How I Use Claude Code to Ship Faster (Solo Founder)
TL;DR
I use Claude Code with a custom skills layer to plan, build, test and deploy across products. Here is my exact workflow and the skills I rely on.
I use Claude Code as the primary editor across every product at Shahriar Labs, backed by a custom skills layer. It compresses planning, coding, testing and deployment into one loop. This post is the exact workflow.
The loop
1. brief -> 2. plan -> 3. implement -> 4. test -> 5. deploy
Each step is a Claude Code subagent or skill. Nothing exits the loop without code review and tests.
The skills layer
A skill is a reusable agent capability. Mine live as open source:
- softco — "act as a full software firm"
- skill-builder — scaffold and audit new skills
- latex-engineer — LaTeX work
- manim-coding-skill — ManimGL animations
- openrouter-free-infer — cheap free LLM routing
- Godot-Skill — typed GDScript for game work
- common-knowledge — Git-backed memory across sessions
The skills layer is the reason this works. Without it, every project rebuilds the same prompts. My AI Agent Skills Stack goes through each.
What I delegate vs what I do myself
Delegate: boilerplate, scaffolding, test generation, documentation, refactors with clear specs, dependency updates.
Do myself: architecture decisions, the first version of any new abstraction, security-sensitive code, anything I cannot review fluently after the fact.
The trust loop
Code generated by an agent gets the same review as code by a contractor. Pre-commit hooks, type checks, tests. The skills enforce style; the human enforces taste.
The tools I built to make AI coding agents 10x goes deeper on the toolchain.
What changes when you ship this way
You stop writing boilerplate. You spend more time on architecture and review. The boring 80% gets faster; the interesting 20% gets the time it deserves.
FAQ
Q: Doesn't Claude Code make junior engineers worse? A: Only if you let it. Used right, it accelerates feedback. Used as a crutch, it stunts.
Q: How expensive is this workflow? A: Cheap calls go through openrouter-free-infer; expensive calls stay on Claude. Cost is meaningful but predictable.
Q: Where do the skills live? A: Open source on github.com/shihabshahrier. Drop them into your Claude Code setup and they work.
Written by Shihab Shahriar Antor. Hire me for AI engineering and product work at Shahriar Labs.
Written by
Shihab Shahriar Antor — AI Engineer & Founder of Shahriar Labs. Creator of LetX, QuantumSketch, and more.