Claude Code Review: A Terminal Agent on Unfamiliar Code
Hands-on notes from using a terminal coding agent for a week on a codebase I did not write: where autonomous multi-file work paid off, and where supervision was cheaper than delegation.
The full Hamzify archive in publication order: AI coding tool reviews, vibe coding experiments, AI agent build logs, comparisons and workflows.
Editor's picks
Hands-on notes from using a terminal coding agent for a week on a codebase I did not write: where autonomous multi-file work paid off, and where supervision was cheaper than delegation.
A full build log of shipping a small SaaS with coding agents doing most of the typing: what the agents handled well, the three places they stalled, and what a human still had to own.
A repeatable five-step loop for working with an AI coding assistant on a real codebase: brief, constrain, generate, verify, integrate — and what belongs in each step.
A hands-on review of Cursor as a daily driver on an existing production codebase — where agent mode earns its keep, where it costs you time, and who should stay in their current editor.
A reference checklist for what an AI coding agent needs to know before it starts: the seven inputs that change output quality, and the ones that only add noise.
A review workflow tuned to the specific mistakes coding models make: a triage order, the six failure patterns worth hunting for, and where to spend your attention.
A use-case comparison of Cursor and GitHub Copilot run through identical tasks — refactoring, a new feature, an unfamiliar codebase and test writing — with a recommendation for each.
A guide to the AI development tool landscape organised by the job each category does — editors, agents, CLI tools, review bots and model routers — and how to tell when you need one.
An experiment in single-prompt development: one paragraph of requirements, an agent, and a Chrome extension that has to actually load in a browser and work.
A build log of rebuilding a developer portfolio in a weekend with AI writing most of the code — the design decisions that had to stay human, and the performance work that got missed.
An experiment in shipping a small real product in a single evening with AI doing most of the implementation — and an honest account of which shortcuts were fine and which were debt.