Summary
Cole Medin demonstrates how to deploy an autonomous AI software factory using GPT-6 Astra and Codeex, arguing that Astra’s significantly improved intent-understanding makes the concept of fully automated code pipelines—PRD in, validated code out—meaningfully closer to practical reality. The video is grounded in a week of firsthand testing comparing Astra against Fable 5.1, with Medin finding Astra consistently ahead on complex autonomous builds and notably better at inferring developer intent without requiring lengthy clarification.
The tutorial walks step-by-step through setting up the open-source software factory project on a remote VPS using Hostinger, including SSH configuration, firewall setup, and dependency installation—all orchestrated through a coding agent that interviews the user for context before beginning deployment. The factory supports both greenfield projects starting from a PRD and existing codebases, with Codeex as the backend coding agent. Medin shows the system running against a demo link-shortener application, illustrating how a GitHub issue can trigger fully autonomous code generation in a 24/7 remote environment.
Medin is careful to push back on Jensen Huang’s AGI declaration around Astra, framing his enthusiasm in practical terms: not AGI, but the first model to meaningfully reduce the friction of autonomous coding workflows at scale. He plans to continue open-sourcing the software factory project with future content on token efficiency and reliability improvements.
📺 Source: Cole Medin · Published September 12, 2026
🏷️ Format: Tutorial Demo
