GenAI · 1 min read
The Overnight Agent Factory — Autonomous Development with Claude Code & Cursor
How I run parallel AI agents overnight to build production software while I sleep.
One of the most transformative shifts in my development workflow has been what I call the Overnight Agent Factory — a setup where multiple Claude Code agents and Cursor instances work in parallel on different parts of a project while I’m asleep.
The Setup
- Claude Code (Max subscription): Primary agentic coding, architecture decisions
- Cursor Pro: Parallel workstreams, code review, refactoring
- MiniMax M2.5 API: Supplementary generation for boilerplate and tests
- tmux + git worktrees: Isolation between parallel agents
The Three-Tier AI Model
For production AI features, I’ve developed a three-tier approach:
- Deterministic rules — Business logic that must be exact (compliance, validation)
- Trained ML models — Pattern recognition on structured data (classification, scoring)
- LLM agents — Natural language understanding, complex reasoning, user interaction
This hierarchy ensures reliability where it matters while leveraging AI flexibility where it adds value.