THE TUESDAY PROTOTYPE
Last Tuesday, I built a working AI agent in two hours that would have taken my old team six weeks and $100,000.
I volunteer at The Human Remembrance Project, an AI nonprofit. Our product Dandelion is an AI agent that helps seniors record their life stories for future generations - while keeping full control over their data and memories.
This prototype uses a pair of agents: one guides users through sharing their life story, while another extracts and organizes key facts. In the background, this will become a chronological narrative, family tree, and digital scrapbook. A third agent casually follows up on details while tracking the narrators cognitive abilities over time.
The product is not polished, scalable or ready for any real usage, but the concept and potential are clear. Two years ago, ChatGPT could barely write anything that sounded remotely human. Now, it can build working software, in two hours.
SOME UNCOMFORTABLE MATH
Why your sprint planning meetings cost more than my entire build
The point is that this amount of work would have taken several weeks with a UX designer, web developer, backend engineer and QA person. We would have had at least two grooming meetings, ten daily standups, and a planning meeting. The UX team would have needed a week of research to design a chatbot.
Let's calculate: 5 people × 6 weeks × 2,885/week = 86,550 in salary alone. Add overhead and you're looking at $100K+ for a prototype.
I did this in two hours.
WHAT STILL BREAKS
Or: Why you still need engineers (for now)
Again, this is not production ready code. Not only should you not put this in front of real customers, but you should also never use the product manager's code. But I am planning to use this to test with real potential users to validate the idea.
It will still take teams of software engineers, architects, UX researchers, and designers to build real software. The difference is the conversation now starts with a working prototype, not a PRD.
THE NEW PM TOOLKIT
When your primary dev environment is a chat window
Software development is probably one of the most complex workflows in the white collar world. Each member of the team is a highly trained specialist representing different categories of knowledge like human behavior, technology and business. Each member makes six figures minimum, and the first figure often starts with a number greater than one.
If AI can replicate work that commands these salaries, we're looking at a fundamental shift in software economics.
But here's what most people miss: the chat window isn't magic. Getting that two-hour result required writing detailed PRDs, defining user personas, creating acceptance criteria, and building a clear technical architecture - before typing a single prompt. The difference? I'm doing the PM work directly with the AI, not through three layers of human translation.
Vibe coding isn't about replacing PM skills. It's about applying them at machine speed.
YOUR MOVE
The conversation every PM should have with their CFO
Less complex workflows with less dependency on highly skilled professionals will be even easier to automate. Small businesses and nonprofits will be able to create their own agents and automations merely by describing what they want.
If I can compress a six-week sprint into two hours for one of the most complex workflows in business, what could you compress in your organization?
Next week: What happened when real seniors actually used it. Spoiler: everything broke in ways I didn't expect, and that's exactly the point.
You had me at "When your primary dev environment is a chat window" 🤣🤣🤣