AI coding assistants and coding agents
AI that suggests, explains, fixes and writes code, from autocomplete to agents that run tasks.
What is it?
AI coding assistants help people write software. The simplest ones complete the line you are typing. Chat-based assistants explain code, find bugs and write functions on request.
Coding agents go further: given a task, they can read files in a project, plan changes, edit several files, run tests or commands, and report back. They work best when they can inspect the real codebase and when a person reviews each change. They can save a lot of time, but they also make confident mistakes, so tests and code review remain essential.
Which models use it?
- GitHub Copilot offers code completion, chat and agent features inside editors.
- Cursor is a code editor built around AI assistance.
- Claude Code from Anthropic is a coding agent that works in the terminal and editors.
- OpenAI offers coding features in ChatGPT and through its Codex tools.
- Gemini models power coding help in Google developer tools.
How to use it
- Give the assistant access to, or paste, the relevant files and explain the goal.
- Ask it to inspect existing code and conventions before proposing changes.
- Request a short plan first, then small, reviewable changes.
- Run the tests, linters and the app yourself, and paste back any errors.
- Review every change line by line before merging, especially security-related code.
- Never paste secrets such as passwords or API keys into prompts.
Benefits
- Faster boilerplate, tests and routine fixes.
- Explains unfamiliar code and error messages in plain language.
- Helps learners understand programming concepts step by step.
- Agents can handle multi-file chores while you focus on design decisions.
Limits and risks
- May invent functions, packages or APIs that do not exist.
- Can introduce subtle bugs or security problems that look correct.
- Suggested code may not match your project's versions or conventions.
- Agents with terminal access can make destructive changes; limit permissions and use version control.
- Licence and privacy rules about sharing company code apply.
Prompt template
Copy it, replace the parts in [brackets] and paste it into any AI chat.
You are helping me with a [language/framework] project. Goal: [what you want to build or fix]. Relevant files: [paste or list files]. Error or current behaviour: [paste error / describe]. First read the code and explain the likely cause. Then propose the smallest safe change as a patch, list any edge cases, and tell me which tests to run. Do not invent functions or packages; ask if something is missing.
More in AI at work
- Writing and editing with AIUsing AI to draft, rewrite, shorten, proofread and adjust the tone of text.
- Data analysis with AIAsking AI to explore spreadsheets and data, write formulas and code, and explain the results.
- AI in customer supportChatbots and assistants that answer customer questions and help support teams work faster.
- AI for learning and tutoringUsing AI as a patient tutor to explain topics, quiz you and give feedback.
- AI for design and creative workUsing AI to brainstorm ideas, make mockups, edit photos and speed up creative work.
- Translation with AITranslating text, documents and conversations between languages with AI.