Claude Code reviews

Most of NEXUPIRA is built with Claude Code. It handles large changes across our whole codebase, runs the tests, and keeps going until the change actually works.

Lancetta was built in Claude Code sessions, and it's one of the two agents Lancetta watches 🧡

for coding, building features and editing ui

I'm a designer, not a full-time engineer, so I needed a tool that could own whole features end-to-end, not just autocomplete lines. Claude Code holds the full project context, so I could describe behavior in plain language and get working native macOS code. It let one person ship what normally takes a small team.

Anthropic's models like Opus 5.5 and Fable are very good in long running coding work and intelligence tasks.

The workhorse the handles the big local jobs.

The productivity boost Claude Code provides has been nothing short of amazing. What I was able to accomplish in 5 months with a $140 Claude Pro Max subscription would've cost $30-50K to hire a developer or a year of work if I coded it myself.

I use Claude Code to build our product - frontend and backend - but it also handles a lot of my non-coding work now: reading meeting notes, drafting emails, research. As a founder I'm context-switching all day, and having one tool that can pick up whatever I'm in the middle of has changed how much I get through. It started as a coding tool for us and became something closer to a general assistant. It catches real things on our codebase - edge cases in auth logic, missed error handling. I wouldn't hand it architecture decisions without reading the output, but as a first pass it flags issues we would have shipped. We use it heavily every day and the cost is easy to justify for what it replaces. Usage limits are the only thing we occasionally bump into on long sessions. The permission model is sensible - it asks before running commands or touching anything outside the project, and you can tighten or loosen that per project. We keep secrets in env files it doesn't read into context. No incidents in months of daily use.

Claude Code was central to building Manta AI — from architecture decisions to implementation. Having an AI that understands context deeply across a large codebase made it the right tool for a product this technically complex.

I used Antigravity for the initial backend database configuration as it saves tokens while I moved to Claude Code later for building the user flow and UI/UX enhancement.

One of the vibecoding tools we lean on hard to ship fast, which is honestly part of why we needed Blop in the first place.

Sipcode is a PreToolUse hook for Claude Code. Spent two months on the hook surface and the transcript JSONL format. Allow + updatedInput and deny + reason are the primitives that made it possible.