PMs Using AI for Applications: 1 Hour Development, 5 Hours Deployment, Threshold Shifted
AI Summary · Serial Entrepreneur Perspective
A PM with zero coding skills launched an MBTI game in 6 hours using Codex + Alibaba Cloud. Key insight: AI has eliminated the "development" barrier, but left the "ops/deployment" barrier for you to tackle. Your existing knowledge stock determines your AI leverage.
Key Insight: The barrier hasn't disappeared, it just moved
As a serial entrepreneur, I've seen too many triumphant tales of "building a product with AI from scratch." But this post is refreshingly honest: a product manager who can't write a line of code went from idea to live launch in just 6 hours using Codex + Alibaba Cloud. Development took 1 hour; deployment took 5.
Behind those numbers lies a counterintuitive truth: AI hasn't flattened the entrepreneurial barrier—it has simply shifted it from "writing code" to "understanding systems" and "specifying requirements."
Why did development take just 1 hour? Because it's a PM's home turf
In development, AI did 40 minutes of work and the human did 20. Why was it so smooth? Because a product manager's core strengths are describing requirements, breaking down logic, and accepting delivery. That's exactly what we do for a living.
OpenAI's data backs this up: among Codex users, non-developers like product managers, lawyers, and analysts make up 20%—and their growth rate exceeds that of programmers. The more precisely you describe, the faster AI delivers.
Why did deployment take 5 hours? Because knowledge gaps get amplified
In deployment, AI handled only 1 hour while the human spent 4. The author got stuck on basics like "how do I get local files onto a cloud server," "what is SSH," and "which machine should I run this command on?"
The issue isn't that AI lacks smarts—it's that AI assumes the person on the other end has ops common sense. Every command it gives is correct, but the author didn't know which steps to run on their own laptop versus on the server. It's like holding a perfectly accurate map but having no idea what city you're in.
The farther your knowledge base is from AI's, the less AI can do on your behalf—and the more mess you'll have to clean up yourself.
Three actionable tips for founders
1. Reverse the order: ask AI for the full picture first, then start building
When I used to hit something unfamiliar, my first instinct was "I don't know how, so I can't do it." Now you should ask AI first: "What does the standard workflow look like?" Let it lay out the map so you immediately see where your gaps are. If the author had asked about the full deployment process upfront, at least half of those 5 hours could have been saved.
2. Don't treat "I don't know" as an endpoint—treat it as a starting point
Those 4 painful hours were essentially catching up. Stuck at every step, ask and learn—each knowledge block you fill will double your human-AI collaboration efficiency next time. Two years ago, learning Linux might have taken three months of grinding; now a few weekends of Q&A is enough.
3. Understand the new value of the "mover"
Future entrepreneurs may just be "connectors" standing in the middle: AI on the left writes code, AI on the right handles ops, and you're the one who understands what the entire pipeline looks like and knows when to pass instructions from one side to the other. The ability to see the big picture is scarcer than knowing how to code.
Original · Every Product Manager Is One: Read original →