Key Takeaways

  • Jeff Harmon builds working product prototypes late at night using AI coding agents, then hands them off before morning meetings start.
  • Angel Studios pairs executives with dedicated AI assistants whose only job is finishing prototypes that founders start.
  • Shaan Puri argues that executive attention, not AI tooling speed, is now the primary bottleneck in modern software builds.
  • Core engineering teams build safety guardrails around executive code rather than writing early MVPs from scratch.
  • Harmon rejects tech layoff predictions: 10x worker productivity makes companies hire aggressively rather than downsize.

The Method

Jeff Harmon runs Angel Studios and helped build campaigns for Squatty Potty and Poo-Pourri. He spends his nights writing code with AI agents. By 1:00 AM, he has a working product test. By 9:00 AM, daily executive fires pull him away.

“I basically got to the point where I would vibe code products and then I would get stuck because I'd get pulled outside to meetings into fires into different things as an executive in the company,” Harmon explained. “And but I love building out the prototypes of the products. And I just asked I was like I need a person who just can take it from there and just go to the next step.”

Instead of abandoning half-built MVPs or dumping broken code on senior engineers, Harmon hired a dedicated assistant to sit directly between executive ideation and production engineering.

Here is how the loop runs:

1. The founder builds a messy prototype with AI agents until hitting a blocker or leaving for morning meetings.

2. The personal vibe coder takes the raw repository, fixes broken dependencies, completes edge cases, and gets the user flow working.

3. The core engineering team reviews the working build and constructs structural guardrails around the parts Harmon broke.

Shaan Puri notes that most companies get executive staffing backwards. Leaders hire chiefs of staff for scheduling and internal memos, but leave their technical creative work stranded in personal notebooks. As Puri put it: “AI is so fast moving and can do so much, but is the bottleneck is you.”

Where This Breaks Down

This system breaks down when the personal assistant lacks basic technical taste. If the helper simply pastes error logs back into LLMs without testing the output, you duplicate the founder's initial problem and double the engineering team's cleanup work.

It also fails in companies where senior software engineers treat messy prototypes as personal insults. “What I do is I end up vibe coding till I break something and then the engineering team builds a new harness to fix what I broke,” Harmon said. If your lead engineers insist on writing every architectural spec from scratch before seeing a working visual demo, this handoff will create cultural friction.

Finally, this workflow requires founders who accept that greater productivity demands more headcount, not less. Harmon dismisses the idea that AI shrinks technical teams. “When you get something that makes everybody 10 times more effective, you're going to become so much more productive that you need more people,” Harmon said. “You don't say, 'I want less people that are 10 times more effective.' You say, 'I want more people that are 10 times more effective.'”

What to Do With This

Look at the last three software experiments or internal tools you abandoned halfway through this month. Pick the one that directly solves a customer problem. Post a contract job for a junior technical builder whose single evaluation metric is taking your raw prompt-built repository and shipping a working staging URL within 48 hours.