🧑‍💻 AI for Coding Essentials

Plan before you prompt, then ship in small reviewable steps — agents for multi-step work, autocomplete for local edits, a human on every merge.

Turn vague requests into reliable code by loading the prompt with real context — files, types, errors, framework version, and clear acceptance criteria.

A confident tone is not correctness — read every diff critically, verify unfamiliar APIs against current docs, and run the code to catch hallucinated packages and plausible-but-wrong logic.

Pick the right tool for the job — autocomplete as you type, chat for questions, agents to run commands and tests — with MCP standardizing how context reaches the model.

Tests are your proof AI-written code actually works — cover edge cases, wire them into CI, and scrutinize AI-authored tests that may just assert the same bug.

Move fast without getting burned — keep secrets out of prompts, watch license and copyright, treat untrusted repo content as a prompt-injection vector, and review for security, not just correctness.