MAESTRO: Autonomous coding agent that argues with itself first

Most AI coding tools start writing immediately. No verification, no pushback. MAESTRO is different. Two frontier models debate before touching your code: - An agent proposes an approach, reading your actual files - Another agent challenges every assumption - You approve the spec - Execution runs inside an isolated Firecracker microVM - Nothing commits until it passes safety checks 95-100/100 scores. Looking for 10 beta testers — free trial.

Hi Product Hunt! I'm Romeo, a professor and researcher who built MAESTRO as a solo project. The core insight: every AI coding tool I tried started writing code immediately. No pushback, no verification. The result was confident but wrong. MAESTRO inverts this. The PM-Agent debate phase is the entire point — GPT-5.4 and Claude reading your actual codebase and arguing before a single file changes. You see the spec before execution starts. You approve it. Only then does the executor run inside an isolated Firecracker microVM. Stack for the curious: FastAPI + Temporal + Firecracker microVMs + Memgraph + Tree-sitter + FAISS + PostgreSQL. Happy to answer any questions about the architecture or the beta.

Built this alone as a researcher who kept losing hours to repetitive coding tasks. MAESTRO debates its own specs before writing a single line — Opus proposes, Gemini challenges, Codex executes. Currently in closed beta with 10 spots. Happy to answer any questions here.