EMZETT.
Login

Emzett Wiki

Code assistants and vibe coding: Opportunities and risks

Properties1
tagsglossar, ki, programmieren

2 min read

Code assistants and vibe coding: Opportunities and risks

In short: AI code assistants suggest code or write whole functions; they speed up routine but need understanding, tests and review, otherwise insecure and unmaintainable programs result.

In more detail: Tools such as GitHub Copilot, Cursor, Claude Code, ChatGPT, JetBrains AI complete code, explain errors, generate tests and with agents build whole features. Vibe coding (Andrej Karpathy, 2025) means: you describe wishes in natural language and accept the code without reading it closely.

In Depth

Where they help

  • Routine and boilerplate: getters, forms, configuration, regex, SQL.
  • Exploring new languages/libraries, generating examples.
  • Explaining errors and suggesting solutions.
  • Writing tests, drafting documentation and comments.
  • Refactoring, translating between languages.
  • Prototypes and small tools quickly (vibe coding suits throwaway projects and experiments).

Risks

  • Wrong or invented code (non-existent functions/packages: “slopsquatting” uses invented package names for malware).
  • Security flaws (SQL injection, missing checks, insecure cryptography, secrets in code).
  • Hard to maintain: bloated code, duplicates, unclear structure; whoever understands nothing can no longer change anything.
  • Licence questions (training data, code with copyleft), privacy (code/data leaves your machine).
  • Loss of learning: whoever never solves things alone does not learn programming.
  • Illusory progress: it runs until it breaks.

Good way of working

  1. Describe the task small and precisely (context, goal, constraints, examples).
  2. Read and understand the code before accepting it; ask for explanations.
  3. Run tests (and your own checks); code review also for AI code.
  4. Check security (inputs, permissions, dependencies, secrets; use scanners).
  5. Version control: small commits, AI changes traceable.
  6. No secrets and customer data in prompts; observe company policies.
  7. Maintain your own skills: practise algorithms and basics yourself. In company/training clarify approval and labelling.

Outlook

Agents take over more (planning, testing, committing), the responsibility stays with the human. Important skills: formulating requirements, understanding architecture, testing, debugging, security, communication. Whoever masters the basics can steer AI; whoever does not is steered.

See also: Using AI assistants sensibly and safely, Code review, SQL injection, Software supply chain