GuideTechnology
11 min read
How can a language model that only produces text check a calendar, search a database or edit a file? Through tool calling: the model asks, your program acts. This article explains function calling, the Model Context Protocol and the agent loop, with a runnable toy loop that shows allowed tools, step limits and what the model never does itself.
GuideTechnology
10 min read
Vibe coding means accepting AI-generated code without reading it, which is fine for a weekend toy and risky for anything that has users, data, money or a future. This article gives a fair account of both ends of the spectrum and a project workflow that keeps the speed while preserving your understanding.
GuideTechnology
11 min read
Generated code that works can still be vulnerable, depend on a package that should not be trusted, or resemble code with a licence you must respect. This article lists the common security weaknesses, shows a SQL injection example you can run, gives a short security review routine and explains licensing questions without pretending to be legal advice.
GuideTechnology
10 min read
AI-written code often passes AI-written tests, because both come from the same assumptions. This article shows how to write tests that check what the code should do, how to use AI as a second reviewer without relying on it, what to look for in AI-generated diffs, and how to refactor safely.
GuideTechnology
9 min read
Pasting an error into an assistant can solve a bug in a minute, or produce a patch that hides it. This article gives a safe debugging workflow: reproduce, share safely, ask for hypotheses, test them, review the fix, add a regression test, with a worked example and a list of dangerous 'fixes'.
GuideTechnology
10 min read
Reading code is the safest and often most valuable use of an AI assistant: nothing changes until you decide it should. This article covers explaining unfamiliar code, mapping a codebase, learning an API or framework, writing documentation and brainstorming architecture, with the checks that catch confident mistakes.
GuideTechnology
9 min read
Autocomplete, chat, edit mode, agents and review bots all promise faster coding. This article explains what each mode does, how to give it the context it needs, what the evidence says about speed and security, and the responsibilities that stay with you no matter who typed the code.
