Your agent needs a SOUL.md you can't write from scratch. I built a 45-minute prompt that writes it for you.

Summary: Your agent needs a SOUL.md you can't write from scratch
By Nate | Nate's Substack | April 15, 2026
๐ง Main Thesis
The biggest unsolved problem in the AI agent revolution isn't technical โ it's human self-knowledge. Most people cannot describe their own work at the resolution an AI agent needs to act on it. Until that gap is closed, agents sit idle after setup.
๐ Key Arguments
The Install Problem Is Solved โ The Spec Problem Isn't
- Tools like OpenClaw (250k GitHub stars), Meta's Manus ($2B acquisition), Anthropic's Dispatch, and Perplexity's hardware product have made deploying personal AI agents trivially easy.
- Yet the most common post in agent communities after a successful install is: "Okay... now what?"
- The bottleneck isn't setup โ it's that users don't know how to write instructions their agent can act on.
The 40-Hour Wall
- When people actually try to spec out their own work for an agent, they hit a wall: they've never articulated what they do at a granular, operational level.
- Every major agent product skips this onboarding step entirely.
The Expertise Trap
- The more senior and experienced you are, the less visible your own mental operating system becomes โ to yourself and others.
- Expertise development intentionally internalises and compresses knowledge, making it harder to externalise on demand.
- This is a feature of how mastery works, not a personal failing.
The Career Problem Hidden Inside the Agent Problem
- Inability to articulate your own work drives: delegation failures, promotion ceilings, and institutional knowledge loss.
- These problems exist with or without AI โ agents just make the gap visible and urgent.
๐ก Practical Takeaway: The SOUL.md
- Nate's solution is a SOUL.md โ a structured document that captures how you actually think, decide, prioritise, and operate.
- You can't write it from scratch because you lack the self-awareness and prompting structure to surface what's implicit.
- He has built a 45-minute interviewer agent prompt that draws this information out through structured conversation, producing a SOUL.md you can then give to any agent as its foundational context.
- The first agent worth running isn't an assistant โ it's an interviewer.
๐ฏ Core Insight
The agent revolution has a prerequisite nobody is selling: the ability to describe yourself and your work with enough fidelity that something else can act on your behalf.
This piece is about that prerequisite โ and provides a tool to meet it.







