Support and Training
AI Competency Centre Resources
Direct Consultation: Request project-specific guidance through our Expression of Interest form. Our technical staff and AI consultants provide guidance for implementing AI coding tools in research workflows.
Training Programmes: Attend our foundational workshops designed for understanding large language models and their applications. We also deliver tailored workshops for departments and colleges covering effective and responsible use of AI tools. Request custom training for your research group through our Expression of Interest form.
Community Engagement: Join the AI Builders User Group (BUG) for technical discussions about building applications using large language models, or the Generative AI Special Interest Group for broader discussions about AI in academic contexts.
AI Coding Tool Support
GitHub Copilot Information: Learn about AI coding assistants and how to access GitHub Copilot Individual (free for academic staff and students) through our FAQ guidance. We provide information about privacy settings and data retention policies, but do not offer hands-on training.
Tool Selection Guidance: Our consultants can advise on choosing appropriate AI coding tools for your project, including platforms like Cursor, ChatGPT, Claude, and local development environments with Continue and aider.
University-Supported AI Tools: Access ChatGPT Edu and Microsoft 365 Copilot through departmental licensing. ChatGPT Edu includes Advanced Data Analysis with Python code interpreter capabilities for research applications.
Additional Resources
API Access Programme: Request access to OpenAI API for research projects with an initial $50 monthly limit per project. Higher limits available based on project needs. Learn more on our Resources for Researchers page.
Partner Training: Connect with Oxford Research Software Engineering (OxRSE) training programmes which are expanding to include ML-specific courses, and Cloud Computing for Research Competency Centre courses relevant for ML project hosting.
Funding Opportunities: Explore the AI Teaching and Learning Exploratory Fund for research projects investigating AI applications in academic contexts. Contact us through our Expression of Interest form to learn about current funding rounds.