9. Coding, Software, and Technical Work

AI coding assistants can explain unfamiliar code, generate examples, suggest tests, document functions, help debug errors, and support prototyping. They can also introduce security vulnerabilities, incorrect assumptions, licensing problems, and subtle bugs.

9.1 Examples of coding tools

GitHub Copilot, Cursor, and coding capabilities within general-purpose assistants support different combinations of autocomplete, chat, codebase search, agentic changes, and test generation. Cursor describes features for multi-file editing, codebase context, terminal commands, and review of proposed changes (Cursor, n.d.). xAI has also described Grok-oriented coding and application-building capabilities (xAI, 2026c). These examples are illustrative; verify current functionality, licensing, enterprise controls, and supported development environments.

9.2 A safe development cycle

1. Explain the intended behaviour, constraints, language, framework, and existing architecture.
2. Ask the assistant to outline an approach before generating code.
3. Review dependencies, permissions, data flows, and security implications.
4. Generate small, testable changes rather than accepting a large opaque rewrite.
5. Run unit, integration, and security tests in a safe environment.
6. Inspect diffs and understand every consequential change before merging.
7. Check licences, secrets, logging, error handling, and accessibility.
8. Document the use of AI where required by project policy.

9.3 Security and reliability

Do not paste passwords, private keys, production credentials, or proprietary code into a tool unless approved under the relevant security and contractual controls. Treat generated code as untrusted until reviewed and tested. An assistant’s explanation of why code works is not proof that it works. For high-impact software, require peer review, automated testing, threat modelling, and change management.